Liveness Detection in Fingerprint Recognition Systems

Transcription

1 Liveness Detection in Fingerprint Recognition Systems Examensarbete utfört i Informationsteori vid Linköpings tekniska högskola av Marie Sandström Reg nr: LITH-ISY-EX Linköping 2004

2

3 Liveness Detection in Fingerprint Recognition Systems Examensarbete utfört i Informationsteori vid Linköpings tekniska högskola av Marie Sandström Reg nr: LITH-ISY-EX Supervisor: Fredrik Claesson Examiner: Viiveke Fåk Linköping 10th June 2004.

4

5 Avdelning, Institution Division, Department Datum Date Institutionen för systemteknik LINKÖPING Språk Language Svenska/Swedish X Engelska/English Rapporttyp Report category Licentiatavhandling X Examensarbete C-uppsats D-uppsats Övrig rapport ISBN ISRN LITH-ISY-EX Serietitel och serienummer Title of series, numbering ISSN URL för elektronisk version Titel Title Författare Author Detektering av Artificiella Fingeravtryck vid Användarautenticiering Liveness Detection in Fingerprint Recognition Systems Marie Sandström Sammanfattning Abstract Biometrics deals with identifying individuals with help of their biological data. Fingerprint scanning is the most common method of the biometric methods available today. The security of fingerprint scanners has however been questioned and previous studies have shown that fingerprint scanners can be fooled with artificial fingerprints, i.e. copies of real fingerprints. The fingerprint recognition systems are evolving and this study will discuss the situation of today. Two approaches have been used to find out how good fingerprint recognition systems are in distinguishing between live fingers and artificial clones. The first approach is a literature study, while the second consists of experiments. A literature study of liveness detection in fingerprint recognition systems has been performed. A description of different liveness detection methods is presented and discussed. Methods requiring extra hardware use temperature, pulse, blood pressure, electric resistance, etc., and methods using already existent information in the system use skin deformation, pores, perspiration, etc. The experiments focus on making artificial fingerprints in gelatin from a latent fingerprint. Nine different systems were tested at the CeBIT trade fair in Germany and all were deceived. Three other different systems were put up against more extensive tests with three different subjects. All systems were circumvented with all subjects' artificial fingerprints, but with varying results. The results are analyzed and discussed, partly with help of the A/R value defined in this report. Nyckelord Keyword biometrics, identification, verification, fingerprints, fingerprint scanners, sensor attacks, artificial fingerprints, liveness detection

6

7 Abstract Biometrics deals with identifying individuals with help of their biological data. Fingerprint scanning is the most common method of the biometric methods available today. The security of fingerprint scanners has however been questioned and previous studies have shown that fingerprint scanners can be fooled with artificial fingerprints, i.e. copies of real fingerprints. The fingerprint recognition systems are evolving and this study will discuss the situation of today. Two approaches have been used to find out how good fingerprint recognition systems are in distinguishing between live fingers and artificial clones. The first approach is a literature study, while the second consists of experiments. A literature study of liveness detection in fingerprint recognition systems has been performed. A description of different liveness detection methods is presented and discussed. Methods requiring extra hardware use temperature, pulse, blood pressure, electric resistance, etc., and methods using already existent information in the system use skin deformation, pores, perspiration, etc. The experiments focus on making artificial fingerprints in gelatin from a latent fingerprint. Nine different systems were tested at the CeBIT trade fair in Germany and all were deceived. Three other different systems were put up against more extensive tests with three different subjects. All systems were circumvented with all subjects artificial fingerprints, but with varying results. The results are analyzed and discussed, partly with help of the A/R value defined in this report. i

8 ii

9 Acknowledgment A number of people have helped me making this thesis work possible. First of all, I would like to thank everybody at ISY, especially my examiner Viiveke Fåk, my supervisor Fredrik Claesson, the Data Transmission group for lending me their camera, and Sören Hansson, who helped me with the PCB production part of the experiments. I would also like to take the opportunity to thank the Fingerprint Group at the National Laboratory of Forensic Science (SKL), especially Lena Hallberg and Göran Kidfelt. I could not have performed the experiments without the participants. Thank you for lending me your fingerprints! The experiments at CeBIT would not have been able to perform without the companies who let me try their products. Thank you all for being patient with me! I would also like to thank the following people who helped me in various ways: Ulf Söderholm, Fredrik Larsson, Björn Mellström, Bo Thunér, Susanne Edlund, Maria Magnusson Seger, Johan Blommé, and Andreas Bergner. Last but not least, I would like to thank Hannes Lindblom, whom I could not have done this thesis without. iii

10 iv

11 Be aware... Before you start reading this report, take a close look at your fingertips. Your papillary lines might form a loop, a whorl, or maybe it looks more like an arch. If you look even closer, you might be able to see some lines that split into two, a delta pattern somewhere, and maybe you can even see some sweat drops coming out of the pores on your fingertips. Your fingerprint patterns are most certainly unique in the whole world. In theory, it is thus possible to identify you with help of a single fingerprint. If it was possible to make a copy of your fingerprint, your identity could then be used. Do you remember every single thing you touched today? Maybe you touched a few door handles, a glass, or a cup. Are you sure nobody has been watching you to be able to steal your fingerprint? Remember that a password can be changed, a new credit card can be bought, but a finger is not as easily changed. v

12 vi

13 Contents 1 Introduction Background Basic terminology Goal Purpose Method Method criticism and limitations Target group Reading guide Notes Biometric overview Identification and verification Methods of identification and verification Results from identification and verification procedures Biometric techniques Physical characteristics Behavioral characteristics Fingerprints History Today Fingerprint characteristics Classification and pattern types Terminology Enhancement techniques Processing techniques Fingerprint scanners Fingerprint images Scanning techniques Optical sensors Solid-state sensors vii

14 viii Contents Ultrasonic sensors Touch versus sweep Algorithms in fingerprint scanners Image enhancement Feature extraction and comparison Sensor attacks and protection schemes Registered finger Unregistered finger A twin s fingerprint or a genetic clone Artificial fingerprint Others Liveness detection Liveness detection in biometric systems Using extra hardware Temperature Optical properties Pulse Pulse oximetry Blood pressure Electric resistance Relative dielectric permittivity Combining ECG, pulse oximetry, and temperature Detection under epidermis Other claims Using existing information Skin deformation Pores Unique characteristic for each individual Perspiration Testing of liveness detection methods Relevance of liveness detection Other methods to limit spoofing Multiple snapshots of the same finger Multiple fingers Challenge-response Supervision Multi-modal biometrics Multiple identification/verification methods Additional comments History of artificial fingerprints Albert Wehde s work Six biometric devices point the finger at security Biometrical fingerprint recognition: don t get your fingers burned.. 50

15 Contents ix 6.4 Impact of Artificial Gummy Fingers on Fingerprint Systems Body Check Biometric Access Protection Devices and their Programs Put to the Test An Investigation Into the Vulnerability of the Siemens ID Mouse Professional Version Spoofing and Anti-Spoofing Measures Fooling Fingerprint Scanners Biometric Vulnerabilities of the Precise Biometrics 100 SC Scanner Evaluation of biometric security systems against artificial fingers Experiment description Making of the artificial fingerprint Enhancing the fingerprint Photographing the fingerprint Image processing Printing the image PCB production Gelatin solution Experiments at CeBIT Extensive experiments Subjects and input Software and hardware Experiment procedure Results CeBIT Extensive experiments Results in numbers Results in percent Discussion and analysis Experiment method Enhancing the fingerprint Photographing the fingerprint Image processing Printing the image PCB production Gelatin solution Experiments at CeBIT Sweeping sensors Extensive experiments Experience Subjects Initial test results The A/R value

16 x Contents Comparison with results from previous studies Additional comments about artificial fingerprints Finding a quality latent fingerprint Alternative acquisition of fingerprint image Economies of scale Forging fingerprints Cooperation using latent print Using the artificial fingerprint Conclusion Final conclusion Future work Liveness detection Artificial fingerprints Fingerprint scanners Alternative biometrics Bibliography 97 A Dictionary 103 B Material 109 B.1 Enhancing the fingerprint B.2 Photographing the fingerprint B.3 Image processing B.4 Printing B.5 PCB production B.6 Gelatin solution C Experiment details 113 C.1 Photographing C.2 Image processing C.3 Fingerprint images before and after image processing C.4 PCB production D Scanners used in extensive experiments 119 E Software used in extensive experiments 121 F Test results 123 F.1 Results per fingerprint scanner F.1.1 Identix F.1.2 Targus DEFCON TM Authenticator TM F.1.3 Precise TM Biometrics 100 MC F.2 Results per subject F.2.1 Real fingerprints

17 Contents xi F.2.2 Artificial fingerprints

18 xii Contents List of Figures 2.1 Enrollment, verification, and identification. [41] The relationship between FRR, FAR, and EER. [34] The three major pattern types: arches, loops, and whorls. [22] Core and delta points. [7] Minutiae details. [7] Sweat pores. [41] Cross-section of a papillary line. [48] An FTIR-based fingerprint sensor. [41] A fingerprint sensor using FTIR with a sheet prism. [41] Fingerprint sensing using optical fibers. [41] Electro-optical fingerprint sensor. [41] Capacitive fingerprint sensor. [41] An ultrasonic fingerprint sensor. [41] A sweeping sensor. [41] Typical structure of a fingerprint recognition system. [3, 42] West Virginia perspiration detection method. [9, 34] An overview of the process of making the mold Soot powder mixture and squirrel hair brush A mold with a gelatin solution on top of it A fingertip with a wafer-thin gelatin fingerprint on top of it The number of successful logins with real fingerprints The number of false acceptances with artificial fingerprints The success rate with real fingerprints The FAR with artificial fingerprints Mean values, in percent, for real and artificial fingerprints Results of unofficial tests with subject S B.1 Materials used during production of PCB. [14] B.2 Gelatin used for making artificial fingerprints. [3] C.1 S1 s fingerprint before and after image processing C.2 S2 s fingerprint before and after image processing C.3 S3 s fingerprint before and after image processing D.1 Identix fingerprint scanner. [30] D.2 Targus DEFCON TM Authenticator TM. [54] D.3 Precise TM Biometrics 100 MC. [2] E.1 Screenshot of BioLogon TM for Windows. [3]

19 Contents xiii E.2 Screenshot of Softex Omnipass. [3] E.3 Screenshot of Precise BioManager TM included in Precise Logon software. [3]

20 xiv Contents List of Tables 6.1 Characteristics of a live finger compared to a gelatin artificial fingerprint. [42] Experiment types. [42] Possible experiment types. [42] Testing order for the scanners in round one and two Results from attacks with artificial fingerprints at CeBIT The A/R value for all subjects, round one The A/R value for all subjects, round two F.1 Results of the Identix fingerprint scanner for real fingerprints F.2 Results of the Identix fingerprint scanner for artificial fingerprints, round one F.3 Results of the Identix fingerprint scanner for artificial fingerprints, round two F.4 Results of the Targus fingerprint scanner for real fingerprints F.5 Results of the Targus fingerprint scanner for artificial fingerprints, round one F.6 Results of the Targus fingerprint scanner for artificial fingerprints, round two F.7 Results of the Precise fingerprint scanner for real fingerprints F.8 Results of the Precise fingerprint scanner for artificial fingerprints, round one F.9 Results of the Precise fingerprint scanner for artificial fingerprints, round two F.10 Sum of values per user for real fingerprints F.11 Success rate, FRR, and FAR, per subject for real fingerprints F.12 Sum of values per subject for artificial fingerprints, round one F.13 Sum of values per subject for artificial fingerprints, round two F.14 Values in percent, per subject for artificial fingerprints, round one F.15 Values in percent, per subject for artificial fingerprints, round two.. 128

21 Chapter 1 Introduction This chapter contains a short introduction to the thesis. The goal, purpose, method, and target group will be presented, method criticism and limitations will be discussed, and a reading guide will give the reader a quick guide to each chapter. 1.1 Background The use of biometric systems are growing every day. Fingerprint scanning is the one biometric identification method available today that is mostly used. The security of fingerprint scanners has however been questioned and previous studies have shown that fingerprint scanners can be fooled with artificial fingerprints, i.e. copies of real fingerprints. The fingerprint systems are evolving and this study will discuss the situation of today. 1.2 Basic terminology The term artificial fingerprint will be used in this report to refer to artificially created fingerprints, as compared to fake fingers/fingerprints which may also include modifications of live fingers. Many other writings in the area use the terms artificial finger or artificial fingertip, but these terms have not been used in this report to emphasize that the artificial fingerprints made from latent fingerprints, are in fact thin small prints and not entire artificially created fingers or fingertips. The terms live finger/fingerprint and real finger/fingerprint will be used to denote a finger/fingerprint which is part of a living body. 1

22 2 Introduction 1.3 Goal Two different approaches to the fingerprint scanner area will be covered in this report. The theoretical approach will discuss liveness detection, i.e. the fingerprint scanners ability to distinguish between live fingers and artificial clones. Different liveness detection methods will be presented and analyzed with regards to attacks with artificial fingerprints. The empirical approach consists of examining the fingerprint scanners ability to withstand an attack of an artificial fingerprint using techniques based on earlier research made in [42] and [57]. More information about the method is found in section Purpose Several reports [3, 42, 57] have noted successful attacks on fingerprint systems using artificial fingerprints. Since the fingerprint scanner market is growing and the technology is evolving, new products that can withstand attacks with artificial fingerprints might have seen the light today. This report will give a further examination of the fingerprint scanner area to clarify whether or not fingerprint systems can be trusted or if they are too insecure to be used today. 1.5 Method For the theoretical approach, a literature study has been performed. proceedings, books, etc. have been read, discussed and analyzed. Articles, Prior to this report, a number of experiments have been performed based on earlier research made in [42, 57]. A latent fingerprint on a piece of glass was the startingpoint for the creation of the artificial fingerprint. This starting-point was chosen to simulate the user s lack of awareness that the fingerprint was being stolen from him/her, as if the latent print was taken from a drinking-glass. The method used is described in more detail in chapter 7 on page Method criticism and limitations Glasslike surfaces is only one of the possible surfaces a fingerprint can be found on. To investigate all possible surfaces, would require an enormous effort and a lot of time, and is therefore not part of this thesis.

23 1.7 Target group 3 Using an artificial fingerprint is only one of the possible attacks on a fingerprint system. Attacks at the sensor level will be described shortly in section 4.5 on page 31, but describing all the other possible attacks is outside the scope of this report. 1.7 Target group This report has a number of different target groups: Manufacturers of fingerprint recognition systems. Companies considering starting to use a fingerprint recognition system. Users of fingerprint recognition systems. Researchers who want to continue researching the field of fingerprint recognition systems, especially when it comes to liveness detection and attacks with artificial fingerprints. Students in the field of computer science, information technology, etc., who have an interest in the security field and especially biometrics. Since this report has so many different target groups, different parts of the report are relevant to different groups of people. The reader is not presumed to have previous knowledge about computer security, biometrics, or fingerprint recognition systems. The reading guide in section 1.8, is recommended for the reader who quickly wants to find the relevant parts for his/her specific purpose. 1.8 Reading guide This section contains a short description of each chapter and appendix in the report. Chapter 1 on page 1 contains a short introduction to the thesis. The goal, purpose, method, and target group are presented, and method criticism and limitations are discussed. Chapter 2 on page 7 gives the reader an introduction to the biometric area, describes important terms, and is a good starting point for the following chapters. Chapter 3 on page 13 presents the historical and present use of fingerprints, physical characteristics of fingerprints, and different enhancement techniques of fingerprints. Chapter 4 on page 21 discusses the use of fingerprint scanners, different scanning technologies, and briefly explains the algorithms used in the scanners.

24 4 Introduction The chapter also describes possible ways of intrusion at the sensor level of fingerprint scanning systems, as well as the available protection schemes. Chapter 5 on page 35 presents and discusses ideas about liveness detection, i.e. fingerprint scanners ability to distinguish between live fingers and artificial clones. Chapter 6 on page 49 summarizes the most important previous work in the field of artificial fingerprints. Chapter 7 on page 55 describes the method used in the experiments, the creation process of artificial fingerprints, and the material and software used in the experiments. Chapter 8 on page 67 presents the results from the experiments described in the previous chapter. Chapter 9 on page 75 analysis and discusses the method used in the experiments and the results acquired. Chapter 10 on page 93 contains a final conclusion and ideas about future work in the fields of liveness detection, artificial fingerprints, and fingerprint scanners. Appendix A on page 103 contains an alphabetized explanatory list of abbreviations, technical terms, and medical terms used in this report. Appendix B on page 109 contains detailed information about the material used in the experiments. Appendix C on page 113 describes some parts of the experiment method in more detail. Appendix D on page 119 contains detailed information about the scanners used in the extensive experiments. Appendix E on page 121 contains detailed information about the software used in the extensive experiments. Appendix F on page 123 presents the detailed data (in numbers) of the results from the extensive experiments. 1.9 Notes An extensive list with explanations of all important technical and medical terms and abbreviations, can be found in appendix A on page 103. The most important terms used will still be explained in the appropriate sections. If nothing else is stated, references placed before a period in a sentence, refers to the sentence only, while a reference placed after the period refers to the whole

25 1.9 Notes 5 paragraph. A reference placed right before the colon before the beginning of a list, refers to the list after the colon. A reference which is placed after the last period in a figure subtitle, refers to the picture included and the whole subtitle. Note that the results from the experiments performed prior to this report, only describe how good the systems are at protecting against attacks with gelatin artificial fingerprints and not against any other attacks. The systems tested, do have other good and bad qualities that must be considered when purchasing a system. The experiments were performed to check the security of the systems with regards to attacks with artificial fingerprints, and not with regards to any other attacks or qualities of the systems.

26 6 Introduction

27 Chapter 2 Biometric overview Biometrics (also known as biometry) is defined as the identification of an individual based on biological traits, such as fingerprints, iris patterns, and facial features [43]. 2.1 Identification and verification Identification and verification (also known as authentication) are both used to declare the identity of a user. Since the two terms identification and verification are easily mixed up, definitions are given below [41]: Identification: In an identification system, an individual is recognized by comparing with an entire database of templates to find a match. The system conducts one-to-many comparisons to establish the identity of the individual. The individual to be identified does not have to claim an identity (Who am I?). [41] Verification (authentication): In a verification system, the individual to be identified has to claim his/her identity (Am I whom I claim to be?) and this template is then compared to the individual s biometric characteristics. The system conducts one-to-one comparisons to establish the identity of the individual. [41] Before a system is able to verify/identify the specific biometrics of a person, the system requires something to compare it with. Therefore, a profile or template containing the biometric properties is stored in the system. Recording the characteristics of a person is called enrollment. [57] The processes of enrollment, verification, and identification are depicted graphically in figure 2.1 on page 8. 7

28 8 Biometric overview Figure 2.1. Enrollment, verification, and identification. [41] Methods of identification and verification As a user, you can be identified or verified on the basis of: Something you know: e.g. a password or a PIN. Something you hold: e.g. a credit card, a key, or a passport. Something you are (biometrics): e.g. a fingerprint or iris patterns. Using something you know and hold are two easy identification/verification solutions widely used today. Using something you know only requires a good memory, but can on the other hand easily be overheard, seen, or even guessed. An item you hold can be stolen and later on used or copied. Using biometrics might at first seem to overcome these problems, since fingerprints, iris patterns, etc. are part of

29 2.1 Identification and verification 9 your body and thus not easily misplaced, stolen, forged, or shared. This report might however give you some new insight about this subject. One way to increase security in an identification/verification system is to combine two or more different identification/verification methods Results from identification and verification procedures When results from identification or verification procedures are discussed, the following terms will be used in this report: Success rate: The rate at which successful verifications or identifications are made compared to the total number of trials. [3] False rejection rate (FRR): The rate at which the system falsely rejects a registered user compared to the total number of trials. [3] False acceptance rate (FAR): The rate at which the system falsely accepts a nonregistered (or another registered) user as a registered one compared to the total number of trials. The FAR is in this report used in the identification version, as a contrast to verification procedures, where it measures if a user is accepted under a false claimed identity. [3] Equal error rate (EER): The common value of the FAR and FRR when the FAR equals the FRR. This is the value where both the FAR and FRR are kept as low as possible at the same time (see figure 2.2). A low EER value indicates a high accuracy of the system. [47] Figure 2.2. The relationship between FRR, FAR, and EER. A big FRR often means a low FAR, and a big FAR often means a low FRR. The small EER value indicates that the security of the system is better. [34]

30 10 Biometric overview 2.2 Biometric techniques Currently, there are many different techniques available to identify/verify a person based on biometrics [57]. These techniques can be divided into physical characteristics and behavioral characteristics. All techniques have in common that acquired data is compared with templates enrolled earlier Physical characteristics The following are examples of biometric techniques based on physical characteristics [3]: Fingerprint recognition: Fingerprint recognition systems scan the fingerprint pattern for recognition. Recognition of hand or finger: Recognition of hand or finger systems scan the entire hand or larger parts of the finger and makes a comparison of patterns in the skin (similar to fingerprint recognition systems). The difference between a fingerprint recognition system and a hand/finger recognition system, lie mostly in the size of the scanner and the resolution of the scanning array. Face recognition: Face recognition systems detect patterns, shapes, and shadows in the face. Face geometry: Face geometry systems work similar to face recognition systems, but focus more on shapes and forms instead of patterns. Vein pattern recognition: Vein pattern recognition systems detect veins in the surface of the hand. These patterns are considered to be as unique as fingerprints, but have the advantage of not being as easily copied or stolen as fingerprints are. Retina recognition: Retina recognition systems scan the surface of the retina and compare nerve patterns, blood vessels and such features. Iris recognition: Iris recognition systems scan the surface of the iris to compare patterns Behavioral characteristics The following are examples of biometric techniques based on behavioral characteristics [3]: Voice recognition: Voice recognition systems use characteristics of the voice, such as pitch, tone, and frequency.

31 2.2 Biometric techniques 11 Signature recognition: Signature recognition systems measure pressure of the pen and frequency of writing to identify a person via a signature. Keystrokes dynamics: Keystrokes dynamics systems use statistics, e.g. time between keystrokes, word choices, word combinations, general speed of typing etc. The authors of the book Handbook of Fingerprint Recognition suggest that all biometric identifiers are a combination of distinctive physiological and behavioral characteristics. For example, fingerprints may be physiological in nature but the usage of the input device (e.g. how a user presents a finger to the fingerprint scanner) depends on the person s behavior. [41]

32 12 Biometric overview

33 Chapter 3 Fingerprints Already at the age of seven months, a foetus fingerprints are fully developed. The characteristics of the fingerprint does not change throughout the lifetime except for injury, disease, or decomposition after death. However, after a small injury on the fingertip, the pattern will grow back as the fingertip heals. [41, 49] This chapter will begin with some important historical events concerning fingerprints, and specifically fingerprints as an identification tool. Then, a short glimpse will be taken at how society today looks at fingerprints. Fingerprint characteristics and enhancement techniques will also be discussed to give the reader a better platform to stand on, before reading the following chapters. 3.1 History It is not justifiable to say that one single person was first to discover fingerprint patterns. Every human being has had papillary lines in front of her eyes for a very long time. It has only been a question of looking down at one s own hands. However, there still exist some important historical events connected to fingerprints, which will be described shortly here. Already in ancient times, fingerprints appeared on pottery and cave paintings in Asia, Europe, and North America to denote authorship or identity [7]. Fingerprints were not described in writing until the 17th century. In 1686, Marcello Malpighi, a professor of anatomy at the University of Bologna, described papillary ridges in his treatise. [7, 44] In 1823, the Czech physician Jan Evangelista Purkyně, classified fingerprint patterns into nine basic types. Purkyně s classification system, laid the foundation for future fingerprint identification systems. [7] 13

34 14 Fingerprints It was not until the later part of the 19th century that fingerprints found its use in personal identification through the two colonials in British India; Sir William Herschel and Dr. Henry Faulds. Dr. Faulds also devised a method of classification. [7, 44] Sir Francis Galton, a British anthropologist and a cousin of Charles Darwin, scientifically proved in the late 19th century that fingerprints do not change over the course of an individual s lifetime, and that no two fingerprints are exactly alike. According to his calculations, the odds of two individual fingerprints being the same, are 1 in 64 billion. Galton identified the characteristics (minutiae) by which fingerprints can be identified, and these characteristics are therefore sometimes referred to as Galton s details today. [7, 44] Galton classified fingerprints as one of the three patterns, arches, loops, and whorls. He found out that approximately 60 percent of all fingerprints are loops, around 30 percent whorls, and the remaining 10 percent are arches. Because of this uneven distribution, Galton then further subdivided the loops into inner and outer loops depending on whether the loop opened up toward the little finger or the thumb. Galton also was the founder of the classical fingerprint cards used in forensics. [7] In 1901, fingerprints were introduced for criminal identification in England and Wales. Galton s observations, and revisions of those by Sir Edward Richard Henry, were used. This was the foundation of the Henry Classification System. [44] In 1918, Edmond Locard wrote that if 12 points (Galton s deatils) were the same between two fingerprints, it would suffice as a positive identification. This is often referred to as the 12 point rule. Different countries have different rules though for identification, including own standards with a minimum number of points. [44] With the introduction of computers in the 20th century, the storing of fingerprint cards became computerized. [44] Sweden has since the 1st of April 2003 abandoned the 12 point rule. Today, a nonnumerical standard is used with no required minimum number of points for positive identification. [12] 3.2 Today Fingerprint usage can be divided into three different areas [3]: Security, as identification of individuals. Forensics, also as an identification method.

35 3.3 Fingerprint characteristics 15 Personal characteristics and dermatoglyphics, often involved with horoscopes and similar nonscientifically proven prophesies. The two first are by far the greatest areas. Fingerprint-based systems, used for security reasons, are so popular today that they have almost become the synonym for biometric systems [41]. Fingerprint-based systems will be further discussed in chapter 4 on page 21. Enormous amounts of information is stored in a fingerprint database. For example, the total number of fingerprint cards (each card contains one impression each of the 10 fingers of a person) in the FBI fingerprint database has now exceeded 200 million, and is growing continuously. Most law enforcement agencies in the world use an AFIS (Automatic Fingerprint Identification System) today. These systems have increased the productivity and greatly reduced the cost of hiring and training human fingerprint experts. [41] Since the discovery of the DNA structure in 1953, DNA has become more and more important in the society as a whole, as well as in forensics. With the science of cloning though, it can be questioned whether or not DNA can actually be used for identification purposes. If individuals can be cloned, DNA typing is as much help as it is in distinguishing identical twins. By definition, identical twins cannot be distinguished by DNA. The same problem does not occur with fingerprints. Even though the fingerprints of identical twins are very similar, automatic fingerprint system can successfully distinguish identical twins though with a slightly lower accuracy than nontwins. It should however be noted that the algorithms in some fingerprint systems may not be robust enough to detect these differences. [7, 35, 41, 51] 3.3 Fingerprint characteristics You have probably looked at your own fingerprint at some point in your life and noticed the papillary lines on it. In fingerprint literature, the terms ridges and valleys are used to describe the higher and lower parts of the papillary lines. The reason we have ridges and valleys on our fingers, is the frictional ability of the skin [48]. The formation of the ridges and valleys is a combination of genetic and environmental factors. The DNA gives directions in the formation of the skin of the foetus, but the exact formation of the fingerprint is a consequence of random events. The exact position of the foetus in the womb at a particular moment, and the exact composition and density of surrounding amniotic fluid, decide how every individual ridge will form. [25] This is also the reason why the fingerprints on different fingers on the same individual are different, and why identical twins have different fingerprints, see section 3.2 on page 14.

36 16 Fingerprints Classification and pattern types Fingerprints can be and have been classified in different ways throughout history, see section 3.1 on page 13. The Henry Classification System was the basis of modern day AFIS classification methods up until the 1990s. In recent years, the Henry Classification System has in most forensic departments been replaced by ridge flow classification approaches. These new classification methods use the distance between core and delta points, minutiae locations, and pattern type (the latter using the Henry Classification System). [22] Fingerprints can be divided into the three major pattern types arches, loops, and whorls, depicted in figure 3.1. Loops are the most common fingerprint pattern [27]. These major pattern types can appear in different variations. For example, you can find plain or tented (narrow) arches, right or left loops, and spiral or concentric circles as whorls. Also, the different pattern types can be combined to form a fingerprint, e.g. a double loop, or an arch with a loop [5]. Figure 3.1. The three major pattern types: arches, loops, and whorls. These major pattern types can be divided further into different subgroups: right or left loops, plain or tented (narrow) arches, and spiral or concentric circles as whorls. There are also combinations of these different pattern types. [22] Terminology To understand the basics of fingerprints, the same approach as [41] uses, will be presented here. A fingerprint can be looked at from different levels; the global level, the local level, and the very-fine level [41]. At the global level, you find the singularity points, called core and delta points, see figure 3.2 on page 17. These singularity points are very important for fingerprint classification, but they are not sufficient for accurate matching [41].

37 3.3 Fingerprint characteristics 17 Figure 3.2. Core and delta points marked on sketches of the two fingerprint patterns loop and whorl. Loops have one delta, whorls have two. Minutiae details are not shown. The number of intervening ridges from delta to core in the leftmost pattern (loop) is 12. A ridge tracing from left to right delta on the rightmost pattern (whorl) determines an inner tracing, meaning that when following a ridge emanating from the left delta, the ridge passes inside the other delta. [7] At the local level, you find the minutiae details (sometimes called minutiae points). One way to classify the minutiae details are in terms of ridge termination, bifurcation, independent ridge, dot or island, lake, spur, and crossover [7]. These are depicted in figure 3.3. The two most prominent minutiae details, are ridge termination (ending) and ridge bifurcation [41]. Figure 3.3. Minutiae details, also known as ridge characteristics, ridge details, or Galton s details. Most of the identifications of fingerprints during this century, were made by matching corresponding minutiae details between two prints. [7] At the very-fine level, you find essentially the finger sweat pores, see figure 3.4 on

38 18 Fingerprints page 18. The position and shape of the pores can be used to help identify a person. To be able to use this information, a high-resolution image of the fingerprint is required. [41] Figure 3.4. Part of a fingerprint image with sweat pores and minutiae details visible. The black lines in the image correspond to the ridges in the fingerprint, and the white lines in the image correspond to the valleys in the fingerprint. The white dots on the ridges correspond to the sweat pores in the fingerprint and are marked with empty circles on a single ridge line. Minutiae details are marked with black-filled circles. [41] Figure 3.5 shows a cross-section of a papillary line. The sweat glands supply the papillary skin with moisture and when touching a surface with a finger, the sweat from these pores is transferred to the pattern of the fingerprint, see figure 3.4. The outer skin layer is called epidermis, and the inner skin layer is called dermis. Figure 3.5. Cross-section of a papillary line of a fingerprint. [48]

39 3.4 Enhancement techniques Enhancement techniques A latent fingerprint results from the reproduction of friction ridges found on fingers. To be able to identify the owner of the fingerprint, the fingerprint must in most cases first be enhanced in order for it to be visible. Enhancing a fingerprint will also be used in the experiments described in chapter 7 on page 55. A print consists of a combination of different chemicals that originate from natural secretions, blood, and contaminants. Some contaminants found in fingerprints result from contact with different materials in the environment. [56] Latent fingerprints can be found on all types of surfaces. In general, surfaces can be characterized as porous, nonporous, or semiporous. Understanding these characteristics helps in deciding the processing technique of the latent fingerprint. [56] Processing techniques In addition to the type of surface, another determining factor in choosing the proper process is the residue of the latent fingerprint, including perspiration, blood, oil or grease, and dust. [56] The condition of the surface also contributes to determining the correct process. Such surface characteristics include dryness, wetness, dirtiness, and tackiness or stickiness. [56] A variety of techniques, including use of chemicals, powders, lasers, alternate light sources, and other physical means, are employed in the detection and development of latent prints. For a detailed description of these different techniques and in which situations to use which techniques, see [56]. Two techniques will though be described more in detail here, since they have been found easily available for nonprofessionals and can be used on nonporous surfaces. Fingerprint powders Powdering is the application of finely ground, colored powder to a nonporous object to make latent prints visible. Powder clings to moisture, oil, and other residues. [56] Different colored powders can be used, e.g. black, white, and gray. The color of the powder depends on the surface, e.g. on a white surface, a black or gray powder will enhance the fingerprint much better than a white powder. The recommended brushes to use with colored powders are fiberglass filament brushes, camel-hair brushes, feather dusters, and squirrel-hair brushes. [1, 56]

40 20 Fingerprints A finely ground magnetic powder can also be used together with a magna brush wand. [56] One important thing to take notice of when using powder and a brush, is to brush in the direction of any ridges that begin to appear. A detailed description of the procedure can be found on pages in [56]. When the fingerprint has been powdered, the print has to be lifted in order to photograph it. If the fingerprint is already placed on a flat surface, you might not have to lift it, but can instead photograph it directly. When lifting the fingerprint, it is important to avoid air bubbles, which will easily form underneath the tape. Cyanoacrylate fuming Cyanoacrylate fuming is also used to develop latent prints on nonporous specimens. This technique is not recommended to perform at home since it includes risks of getting allergic reactions and the fumes are life threatening. It should however be noticed that it is in fact possible to do at home with materials a nonprofessional can buy. Liquid cyanoacrylate can be found in adhesives available at most hobby shops. Since cyanoacrylate fuming was only tried out in the experiments prior to this report, and not used to the same extent as powdering, it will not be described further here. A detailed description of the processing procedure can be found in [56] and a more amateur approach can be found in [20].

41 Chapter 4 Fingerprint scanners Even though the first fingerprint scanners were introduced more than 30 years ago, it is not until the recent years that the interest for fingerprint scanning has increased considerably [41]. With the terrorist attack in New York on September 11, 2001, the US Government and other governments and organizations, became increasingly interested in the biometrics industry. Passport, border control, and identification cards are areas were fingerprints, as a means of authentication, have become increasingly interesting. The fingerprint scanner market has grown rapidly the last years. With this development, the scanners are shrinking in size, the price is going down, and fingerprint systems are being integrated into electronic equipment such as laptops, mouses, and keyboards. A fingerprint scanner has basically two tasks; to acquire an image of a fingerprint, and to decide whether or not this image matches the image of a previously enrolled fingerprint. The decision phase is done by extracting features from the image and then comparing these features to templates stored in a database. A fingerprint contains a lot of information. Storing and using all this information, would take too much space and unnecessary effort when a lot of the information in fact is redundant. Instead, fingerprint scanners focus on the essential information to make the fingerprint as unique as possible and thus useful in identification and verification situations. [3] This chapter will describe the characteristics of a digital fingerprint image, the different scanning techniques used today, the algorithms behind the surface of the scanners, protection schemes, and possible ways of intrusion. 21

42 22 Fingerprint scanners 4.1 Fingerprint images A digital fingerprint image can be characterized by the following main characteristics [41]: Resolution: The minimum resolution for FBI-compliant sensors are 500 dots per inch (dpi), and this is also met by many commercial devices. The sensors used in the extensive experiments have resolutions of 250 dpi and 380 dpi. Area: The larger the area, the more ridges and valleys are captured, and the more distinct the pattern becomes. The minimum area size required by FBI specifications is 1 1 square inches. Many sensors today have an area a lot smaller than that, thus making it impossible for the entire print to be captured. A small area keeps the cost and size down, but does also lead to unnecessary false rejections. The sensors used in the extensive experiments have area sizes of 9.8 9,8 mm, and mm. Dynamic range (or depth): The number of bits used to encode the intensity value of each pixel. Grayscale is used and the FBI standard for pixel bit depth is 8 bits. Some sensors capture however only 2 or 3 bits of information. Geometric accuracy: Can be defined as the maximum geometric distortion introduced by the acquisition device, and is expressed as a percentage with respect to x and y directions. Image quality: Difficult to measure, especially since it is hard to decouple it from the intrinsic finger quality or status. All the characteristics mentioned above work together to set the accuracy of the system. 4.2 Scanning techniques While the first generation scanners used optical techniques, a variety of sensing techniques are used today and almost all of them belong to one of the three families: optical, solid-state, and ultrasound. [41, 57] The main technologies used today are optical and solid-state sensors (mainly capacitive sensors). Solid-state sensors are now gaining great popularity because of their compact size which facilitates in embedding them into laptop computers, cellular phones, smart cards, and the like. [25, 41] Optical sensors The advantages with optical sensors include withstanding temperature fluctuations (to some degree), a fairly low cost, resolutions up to 500 dpi, better image quality,

43 4.2 Scanning techniques 23 and the possibility of larger sensing areas. [24, 41] The drawbacks of optical sensors are size and problems with latent prints [24, 53, 55]. Cuts, abrasions, calluses, and other damage, as well as dirt, grease and other contamination, can also be a problem with optical scanners [29]. Frustrated Total Internal Reflection (FTIR) When you place your finger on an FTIR-based optical sensor (see figure 4.1), the ridges will be in contact with the prism surface, while the valleys will remain at a distance. One side of the prism is illuminated through a diffuse light (a bank of light-emitting diodes (LED) or a film planar light). The light is reflected at the valleys and randomly scattered (absorbed) at the ridges. The lack of reflection from the ridges, makes it possible to acquire an image of the fingerprint. In the early days FTIR sensors, a CCD camera was used to acquire the fingerprint image. Today, the FTIR sensors have shrunk considerably in size and cost with help of the new CMOS technology. [41, 57] Since FTIR devices sense a three-dimensional surface, it is difficult to fool them with a photograph or image of a fingerprint [41]. Latent prints are however still a problem [53, 55]. Furthermore, it is difficult to make a small enough FTIR device suitable to embed into a PDA or a mobile phone, even though they can be used in mouses and keyboards. [41] Figure 4.1. An FTIR-based fingerprint sensor. [41] FTIR with a sheet prism This type of optical sensor, use a sheet prism made of a number of primlets adjacent to each other, instead of a single large prism, see figure 4.2 on page 24.

44 24 Fingerprint scanners With the advantage of size reduction, the quality of the acquired images is however lower than traditional FTIR techniques using glass prisms. [41] Figure 4.2. A fingerprint sensor using FTIR with a sheet prism. [41] Optical fibers This technique uses a fiber-optic plate (see figure 4.3) instead of a prism and lens. The finger is in direct contact with the upper side of the plate, while the lower side of the plate is tightly coupled with a CCD or CMOS camera, which receives the light conveyed through the glass fibers. Since the CCD/CMOS is in direct contact with the plate (without any intermediate lens as in the FTIR techniques), its size has to cover the whole sensing area. High costs will thus be the downside of producing large area sensors with this technique. [41] Figure 4.3. Fingerprint sensing using optical fibers. Residual light emitted by the finger, is conveyed through the glass fibers to the CCD/CMOS camera. [41] Electro-optical These type of sensors, consist of two layers: a light-emitting polymer, and a photodiode array, see figure 4.4 on page 25. When the polymer is polarized with the proper voltage, it emits light that depends on the potential applied on one side. As the ridges touch the surface, and the valleys do not, the potential, and thus

45 4.2 Scanning techniques 25 also the amount of light, will be different. The photodiode array (embedded in glass) receives the light and generates the digital fingerprint pattern. Some commercial sensors use the light-emitting polymer together with an ordinary lens and CMOS instead of the photodiode array. Images acquired electro-optically, are yet not comparable in quality with FTIR images. [41] Figure 4.4. Electro-optical fingerprint sensor. [41] Direct reading A variation of optical sensors are the not so common touchless sensors. Instead of pressuring the finger against a plate, the finger is put on an area with a hole, about 2-3 inches from the optics behind. This technique may seem more hygienic, and saves time by not having to clean the sensor surface. It is however very difficult to obtain well-focused and high-contrast images. [21, 41] Solid-state sensors Solid-state sensors (also known as silicon sensors), were first introduced to overcome the problems with size and cost of optical sensors. However, considering a highsecurity device, a large sensing area is needed, and thus the cost will in fact not be any smaller for solid-state sensors than for optical sensors. [41] All silicon sensors consist of an array of pixels, where each pixel is a tiny sensor itself. Four different types of silicon sensing techniques have been proposed to convert the physical information into electrical signals: capacitive, thermal, electric field, and piezoelectric. [41] Capacitive sensors A capacitive sensor consist of a two-dimensional array of micro-capacitor plates embedded in a chip, see figure 4.5 on page 26. The finger skin works as the other side of each micro-capacitor plate. This way, variations in electrical charge will appear due to distance variations from a ridge on the fingerprint to the sensor, and

46 26 Fingerprint scanners from a valley on the fingerprint to the sensor. These small capacitance differences is then used to acquire an image of the fingerprint. [57] Figure 4.5. Capacitive fingerprint sensor. [41] Even though being widely used nowadays, capacitive sensors do have a number of disadvantages: Small sensor area: It can be questioned whether or not a small image scan area is enough to accurately identify an individual. The reduction in sensor size does also require more carefully performed enrollments. A poor enrollment may not capture the center of the fingerprint, thus forcing the subsequent identification/verification fingers to be misplaced in the same way. The sensing area can of course be increased, however resulting in a higher cost. [24, 29, 41] Electrostatic discharge (ESD): Electrostatic discharges from the fingertip can cause large electric fields that could severely damage the device. [41] Chemical corrosion: The silicon chip needs to be protected from chemical substances (e.g. sodium) that are present in fingerprint perspiration. Protecting the surface with a too thick coating will increase the distance between the pixels and the finger too much and make it more difficult to distinguish between a ridge and a valley. Therefore, the coating must be as thin as possible, yet not too thin, or it will not be resistant to mechanical abrasion. [41] Thermal sensors Thermal sensors are made of pyro-electric material that generates current based on temperature differentials. The temperature differentials between the skin (the ridges) and the air (in the valleys) is used to acquire the fingerprint image. Since thermal equilibrium is reached quickly, it might be necessary to use a sweeping technique when it comes to thermal sensors. Thermal sensors are not sensitive to ESD, nor do they have any problems with a thick (10 to 20 microns) protective coating. [41]

47 4.2 Scanning techniques 27 Electric field sensors The problems optical and capacitive sensor have with dry skin conditions, calluses, cuts, etc. is not the case of electric field sensors. These sensors enter the skin and creates a fingerprint image from below the damaged surface layer. The variations of the electric field is measured in the conductive layer, the boundary between the outer layer of damaged skin and the pristine skin. [4] Piezoelectric (pressure) The sensor surface is made of a non-conductive dielectric material. When pressure is applied by the finger, a small amount of current, dependent on the pressure, is generated (this effect is called the piezoelectric effect). The different pressure from the valleys and ridges, therefore result in different amounts of current. One of the disadvantages of this technique, is the materials used, which are often not sensitive enough to detect the differences between ridges and valleys. Additionally, the protective coating blurs the resulting image. [41] Ultrasonic sensors In an ultrasonic sensor (see figure 4.6), a transmitter sends acoustic signals toward the fingertip, and a receiver detects the echo signals which bounce off the fingerprint surface. The difference in acoustic impedance of the skin (ridges) and the air (valleys) is used to measure the distance, thus acquiring an image of the fingerprint. The frequency range used by these sensors, varies from 20 kilohertz to several Gigahertz. The top frequencies are required to get the required resolution to be able to differentiate fingerprints from each other. [41, 57] Figure 4.6. An ultrasonic sensor uses sound waves which penetrate materials and give a partial echo at each impedance change. [41] It has been stated that the improved image quality from ultrasonic sensors results

48 28 Fingerprint scanners in accuracy rates approximately a factor of 10 better than any other fingerprint sensing technology on the market today. [29] Except electric fields, ultrasound is one of the few technologies that images the subsurface of the finger skin, thus penetrating dirt, grease, etc. on the sensor surface and the finger. Ultrasound technology, though considered perhaps the most accurate of the fingerprint technologies, is not yet widely used due to large size and a quite high cost. Moreover, it takes a few seconds to acquire an image. [24, 41] 4.3 Touch versus sweep Most sensors used today are touch sensors (area sensors). When using a touch sensor, you simply put your finger on the sensor and hold it for a few seconds without moving it. Very little user training is required to use a touch sensor. However, there are a few drawbacks with touch sensors as well: The sensor quickly becomes dirty and must be cleaned. Some users might have issues with using the device if it does not look clean. [41] Problems with latent prints exist. Depending on the type of sensing technique, studies have shown that it is possible to reactivate a latent print on a fingerprint sensor. [53, 55] Rotation of the finger may be a problem for recognition. Some matching algorithms do not accept large rotations (e.g. more than 20 degrees) of the finger. [41] Tradeoff between cost and size of the sensing area. This is especially true for solid-state sensors, where the cost mainly depends on the area of the chip die. [41] Because of these drawbacks, a new type of sensor was introduced: the sweeping sensor, see figure 4.7 on page 29. Sweeping sensors are as wide as a finger, but only a few pixels high. Therefore, the main advantage of sweeping sensors, especially in silicon sensors, is reduced cost. The sweeping consists of a vertical movement only. At the end of the swipe or on-the-fly, the fingerprint image is reconstructed from all the images earlier acquired. [41] The sweeping method was originally introduced in conjunction with thermal sensors, but is nowadays used in many different types of sensors. Unlike touch sensors, sweeping sensors look clean since each user s finger cleans the sensor during sweeping. No problem with latent prints exist with sweeping sensors, and in most cases, rotation of the finger is neither a problem. Sweeping sensors do still have some drawbacks as well [41]:

49 4.4 Algorithms in fingerprint scanners 29 Figure 4.7. When a user is sweeping his/her finger on a sweeping sensor, a number of image slices are combined to form an image of the entire fingerprint. [41] Learning time. It takes a number of tries, before a user gets used to sweeping properly (i.e. without sharp speed changes, or discontinuity). The interface must be able to capture a sufficient number of fingerprint slices to follow the finger sweep speed. Reconstructing the fingerprint image from the slices is a time consuming process which usually produces errors. 4.4 Algorithms in fingerprint scanners A typical fingerprint recognition system (see figure 4.8) consists of a scanning device (capture and enhancement), a feature extraction part, and a comparison part where an identification/verification decision is taken. This section will shortly describe these different parts in more detail. Figure 4.8. Typical structure of a fingerprint recognition system. [3, 42]

50 30 Fingerprint scanners Image enhancement When a fingerprint image is captured, it contains a lot of redundant information. Problems with scars, too dry or too moist fingers, or incorrect pressure must also be overcome to get an acceptable image. Therefore, a number of filters, some of which will be described below, are applied to the image. [3] Normalization: By normalizing an image, the colors of the image are spread evenly throughout the gray scale. A normalized image is much easier to compare with other images, and the quality of the image is easier determined. [3] Binarization: Making an image binary, transforms the gray scale image into a binary image (black and white). Either a global or localized threshold value is used. [3] Low pass filtering: The process of low pass filtering smoothens the image to match the pixels nearby so that no points in the image differ from its surroundings to a great extent. By low pass filtering an image, errors and incorrect data are removed, and it simplifies the acquisition process of patterns or minutiae. [3, 27] Quality markup: Redundant information needs to be removed from the image before further analysis can be performed and specific features of the fingerprint can be extracted. Therefore segmentation, i.e. separating the fingerprint image from the background, is needed. Furthermore, any unwanted minutiae (can appear if the print is of bad quality) needs to be removed. [3, 27] Feature extraction and comparison Many algorithms have been developed to match two different fingerprints and they can be divided into the following groups: Minutiae-based matching: This is the most popular and widely used matching method, partly because it is the same technique as used by fingerprint examiners. As described in section on page 16, a fingerprint pattern is full of minutiae points, which characterize the print. In minutiae-based matching, these points are extracted from the print, stored as sets of points in the two-dimensional plane, and then compared with the same points extracted during the enrollment phase. It is very unlikely that the fingerprint during enrollment and the fingerprint during identification/verification had the exact same angle, horizontal and vertical placement. Therefore, the core point (see section on page 16), is used as a reference point for the coordinate system and the distance and angle from the core point is calculated and used

51 4.5 Sensor attacks and protection schemes 31 for each minutiae point. For identification/verification a certain number of minutiae points should match for the user to be successfully logged in. [3, 41] Correlation-based matching: The fingerprint image to be identified/verified, is superimposed with the fingerprint image acquired during the enrollment. The correlation between corresponding pixels is computed for different alignments (e.g. various displacements and rotations). [41] Ridge feature-based matching: This matching method uses features of the ridge pattern, e.g. local orientation and frequency, ridge shape, and texture information. Even though minutiae-based matching is considered more reliable because of its indistinctness, there are cases where ridge feature-based matching is better to use. In very low-quality fingerprint images, it can be difficult to extract the minutiae points, and using the ridge pattern for matching is then preferred. Ridge feature-based matching can be conceived as a superfamily of minutiae-based matching and correlation-based matching. [41] 4.5 Sensor attacks and protection schemes A biometric system can be defeated in different ways, from attacks at the sensor level, to replay attacks on the data communication stream, and attacks on the template database [55]. This report focuses on the attack at the sensor level, which can be performed in a number of ways. Independent of the attack method, the false acceptance rate (FAR) and the false rejection rate (FRR), should be kept low Registered finger The intruder can force the legitimate user to press his/her finger against the fingerprint sensor under duress [42]. Also, the intruder can give the legitimate user a sleeping drug, in order to use either the finger directly against the sensor, or by making a mold of the finger as described in [3, 42, 57]. Another way of using the registered finger, is by separating the finger from the legitimate user s body [42]. This finger can then be used directly on the sensor, or by one of the methods described in [42]. To make it more difficult to attack a fingerprint system as described above, a fingerprint scanner can be combined with another authentication method, e.g. a PIN, a password, or an ID card. Another way to deter these crimes, is by using a way to alarm when under duress, e.g. with help of a special secret code or manner. Also, a two-persons control, where the system requires e.g. fingerprints from two different persons, would be helpful. Using a two-persons control is however very inefficient and not realizable in most situations. [42]

52 32 Fingerprint scanners Unregistered finger This attack means that the intruder uses his/her own finger to try (intentionally or unintentionally) to log in as another user. An important indication for how easy this type of attack is on a special system, is the false acceptance rate, described in section on page 9. Also, by knowing the pattern type of the legitimate user, an intruder with the same pattern type (see section on page 16) will have a higher probability of successfully logging in as the legitimate user. [42] A twin s fingerprint or a genetic clone As described in section 3.2 on page 14, the fingerprints of identical twins are very similar, even though not identical. Using a genetic clone of a fingerprint or the identical twin s fingerprints to deceive a system could be possible if the algorithm used is not robust enough to distinguish the live finger from the intruder s finger. The attack with a genetic cloned finger could be detected with help of a liveness detection mechanism in the system. Using a combination with another authentication method, or using a two-persons control, would also be helpful to deter these crimes. However, protection against the identical twin is not as easy as protection against a genetic clone. [42] Artificial fingerprint An artificial fingerprint, is a fingerprint made to imitate a real (living) fingerprint. It can be made of gelatin, silicone, play-doh, clay, or other materials. There are two ways to make an artificial fingerprint; either by directly making a mold of the legitimate user s finger, or by using a residual fingerprint to produce an artificial fingerprint [42]. The experiments prior to this report, focus on attacks with artificial fingerprints made from a residual fingerprint. Again, liveness detection in the system, a combination with another authentication method, or a two-persons control would be helpful to deter these crimes. [42] Others Some fingerprint systems can be fooled by flashing a light against the scanner, heating up, cooling down, humidifying, impacting on, or vibrating the scanner outside its environmental tolerances. Another way to fool a fingerprint system is to use a residual fingerprint on the sensor surface to reactivate the fingerprint. This can be done by breathing on the sensor s surface, placing a thin-walled water-filled plastic bag on the sensor s surface, or by dusting graphite powder and then pressing an adhesive film on the sensor s surface. On an optical sensor, the method with

53 4.5 Sensor attacks and protection schemes 33 graphite powder and adhesive film can be used together with a halogen lamp to create a kind of snow blindness in the sensor. [42, 55] To protect against attacks using reactivation of a latent fingerprint, a sweeping sensor can be used instead of an area sensor.

54 34 Fingerprint scanners

55 Chapter 5 Liveness detection Liveness detection (sometimes called vitality detection) in a biometric system means the capability for the system to detect, during enrollment and identification/verification, whether or not the biometric sample presented is alive or not. Furthermore, if the system is designed to protect against attacks with artificial fingerprints, it must also check that the presented biometric sample belongs to the live human being who was originally enrolled in the system and not just any live human being. Many people believe that biometric systems can detect liveness in biometric samples. Some manufacturers of biometric system also claim that they have liveness detection in their system. It has however been shown that fingerprint systems can be fooled with artificial fingerprints, that static facial images can be used to fool face recognition systems, and that static iris images can be used to fool iris recognition systems [23]. After a general description of the concept liveness detection in biometric systems, this chapter will focus on liveness detection in fingerprint systems, which techniques fingerprint scanners can use to detect liveness, and how these fingerprint scanners can be fooled. This chapter also includes a discussion of liveness detection, as compared to the other chapters in the thesis which are discussed in chapter 9 on page Liveness detection in biometric systems Liveness detection can be performed either at the acquisition stage, or at the processing stage. For example, an optical fingerprint scanner would create an image of an eraser, but not extract any features; the liveness detection takes place at the processing stage. A capacitive fingerprint sensor on the other hand, would not even 35

56 36 Liveness detection create an image of the eraser; the liveness detection takes place at the acquisition stage. [23] There are two approaches in determining if a finger is alive or not; liveness detection and non-liveness detection. The material or data used to spoof a system often have a number of different non-liveness characteristics that could be used to detect nonliveness. An example of a non-liveness detection detection method would be to detect air bubbles in gelatin artificial fingerprints. Most biometric systems today have a decision process which first checks liveness: if data = live perform acquisition and extraction else if data = not live do not perform acquisition and extraction This means that an intruder has the simpler task of imitating a live finger than circumventing a non-liveness detection mechanism. In fact, any detection mechanism can and will be defeated according to [23]. There are essentially three different ways to introduce liveness detection into a biometric system [51]: Using extra hardware to acquire life signs. Using the information already captured by the system to detect life signs. Using liveness information inherent to the biometric. The first of these methods introduces a few other problems; (1) it is expensive, (2) it is bulky, and (3) it could still be possible to present the artificial fingerprint to the fingerprint sensor and the real fingerprint of the intruder to the hardware that detects liveness. Also, in some cases it is still possible to fool the additional hardware with a wafer-thin artificial fingerprint. The second method does not have these disadvantages, except maybe that it could be possible to still fool with an artificial fingerprint. It is on the other hand a bit more complicated to extract the life signs using no additional hardware. The third method of using inherent liveness information to the biometric, is not applicable to fingerprint recognition. Other biometric systems including facial thermograms, gait, body odor, keystroke dynamics, etc. use this however. These technologies are not widely implemented and still need to be validated as reliable biometric identifiers [34]. The main problem of distinguishing between an artificial fingerprint and a real fingerprint, is that the epidermis (outer skin) of the finger is in fact not alive either. Many different techniques have been suggested to detect liveness, and some of them will be presented in the following sections. For each of these techniques, a method to fool the system with an artificial fingerprint will also be suggested.

57 5.2 Using extra hardware Using extra hardware The main problem with liveness detection methods based on extra hardware, is that the scanners have to be adjusted to operate efficiently in different kinds of environments, leading to problems when using a wafer-thin artificial fingerprint glued on to a live finger. Furthermore, using extra hardware will in many cases be inconvenient for the user. Only in the US, many patents have been filed the last years concerning liveness detection in fingerprint scanners based on physiological properties. Some of the most common methods will be described here. Most of these systems are not available commercially Temperature The temperature of the epidermis is about C. When using a thin silicone artificial fingerprint, this results in a decrease by a maximum of 2 C of the temperature transfer to the sensor. Obviously, it will not be difficult to have the temperature of the artificial fingerprint within the working margins of the sensor. Sensors that are used outdoors often have a broader working margin, giving the intruder even better prerequisites. [57] Even though [57] writes about silicone artificial fingerprints, it can be expected that the same reasoning can be done when it comes to gelatin artificial fingerprints Optical properties Optical sensors can use the optical properties of human skin versus other materials as a liveness detection method. These properties include e.g. absorption, reflection, scattering, and refraction under different lighting conditions (such as red, blue, green, infrared, laser lights). A gelatin artificial fingerprint does however have optical properties which are very similar to human skin. [41] Pulse The pulse in the tip of the finger can be detected and used as a liveness detection method. With a wafer-thin artificial fingerprint, the underlying finger s pulse will however be sensed. Also, practical problems arise due to changes in the pulse. A person with a pulse of 40 beats per minute implicates that the finger must be held for at least four seconds on the sensor for the pulse to be detectable. The same person could have a pulse of 80 beats per minute if he or she worked out immediately before the fingerprint scanning. The emotional state of the person also affects the pulse. [57]

58 38 Liveness detection A US patent entitled Anti-Fraud Biometric Sensor that Accurately Detects Blood Flow by SmartTouch LLC describes how two light emitting diodes (LEDs) and a photo-detector are used to determine whether blood is flowing through the finger. Earlier similar solutions have been possible to fool by simulating blood flow (through the use of a flashing light or by moving the imposters finger). This patent declares to have solved these problems by checking if the background light level is above a threshold and by detecting movement of the finger. This liveness detection method basically implements pulse oximetry, but only uses the pulse rate information, see section [37] Pulse oximetry Pulse oximetry is used in the medical field to measure the oxygen saturation of haemoglobin in a patient s arterial blood. A pulse oximeter also measure the pulse rate. The technology involved is based on two basic principles. First, haemoglobin absorbs light differently at two different wavelengths depending on the degree of oxygenation. Second, the fluctuating volume of arterial blood for each pulse beat adds a pulsatile component to the absorption. [11] Detection of pulse oximetry can be fooled using a translucent artificial fingerprint (e.g. gelatin) which covers only the live finger s fingerprint. The pulse oximetry will measure the saturation of oxygen of haemoglobin in the intruder s finger s blood. [51] Blood pressure Apart from the same disadvantages as with measuring the pulse, measuring blood pressure adds another problem. The sensors available today (excluding the single point sensors that must be entered directly in the vein), require measurement at two different places on the body, e.g. on both hands. Also, blood pressure measurement devices are easy to fool by using a wafer-thin artificial fingerprint and the underlying finger s blood pressure Electric resistance The electric resistance of the skin can range from a couple of kilo-ohms to several mega-ohms depending on the humidity of the finger. With some people having dry fingers, and others being sweaty, it is easy to realize that the span of allowed resistance levels will be great enough for an intruder to easily fool the system. For example, by putting some saliva on the silicone artificial fingerprint, the system will be fooled into believing it is the live finger. [57]

59 5.2 Using extra hardware 39 In [42], the electric resistance was measured to 16 MOhms/cm (sic!) in a live finger and 20 MOhms/cm (sic!) for the corresponding gelatin artificial fingerprint. In other words, the difference is so small between the two that it would be impossible to create liveness detection with this method without getting a too high FRR. Matsumoto and colleagues also showed that a live finger has a moisture level of 16 %, while a gelatin fingerprint has a moisture level of 23 % [42]. Since the moisture affects the resistance, and the weather conditions and psychological conditions can change the dryness or sweatiness of human skin, the difference in moisture level between live fingers and gelatin artificial fingerprints, is small enough to be able to fool sensors with gelatin prints Relative dielectric permittivity The relative dielectric permittivity (also known as relative dielectric constant or RDC), is a measurement of the degree to which a medium resists the flow of electric charge divided by the degree to which free space resists such charge [50]. The different values of RDC between a live finger and an artificial fingerprint is the basis of this liveness detection method. Just like electric resistance, the RDC is also affected by the humidity of the finger, so to get an acceptable FRR, the range of acceptable RDCs will include the RDC of a gelatin fingerprint. An artificial fingerprint made of silicone on the other hand, has to be prepared with a solution of 90 % alcohol and 10 % water to fool a system. The RDCs of alcohol and water are 24 and 80 respectively, while the RDC of human skin has a value in between these. Since the alcohol will evaporate quicker than water, the RDC will soon be within the acceptance range of the sensor. [42, 57] Combining ECG, pulse oximetry, and temperature A US patent from 1998, suggests using one or preferably more biometrical features for liveness detection [46]. Many examples of non-specific biometric parameters are given, but most preferably a combination of pulse oximetry, electrocardiography (ECG), and a temperature sensor is used. A CCD camera is used for the fingerprint identification/verification, and the skin temperature, pulse (both from ECG and optical readings which should correlate), and oxygen saturation of haemoglobin in the arterial blood, are used for a liveness measurement. As mentioned earlier, the temperature sensor can be easily fooled with an artificial fingerprint. Also, detection of pulsation, pulse oximetry, and electrocardiogram can be fooled using a translucent artificial fingerprint (e.g. gelatin) which covers only the intruder s live finger s fingerprint [51]. Additionally, because of the ECG sensor, the user has to hold his/her finger still for six to eight seconds. This is quite a long time when it comes to these types of

60 40 Liveness detection applications. If the user moves the finger, the measurement has to be started all over again. Because of this and other various reasons, the project was discontinued. [34] Detection under epidermis The fingertip consists of bone, fatty tissue, and two layers of skin; the epidermis (the outer layer) and the dermis (the inner layer). The fingerprint pattern is found not only in the epidermis, but the exact same pattern can also be found in between these layers, and this is the information that some liveness detection systems use. [38] Two different kinds of sensors using information underneath the epidermis, will be mentioned here; the ultrasonic sensor, and the electric field sensor. Ultrasonic sensors focus on the fact that the underlying layer is softer and more flexible than the epidermis. Electric field sensors focus on the higher electric conductivity of the layer underneath the epidermis as compared to the epidermis. Even though these types of sensors use the information underneath the epidermis, this does not mean that they cannot be fooled with an artificial fingerprint. In [57] it is suggested that with a knowledge of the liveness detection method used, two different layers of artificial fingerprints with the appropriate characteristics, can be created to fool the scanner. E.g. to fool an ultrasonic sensor, first a more flexible and soft print is made, and then a second regular artificial print is made and attached to the first while making sure that the two line patterns are in exact matching positions. Matching the patterns should be no problem for a dental technician. [57] Interesting to note here is that in the experiments performed, one electric field sensor (in two different scanners) was tested in the extensive experiments, and four different electric field sensors were tested at CeBIT. All of them were deceived without using a two-layered artificial fingerprint, but simply an ordinary gelatin artificial fingerprint. What it does not show however is whether or not ultrasonic sensors could be deceived in the same easy manner. Testing an ultrasonic sensor by a third party with regards to artificial fingerprints is needed Other claims Some manufacturers claim to have other liveness detection methods than the ones described in this section. Additionally, some of them refuse to reveal that secret method. Security by obscurity (keeping things secret by keeping the method secret), will make the system more difficult to break in the beginning, but will in the end often be broken and is therefore not appropriate as a security method. [57]

61 5.3 Using existing information Using existing information To the author s knowledge, there exists only one thoroughly researched method for liveness detection using existing information today. This method, using perspiration as liveness detection, is therefore the only method which will be presented in detail in this section. Some other methods will however first be mentioned Skin deformation This liveness detection method uses the information about how the fingertip s skin deforms when pressed against a surface. If for example, the user is required to place his/her finger on the sensor twice, or to move it once it is in contact with the sensor surface, there will be some non-linear distortions between the two fingerprint impressions. Using a comparably thick artificial fingerprint with the same type of requirements, will only give a rigid transformation between the two fingerprint impressions. Using a thin artificial fingerprint glued on to a live finger, will on the other hand still produce quite similar non-linear deformations as a live finger would. [41] Pores By using a fingerprint sensor which can acquire an image of the print with a very high resolution, it is possible to use details in the fingerprint, such as sweat pores, as a liveness detection method [41]. These fine details might be difficult to copy in artificial fingerprints. According to [41], the work by Matsumoto et al. [42], showed that a coarse reproduction of intra-ridge pores is feasible with gelatin artificial fingerprints. The experiments performed prior to this report, showed that it is difficult to reproduce the exact size and position of the pores on the mold and thus also on the gelatin print, see section on page 79. The pores can however be coarsely reproduced, and even this should make you think twice before using a fingerprint device which uses the position and size of pores as liveness detection Unique characteristic for each individual In [41], the authors argue that a good liveness detection method should depend on characteristics that are unique to each individual and which are also difficult to copy. They suggest a method where the recognition is done using the ordinary print on the fingertip, but when it comes to liveness detection, a side impression (near the nail) which has been enrolled earlier, should also be subject to recognition. [41] The advantage of this method is that people usually do not leave their side impressions as latent prints very often. Therefore, the problems with artificial fingerprints

62 42 Liveness detection made from latent prints will dramatically decrease. Also, this method would not require any major changes in the software of the fingerprint scanner. [41] The main drawback with this method is artificial fingerprints created with help of a willing or coerced legitimate user. Also, the acquisition and processing time would be longer Perspiration The Biomedical Signal Analysis Laboratory at West Virginia University, USA, is developing a liveness detection algorithm which is based on the detection of perspiration in a time progression of fingerprint images. To be able to fully understand the algorithm developed at the West Virginia University, a theory background of perspiration will be presented first. Skin characteristics There are about 600 sweat glands per square inch, and the sweat (a dilute sodium chloride solution) diffuses from the sweat glands on to the surface of the skin through small pores. Skin pores do not disappear, move, or spontaneously change over time. The pore-to-pore distance is approximately 0.5 mm over the fingertips. [9] Sweat has a very high dielectric constant and electrical conductivity compared to the lipid-soluble substances absorbed by the outmost layer of the skin. Generally, the dielectric constant of sweat is around 30 times higher than the lipid. [9] Fingerprint scanner When laying a fingertip with moist skin on a capacitive sensor, the capacitance will be much higher (resulting in a darker captured image), than if the skin was not moist. The reason is the high dielectric constant of sweat. Because of this, capacitive scanners are specifically suited for detection of perspiration. [9] Perspiration over time In live fingers, perspiration starts from the pores. The sweat then diffuses along the ridges during time, making the semi-dry regions between the pores moister or darker in the image. The perspiration process does not occur in cadaver fingers or artificial fingerprints. [9] There are mainly two ways to use the perspiration information. Either you can use the fact that perspiration starts from the pores (static approach), or you can use the

63 5.3 Using existing information 43 fact that perspiration changes the image darkness over time (dynamic approach). [9] The algorithm The algorithm maps a 2-dimensional fingerprint image to a signal which represents the gray-level values along the ridges. A pair of consecutive fingerprints are captured in 5 seconds. The last image collected (at time 5 s) is used to determine the location of the ridges, since it usually has darker ridges and better quality. Variations in gray levels in the signal correspond to variations in moisture both statically (in one image) and dynamically (difference between first and last image). A fourier transform of the signal (see figure 5.1) is used to quantify the static variability in gray level along the ridges due to the pores and the presence of perspiration. In particular, the algorithm focuses on frequencies corresponding to the spatial frequency of the pores. Secondly, the dynamic features quantify the change in the local maxima and minima in the ridge signal. [9] Figure 5.1. West Virginia perspiration detection method. The two plotted lines are the capacitance plots across a ridge of the live finger, measured five seconds apart. The local maxima in the plot corresponds to the pores in the fingerprint ridge that are saturated with moisture. The two sensor readings (solid line=initial reading and dashed line=reading after five seconds), show that the areas between the pores tend to fill up with perspiration over time as the moisture spreads across the ridges. If this tendency cannot be observed, the fingerprint is assumed to be fake. [9, 34] The algorithm develops one static measure and four dynamic measures. Classification can be performed based on each of the individual measures developed. While the individual measures give equal error rates (EERs) of between 5.56 % and %, much lower EERs can be achieved by combining all measures. To classify the finger as live or fake/dead, a back-propagation neural network (BPNN) is used with the static measure and dynamic measures as input. [9]

64 44 Liveness detection Testing of the algorithm Using a training set of 12 subjects each for live fingers, and equally many cadaver fingers and artificial prints, the BPNN was trained using a number of iterations. A special criterion can be set to decide how good the BPNN should be, and the BPNN is then trained with as many iterations as needed until this criterion is met. The artificial fingerprints were made from play dough using rubber-based casts. Using a test set of six live fingers, and equally many cadaver and spoof fingers, the BPNN classified all of the cases correctly. [9] Discussion of perspiration This liveness detection method does only require a software upgrade and not any extra hardware. There are however a few things to keep in mind before actually starting to use this liveness detection method outside the laboratory. First of all, more testing is needed, and the team that developed the algorithm has already started working on this. Still, one could argue that the algorithm should also be tested by an independent party. The perspiration process will be somewhat different between different subjects, and will also depend on the initial moisture level of the skin. Therefore, a much larger test set is needed, including subjects with different skin conditions in different climates and seasons as well as a more diverse background (race, age, etc.). This is the work of an ongoing study. [9] Perspiration disorders (finger too moist or too dry) and other abnormal skin conditions could be a problem when it comes to the algorithm developed. However, people with these skin conditions usually have problems with using fingerprint scanners in general. When it comes to too moist fingers, wiping the finger before scanning, is sufficient to overcome these problems. [9] The algorithm could be optimized to make it more time efficient [9]. Also, the time issue when using a scanner is crucial. Tradeoffs between precision and speed of liveness verification/identification will have to be made [9]. How close can two captures be made for the algorithm to still work moderately? To keep your finger on a scanner for five seconds is a comparably long time and this time must be decreased to make the method worth using. The current algorithm averages data from the whole image. Using e.g. the top 25 % of ridge signals which exhibit the most variation could optimize the algorithm. Another optimization could be to develop device specific algorithms. [16] More research is needed in the area of creating artificial fingerprints which have pores and imitate the perspiration process.

65 5.4 Testing of liveness detection methods Testing of liveness detection methods Many liveness detection methods have been suggested and some have been implemented. An independent testing on the effectiveness of these methods for performing liveness detection is still missing. Liveness detection is only a part of the fingerprint recognition system, and must therefore be tested with regards to the effects it has on the whole system when it comes to FAR, FRR, failure to enroll, and other statistics. In addition, user convenience, universality, cost, time, etc., need to be considered for evaluation. [51] 5.5 Relevance of liveness detection Imagine a system that is used for passport bearer authentication, air travel authentication, or access to nuclear facilities. This system would require a high level of security and the consequences would be severe if the system was defeated. In fact, all these usage areas have been considered for biometric applications and might in a few years be reality. In a system requiring a high security level, liveness detection would be necessary if the system was not combined with other authentication methods or other biometrics. On the other hand, a system used for logging in at your home computer might not be interesting enough for an intruder to spend days in creating an artificial fingerprint. In this case, the ease of use of the system is probably more important for the user than having a system with liveness detection. Stephanie Schuckers, Ph.D. at the Clarkson University and West Virginia University in USA, puts the usage of fingerprints in perspective: While someone could steal and make a copy of my office key to gain unauthorized entry, this does not discredit the use of keys [51]. However, you do not leave your keys all over the place for anyone to gain access to them like you do with your fingerprints. Furthermore, a key and lock is much easier to change than a fingerprint. 5.6 Other methods to limit spoofing Liveness detection has been thoroughly discussed in this chapter, but there are also other methods to limit the impact of attacks with artificial fingerprints on fingerprint systems. Some of these other methods will be presented in this section Multiple snapshots of the same finger By taking multiple snapshots of the same finger and requiring each of these to be identified/verified correctly, the FAR when it comes to artificial fingerprints,

66 46 Liveness detection could be decreased. This would probably increase the FRR however, leading to inconvenience for ordinary users of the system Multiple fingers If users were required to enroll more than one finger, then identification/verification can be performed in two different ways to increase the security of the system. The first scenario involves randomization of requested fingers to identify/verify. The other scenario involves requesting all fingers enrolled for identification/verification. [23] Challenge-response Challenge-response is another method to determine the presence of a person. The response can be either voluntary (behavioral) or involuntary (reflexive). In a voluntary challenge-response system, the user will hear, see, or feel something and do something in response. In an involuntary challenge-response system, the user s body automatically responds to a stimulus. Examples of this are muscles responding to electrical stimulation, the dynamic change in the color of skin when pressure is applied, and the reflex of a knee when struck. [34] An implemented example of an involuntary challenge-response is found in the US patent Detector for recognizing the living character of a finger in a fingerprint recognizing apparatus by Kallo et al. The method involves a small impulse current being applied to the finger and the finger s electrical reaction to the impulse is the involuntary response. If the signals returned by the finger are outside the predefined range of acceptable values, the fingerprint is assumed to be fake. Guardware Systems Ltd. uses this patented liveness detection method in their products. [36, 34] One problem with this method is to know that the person is in fact the same as the true owner of the fingerprint presented to the sensor. Furthermore, methods involving shocking for example, are probably not comfortable for the users. [51] Supervision Both identification, verification, and enrollment, can be subject to supervision to increase security. It will be more difficult to bypass a system when being watched. Still, when using a thin transparent gelatin artificial print glued on to a live finger, it should be very difficult for a supervisor to detect it, especially if being tired after having supervised the system many hours already without detecting anything suspicious.

67 5.7 Additional comments Multi-modal biometrics By using multi-modal biometric systems, i.e. systems combining two or more biometric techniques, security can be increased. For example, by combining fingerprint recognition with iris recognition, the security of the system will be higher than using one of those systems alone. When it comes to fooling a multi-modal biometric system with artificial biometrics, it will be more difficult to create both an artificial fingerprint and an artificial iris that will be accepted by the system. Much research is being performed in the area to find the best way to combine the available biometric methods. Except choosing the biometric traits to combine, the method to combine them must also be decided. The methods can for example be combined at the feature extraction level or at the decision level. [51] According to [23], implementing multiple biometrics is currently much more difficult than it seems. Reasons mentioned are environmental issues, cost, and equipment limitations Multiple identification/verification methods Combining a biometric method with something you know and/or hold, will increase the security of the system. For example, a fingerprint recognition system can be combined with a smart card and a password. To deceive that system, the intruder would have to get hold of both the smart card, the password and the fingerprint. Furthermore, if the fingerprint template is stored on the smart card, it will be impossible for an intruder to attack the non-existent centrally stored template database. The downside of this method is the inconvenience for the user of the system, possible high FARs and FRRs, and that it is not possible to use in identification systems [23]. 5.7 Additional comments The discussion about liveness detection has yet been very hidden and companies do often not openly discuss their liveness detection solutions, if they have any. What kind of liveness traits that are measured is of less relevance than how the liveness detection method is implemented. If possible, it is best to acquire the fingerprint information and the liveness information simultaneously, i.e. at the same time and place. Furthermore, to stay one step ahead of adversaries, living techniques that evolve over time could be used for liveness detection. [34] It must always be remembered that liveness detection is not a definite solution to a perfectly secure system. Liveness detection does minimize the risk of successful

68 48 Liveness detection attacks with artificial fingerprints, but it is not guaranteed that it can not be broken. When considering the security of a fingerprint recognition system, the entire system must be looked at, and not only attacks with artificial fingerprints. Comparing the security of fingerprint recognition systems with other biometric systems or other authentication methods, all systems will have its weaknesses and strengths and it is impossible to say that one system is the best.

69 Chapter 6 History of artificial fingerprints The work of creating artificial fingerprints, started earlier than most people know. This section will give an historical overview of how artificial fingerprints have developed over the years and how successful they have been previously in fooling fingerprint systems. It should be noted that the degree of how scientific the following works are, varies. 6.1 Albert Wehde s work According to [7], it was already in the 1920 s that Albert Wehde, an engraver, photographer, former self-described political prisoner, and worker for the identification bureau at Leavenworth, devised a way to forge fingerprints. (It was common practice to employ prison inmates as fingerprint clerks.) Wehde s training as an engraver and photographer came in handy for his method. He left a fingerprint impression in grease on a piece of black tin, dusted the print with white powder, and photographed it. Finally, he made a copper etching of the negative. The copper plate could then be used to forge latent prints. [7] Notice that the information about Wehde s forgery comes originally from the book Finger-Prints Can Be Forged, written by Wehde himself together with John Beffel, a radical journalist. [7] 49

70 50 History of artificial fingerprints 6.2 Six biometric devices point the finger at security By D. Willis and M. Lee, Network Computing, June Fooling fingerprint scanners with help of artificial fingerprints is not a new method. Already in 1998, David Willis and Mike Lee at the Network Computing magazine, noted that artificial fingerprints can be created by making a wax mold and using silicone to create the artificial fingerprints [60]. Their artificial fingerprints fooled four out of six tested fingerprint scanners. The testing team also managed to get past two of the six fingerprint scanners using another method. Latent prints on a table were enhanced with help of a fine brush and dry toner from a laser printer cartridge and then lifted with adhesive tape. The images were transferred to a transparency material on a photocopier and by wetting the ink side of the transparency, this could be used to fool the scanners. [60] 6.3 Biometrical fingerprint recognition: don t get your fingers burned By T. Putte and J. Keuning, September Whether or not Wehde s or Willis and Lee s work inspired Ton van der Putte and Jeroen Keuning in 2000 to do further experiments, is not known. Putte and Keuning used two methods for creating artificial fingerprints [57]: Duplication with cooperation: Using plaster, a mold is formed. The mold is filled with silicone and a pounder is used to make the artificial finger waferthin. Duplication without cooperation: A powder and a brush is used to enhance the latent fingerprint. Scotch tape is used to remove the powder from the background. The tape is placed on the photosensitive side of a film, and with help of a camera, a photo is taken. The negative is attached to a printed circuit board (PCB), exposed to UV light, and the PCB is developed and etched. The slim profile of about 35 micron is deepened and the mold is finished. Again, silicone is used to create the artificial fingerprint. Six fingerprint sensors were tested (optical and solid-state sensors), all of which accepted a silicone fingerprint as a real finger, almost all at the first attempt. More optical sensors have also been tested at various fairs (mainly at the CeBIT trade fair in Hannover, Germany) and all sensors tested, accepted the silicone fingerprint at the first attempt. [57]

71 6.4 Impact of Artificial Gummy Fingers on Fingerprint Systems Impact of Artificial Gummy Fingers on Fingerprint Systems By T. Matsumoto, H. Matsumoto, K. Yamada, and S.Hoshino, January Even though Putte and Keuning were the first well-known to perform experiments simulating the legitimate user not cooperating, the real breakthrough for artificial fingerprints came with the results from a research group led by Tsutomu Matsumoto at the Yokohama National University in Japan. Matsumoto and colleagues tested fingerprint systems with silicone artificial fingerprints. From the results they concluded that systems with capacitive sensors and some systems with optical sensors could reject silicone fingerprints. In order to investigate the security of the systems further, they carried out experiments with gelatin fingerprints to determine whether or not fingerprint systems could detect a gelatin artificial fingerprint or not. Gelatin is made by dissolving collagen (a protein found in bone and connective tissues) in a hot solution. Since gelatin is made out of collagen, it resembles the surface of human skin in ways of moisture, electric resistance, and texture (see table 6.1). [3, 42, 45] Type Moisture Electric resistance (sic!) Live finger 16 % 16 MOhm/cm Gelatin artificial fingerprint 23 % 20 MOhm/cm Table 6.1. Characteristics of a live finger compared to a gelatin artificial fingerprint. [42] As with Putte s and Keuning s work, Matsumoto and colleagues also made two types of experiments [42]: Cloning with a plastic mold: With help of the legitimate user s cooperation, a molding plastic was used to make a mold. Solid gelatin leaves were solved in hot water and this solution (50 % gelatin and 50 % water) was then poured into the mold to create an artificial fingerprint. Cloning from a residual fingerprint: A latent fingerprint on a glass plate was enhanced with cyanoacrylate adhesive, as described in section on page 20. A digital microscope was used to make the fingerprint digital. After using an image processing software to improve the contrasts etc., the picture was printed and used as a mask to create a photo sensitive coated PCB with a copper fingerprint on it. Having the mold ready, a gelatin solution of 40 % gelatin and 60 % water, was used to create the artificial fingerprint. The different types of experiments performed are shown in table 6.2 on page 52. The experiments were performed with five subjects for the method of cloning with a plastic mold and one subject for the method of cloning from a residual fingerprint. They all attempted one to one verification 100 times in each type of experiment

72 52 History of artificial fingerprints for each fingerprint system, thus acquiring acceptance rates in verification for the fingerprint systems. Eleven fingerprint systems were tested, all of which used either optical or capacitive scanning techniques. When different security levels were available, the highest was used. [42] Experiment type Enrollment Verification/Identification 1 Live fingerprint Live fingerprint 2 Live fingerprint Artificial fingerprint 3 Artificial fingerprint Live fingerprint 4 Artificial fingerprint Artificial fingerprint Table 6.2. Experiment types. [42] All fingerprint scanners tested, falsely accepted the artificial fingerprints. For both types of experiments (cloning with a plastic mold and from a residual fingerprint), the artificial fingerprints were all enrollable and in experiment type 2, the fingerprint systems accepted the artificial fingerprints more than 67 % of the time. [42] 6.5 Body Check Biometric Access Protection Devices and their Programs Put to the Test By L. Thalheim, J. Krissler, and P-M. Ziegler, c t magazine, May Lisa Thalheim, Jan Krissler, and Peter-Michael Ziegler for c t magazine, tested eleven fingerprint scanners at the CeBIT trade fair in Hannover, Germany, in Six capacitive scanners, two optical scanners, and one thermal scanner, were tested. Several of the capacitive scanners could be fooled by breathing on the latent print, using a water bag on a latent print, or by dusting with powder and using an adhesive film as described in section on page 32. One optical scanner was also fooled using the adhesive film method. [55] Attacks with silicone artificial fingerprints were also performed where the molds were made by heating the wax on tea-warming candles. Both the optical and thermal scanners were fooled using this method. [55] 6.6 An Investigation Into the Vulnerability of the Siemens ID Mouse Professional Version 4 By A. Ligon, September The security of the Siemens ID Mouse Professional Version 4 was examined with help of the following experiments [39]:

73 6.7 Spoofing and Anti-Spoofing Measures Latent print reactivation through breathing. 2. Latent print reactivation with a water-filled plastic bag. 3. Latent print reactivation with latent print powder. 4. Print lifting with latent print powder. 5. Gummy finger from a live finger mold. 6. Gummy finger from a photolithographic PCB mold. The first four types of attacks are described in section on page 32. In the fifth experiment, Sculpy clay was used to make the mold, and a gelatin solution of 50 % was used to make the artificial fingerprint. Due to lack of equipment, the picture taken by the scanner itself was used to produce the PCB in the sixth type of experiment. [39] The results showed that none of the first four gave an acceptance rate of more than 10 % (out of forty trials). The fifth experiment however, was very successful and managed to fool the scanner up to 90 % of the trials. The sixth method was unsuccessful in fooling the scanner. The probable reason is the bad picture quality produced by the scanner itself. [39] 6.7 Spoofing and Anti-Spoofing Measures By S. A. C. Schuckers, December The Biomedical Signal Analysis Laboratory at West Virginia University, USA, has developed spoofing techniques (i.e. attacks with artificial fingerprints) in order to test a new liveness detection algorithm, see section on page 42. The molds are made from dental impression material (combination of type 0 and 3) and casts are made from Play-Doh and clay, since they are moisture based and most fingerprint scanners were able to image them. Eleven different subjects were used, and six casts were made for each of these subjects. Various fingerprint scanners were tested, including capacitive DC, capacitive AC, optical, and opto-electronic technologies. For certain fingerprint scanners, most subjects casts were able to spoof the system. For all technologies, at least 3 of 11 subjects casts were of sufficient quality to spoof fingerprint devices at least once. [51] Schuckers and colleagues, also tested the systems with cadaver fingers in an attempt to address the possibility that dismembered fingers could be used to spoof fingerprint devices. For one device, 6 out of the 14 available cadaver fingers, were not able to enroll. For the other scanners and fingers, cadaver fingers were falsely accepted % of the verification trials. [51]

74 54 History of artificial fingerprints 6.8 Fooling Fingerprint Scanners Biometric Vulnerabilities of the Precise Biometrics 100 SC Scanner By A. Stén, A. Kaseva, and T. Virtanen, March The next interesting work in the field of artificial fingerprints, was performed by Antti Stén, Antti Kaseva, and Teemupekka Virtanen at the Helsinki University of Technology. To examine the security of fingerprint scanners, they chose an example device and examined it using the following methods [53]: 1. Using grease stains left on the pad. 2. Creating a mold using a live finger. 3. Creating a mold using a latent fingerprint. The first of these attacks was performed in two ways; breathing on the latent print on the sensor, and pressing a gummy bear on the latent. In neither case, the attack was successful. [53] In the second and third attack, gelatin was used to make the artificial fingerprint. A similar method as Matsumoto and colleagues used (see section 6.4 on page 51), was also used in these experiments. Both attacks managed to fool the scanner in a few out of a hundred trials (the third attack had a FAR with artificial fingerprints of 2 %) [53]. 6.9 Evaluation of biometric security systems against artificial fingers By Johan Blommé, October [3] This report evaluates the security of fingerprint scanners with a focus on artificial fingerprints. The experiments were based on cooperation from the legitimate user. The molds were made from a silicone clay, and the artificial fingerprints were made of a gelatin solution. Artificial fingerprints were based on ten subjects whose real fingerprints and artificial counterparts were tested on three different fingerprint scanners, one optical (FTIR with a sheet prism) and two electric-field. All scanners tested accepted artificial fingerprints as substitutes for real fingerprints. Results varied between users and scanners but the artificial fingerprints were accepted about % of the trials. [3]

75 Chapter 7 Experiment description This chapter explains how the experiments were performed. First, a description of how to make the artificial fingerprints used in the experiments, is presented. Then, the two types of experiments performed are described. The first type of experiment tested nine fingerprint systems at the CeBIT trade fair in Germany, The second type of experiments was more extensive than the first, but with fewer fingerprint systems. This chapter will together with appendix B on page 109, and appendix C on page 113, make it possible to fully recreate the performed experimental procedure. 7.1 Making of the artificial fingerprint The making of an artificial fingerprint without a subject s cooperation is quite complicated and requires a number of steps. The process is graphically depicted in figure 7.1 on page 56, and all the steps will be described in detail in this section Enhancing the fingerprint The material used in these experiments are partly due to a limited budget. With more money, more advanced methods and materials could be used for the enhancement of the fingerprints, see [1]. A limit budget would also show that, if possible to circumvent the tested systems, a very advanced equipment or a lot of money is not needed. The subjects right index fingers were checked for scars and dirt before the testing began. If any scars or dirt would have been found, the right middle finger would have been chosen instead. This was however not the case for any of the subjects. 55

76 56 Experiment description Figure 7.1. An overview of the process of making the mold. The subjects were informed to press their right index finger on a micro slide (a small glass plate) three to four times. Before pressing the finger on the micro slide, they were however informed to rub the finger to the side of the nose. This was done to get some substance, like fat and grease on the finger, so that a latent print of good quality would be achieved.

77 7.1 Making of the artificial fingerprint 57 Powdering A soot powder mixture together with a squirrel hair brush (see figure 7.2), both from KTM, were used during the experiments. This equipment is used by forensic laboratory assistants. For more detailed information about the powder and brush, see appendix B on page 109. With help of the squirrel hair brush, the soot powder mixture was carefully brushed onto the fingerprints on the micro slide. Brushing a fingerprint might seem easy at first, but it is more complicated than simply dipping a brush into the jar of powder and then painting it onto a surface [59]. First, the brush was dipped carefully in the jar of powder to load the brush fibers with some powder. Then, before starting to paint, the brush was tapped against the dish to get rid of the excess powder. While letting the brush gently touch the surface, the brush was moved in the direction of the papillary lines where possible. For some fingerprints, this motion becomes similar to twirling the brush. Care was taken at this step, not to overdevelop nor erase the latent prints. The trick lies in letting the brush extremely gently touch the surface, and not having superfluous powder on the brush. Figure 7.2. A soot powder mixture and a squirrel hair brush were used in the experiments to enhance the latent fingerprints. Lifting the fingerprint Adhesive tape from KTM was used to lift the dusted prints from the glass surface to a white piece of paper. This might also seem very easy at first, but proved to be a bit tricky as well, see section on page 76. When applying the tape to the micro slide, a ruler with a piece of cloth on it (to avoid making scratches on the tape), was used. After having applied the tape to

78 58 Experiment description the end of the micro slide, the ruler with the cloth was slid at a constant speed (to avoid folds) over the latent enhanced fingerprints. When removing the tape from the micro slide, again a constant speed was used to avoid folds on the tape. The tape was then attached to a white paper in the same manner as when the tape was attached to the micro slide. Care should also be taken, not to let any grease or dust get attached to any of the sides of the tape. Therefore, make sure your hands are clean Photographing the fingerprint The next step was to photograph the lifted fingerprint. A Minolta DiMAGE 5 digital camera was used with the detailed settings described in appendix C on page 113. Since taking close-up photographs can be very difficult, a tripod was used so that a lower ISO value could be used. The photos were taken indoors with some light from outside together with fluorescent lighting. The camera was held about 25 cm from the print since that was the closest you could get in macro mode with the camera used Image processing Adobe r Photoshop r CS from Adobe Systems Inc., was used for the image processing performed. To start off with, a picture in jpeg format with the size of pixels, was used. The main steps used in the image processing were the following: The image was sharpened with the filter unsharp mask. The image was reversed by flipping the canvas horizontally. The papillary lines were sharpened and pores were removed with help of the brush tool. The colors of the fingerprint were inverted. The size of the image was adjusted to the real size of the image. With help of the threshold option, the image was turned black and white instead of being grayscaled. Any remaining traces from pores or the soot powder mixture were erased. Both the reversing and the inverting of the image are needed to make the outcome of the etching reversed and inverted, in order to make the gelatin artificial fingerprint identical to the real fingerprint. Subject S2 has very wide ridges compared to the valleys. Therefore, the valleys (showing up as black lines) on S2 s fingerprint image, had to be widened a bit to make the etching possible. The detailed description

79 7.1 Making of the artificial fingerprint 59 of the image processing, can be found in appendix C.2 on page 114. Images of part of the subjects fingerprints before and after image processing, can be found in appendix C.3 on page Printing the image The printer used in the experiments was a HP LaserJet 5Si/5Si MX PS with a resolution of 600 dpi. The picture of the fingerprint was printed on a transparency for the etching to work the best PCB production An epoxy laminate, which created a copper thickness of 35 µm, was used for subject S1, and epoxy laminates, which created a copper thickness of 70 µm, were used for subjects S2 and S3. See section on page 80, for an explanation of the chosen thicknesses. The epoxy laminate is in fact a copper clad board coated with a layer of lacquer, i.e. photoresist. The photoresist is sensitive to UV light and even though it is not very sensitive to ordinary light, it should not be exposed to ordinary light for a very long time. [15] The procedure for the production of the PCB can be summarized as follows: The developer and etching solution were mixed. The laminate was exposed for 3 minutes in the UV light box. A first developing was performed until the pattern appeared. Then, a second developing was done to get rid of any remaining photo resist. The board was washed with water. The laminate was put in the etching solution, and to keep a temperature of about +50 C (for the best result), the etching bowl was placed in a hot water bath in the sink. After having stirred for about minutes, the board was washed with water again. The detailed description of the production of the PCB, can be found in appendix C.4 on page Gelatin solution This section discusses the choice of gelatin as material to make the artificial fingerprints. It also describes how the gelatin solution was prepared and how the artificial fingerprint was made.

80 60 Experiment description Material choice When making an artificial fingerprint for fooling a fingerprint scanner, the artificial fingerprint need not resemble a real fingerprint to every detail. The main focus lies in getting the same appearance as a live finger would on the scanner s surface. [3] In previous studies, a range of different materials have been used, see section 6 on page 49. These studies show that different materials work better or worse for different types of sensors. Knowing what kind of sensor you want to fool is thus an advantage. In this study, many different sensors were tested, and the artificial fingerprints could therefore not be made to fool only one specific sensor. As described in section 6.4 on page 51, gelatin is one of the best materials to create an artificial fingerprint out of. Making of the gelatin solution Since the fingerprint scanners can be sensitive to humidity, the same percentage of water as found in the human skin, had to be used in the artificial fingerprints. Therefore, a solution of about 44 % gelatin and 56 % water was used to make the artificial fingerprints (e.g. 17 g of gelatin together with 21.5 ml of water). In previous studies performed, the amount of gelatin has ranged from 40 % to 50 % and since no clear distinction between 40 % and 50 % of gelatin could be found in the early testing stage, a value in between which was easy to measure, was chosen. The solid gelatin leaves were soaked in the water in a square plastic container for about five minutes, and then both the gelatin leaves and the remaining water was put in a glass jar with the cap on to avoid reducing any water. The glass jar with gelatin and water was then put in a water bath in a sauce pan on a stove. The sauce pan and contents were heated and the jar was left in the hot water until the gelatin had dissolved. The jar was rotated and moved gently during the dissolving of the gelatin. A violent moving of the jar caused a lot of air bubbles to form. Air bubbles are not wanted since they would appear on the artificial fingerprint and could make the fingerprint scanners realize an artificial fingerprint is being used. Sometimes the gelatin formed lumps that were difficult to dissolve, and in those cases a kitchen knife was used to help stirring. When the gelatin had dissolved in the water, often there were still air bubbles in the solution. Heating the solution up (still in the jar in a water-bath) and then letting it cool down a couple of times made most bubbles disappear. Making of the gelatin artificial fingerprint The liquid gelatin solution was poured on top of the mold (see figure 7.3 on page 61) and thinned with help of a knife. The thinner the gelatin fingerprint becomes, the more invisible it will be for any person watching you using it. In these experiments

81 7.1 Making of the artificial fingerprint 61 however, the gelatin was used for quite some time, and it was then easier to handle a somewhat thicker gelatin fingerprint. The thickness of the gelatin prints used in these experiments, was about 1-2 mm. Less, and it would have been more difficult to peel it off afterwards without destroying the print. A thinner gelatin print would also have been more difficult to handle without destroying the print, and last but not least, it would have dried out too fast (a gelatin fingerprint had to last about 45 minutes in room temperature and humidity). Figure 7.3. A mold with a gelatin solution on top of it. The mold with the gelatin solution was placed in a refrigerator for about ten minutes until the gelatin had stiffened. The gelatin can also be left in room temperature to stiffen, but it will become drier then. With help of a knife, the gelatin was peeled off from the PCB. Care had to be taken when peeling the gelatin off from the PCB so it would not be damaged. After having peeled off the gelatin, it was cut with a pair of scissors to a small fingerprint. Some surroundings of the fingerprint was however left, because of the fingerprint drying out too fast otherwise during the time consuming experiments. A gelatin artificial fingerprint on top of a live finger is shown in figure 7.4 on page 62. The remainder of the gelatin solution can be stored in the glass jar in a refrigerator for a couple of weeks, and heated in a water-bath the next time needed. Storing a gelatin solution in the refrigerator for more than a couple of days, will however make it go mouldy.

82 62 Experiment description Figure 7.4. A fingertip with a wafer-thin gelatin fingerprint on top of it. Storing of the artificial fingerprint The artificial fingerprints were stored in a refrigerator in small film jars with a piece of moist cloth covering about 75 % of the jar. At the top, the gelatin fingerprint was placed with the flat side down and the print side turned upwards to avoid making the print too moist and risk destroying the pattern. With the lid on, the fingerprint can be stored a couple of days in this environment. After about a week, the gelatin will have gone mouldy, and will soon after that not be able to fool the fingerprint scanners any more. 7.2 Experiments at CeBIT Some fingerprint systems were tested at the CeBIT trade fair in Hannover, Germany, year The products tested were not chosen due to any special reasons, but simply because they were available and were allowed to be tested. The gelatin fingerprints brought to CeBIT were made on the 19th of March 2004 and tested the two following days. During this period, the artificial fingerprints were stored in a cool bag in small film jars with a piece of moist cloth, as described in section on page 62. Nine different sensors were tested. The software and security levels (thresholds) used during the experiments are unknown. For each system tested, first subject S2 s finger was enrolled and identified/verified. Then subject S2 s gelatin artificial fingerprint was tested for identification/verification. If the system could not be fooled with S2 s artificial fingerprint in a few tries, subject S1 s fingerprint was tested in the same manner.

83 7.3 Extensive experiments Extensive experiments The extensive experiments were performed with three different subjects and three different fingerprint scanners. For each subject, 50 identifications/verifications were performed for each fingerprint scanner, both using their real fingerprints, and using two different copies of their artificial counterparts. The reason for testing two different gelatin prints for each subject, is that the results depend a lot on the quality of the gelatin. The tests with the two different gelatin prints, will be referred to as round one and round two in the remaining of the report Subjects and input Since there were only three subjects (due to limited experimental time), it was not possible to get a standardized representation of the population. Still, subjects were chosen depending on age and gender. The subjects varied between 18 and 65 years of age and both genders were represented. In table 7.1, four types of experiments are shown. Matsumoto and colleagues, performed all four types of experiments, see section 6.4 on page 51. In the experiments prior to this report, only experiment type number 1 and 2 were performed since these are the two types of experiments that resemble biometric security systems normal use. Experiment types number 3 and 4 are normally used to check if enrollment of an artificial fingerprint might complicate or simplify recognition [3]. Experiment type Enrollment Verification/Identification 1 Live fingerprint Live fingerprint 2 Live fingerprint Artificial fingerprint 3 Artificial fingerprint Live fingerprint 4 Artificial fingerprint Artificial fingerprint Table 7.1. Possible experiment types. Only experiment type number 1 and 2 were performed. [42] If the FRR was very high for a special subject and fingerprint scanner, the enrollment for that subject and fingerprint scanner was redone. The reason being that a poor enrollment may not for example capture the center of the fingerprint, thus forcing the subsequent verifications/identifications to the same type of misplacement of the finger. For each subject, one mold and two different artificial fingerprints were created and used. If the quality of either the mold and/or the artificial fingerprints were not satisfactory, they were remade. The identification/verification of both the real fingerprints and the artificial fingerprints were performed 50 times for each fingerprint scanner. The verification/identification attempts were performed while counting the number of times

84 64 Experiment description the system accepted the fingerprint, rejected the fingerprint, or falsely logged in another person. Experiment type number 1 is important for showing the FRR. The success rate in this experiment should be as high as possible, ideally 100 %. This experiment type resembles the normal use of the fingerprint scanner. In experiment type number 2, the artificial fingerprint should ideally be rejected every time, thus achieving a FAR of 0 % Software and hardware The same software and hardware as in the experiments described in [3], were used in the extensive experiments. Since the fingerprint scanners tested, are marketed at different levels of security and areas of use, a single unifying program could not be used to test the scanners at equal terms. Instead, the provided software for each scanner was used. Both the Targus DEFCON TM Authenticator TM and the Precise TM Biometrics 100 MC fingerprint scanners use an electric field sensor, while the Identix fingerprint scanner is equipped with an optical sensor using FTIR with a sheet prism. More detailed information about the sensors can be found in appendix D on page 119. Two of the softwares used, had more than one security level to choose from. Since no specific use of the systems was simulated, the default security settings were used. BioLogon TM Security System & Identix BioLogon TM Security System (version 2.03) for Microsoft Windows, is the software that was used together with the Identix fingerprint scanner. BioLogon replaces the standard Windows log on process and allows users to use their fingerprints or a master password instead of remembering a number of passwords for each program and site. [30] To identify a new user, you have to log off the current session and log on as a different user. To speed up the log on/log off process, all unnecessary executables were stripped from the log on process, and a script was created to automatically log off the user directly after having logged on. This also required a registry key to be added. BioLogon has three different security settings, and the default setting 2 was used in the experiments. BioLogon was run on Microsoft Windows XP during the experiments.

85 7.3 Extensive experiments 65 Softex Omnipass & Targus DEFCON TM Authenticator TM Softex Omnipass (version ) is the software that was used together with the Targus DEFCON TM Authenticator TM fingerprint scanner. Softex Omnipass main feature is that you can use a fingerprint (or a master password) to access password protected programs or sites, or log on to a computer using an existent Microsoft Windows account. Softex Omnipass works with Microsoft Windows based operating systems and during the experiments it was run on Microsoft Windows XP. [52] It was possible to use standard Windows together with Softex Omnipass to log on a user, but during the experiments, identification in Softex Omnipass was made by first clicking on an icon in the system tray and then presenting the fingerprint. To speed up the process of having to click on an icon and then on a menu to log in, a macro to perform these operations was used. This allowed a single key on the keyboard to be used to perform these specific commands. Precise TM Logon & Precise TM Biometrics 100 MC Precise TM Logon (version 2.1) is the software that was used together with the Precise TM Biometrics 100 MC fingerprint scanner. Precise Logon is a demonstration software that had to be run on Microsoft Windows 2000, but it has full features when it comes to logging on [3]. The software does only work in verification mode, thus making it impossible to compare the FAR with the other scanners used in the extensive experiments. Precise Logon gives feedback if an incorrect placement or moisture level is used during enrollment or verification. Precise Logon is integrated in the Windows log on process. Like the time issues between trials discussed in section on page 64 concerning the testing of BioLogon, the same problems arose when using Precise Logon. Since the software could only be used in verification mode, only one session had to be stripped of executables started at log on. Using an automatic log off script was not possible using Microsoft Windows 2000 [3]. Precise Logon has 7 different security levels, and the default setting 4 was used during the experiments Experiment procedure Prior to the real experiments, the subjects tried the different systems under supervision to learn how to place and press their fingers on the sensors to get acceptable images of the fingerprint. This way, unfamiliarity with the systems as an erroneous source, could be excluded.

86 66 Experiment description After having learnt how the scanners worked to get accepting results, the subjects were however allowed to use their own techniques when presenting the finger to the scanners. This way, a normal working environment was simulated. During enrollment or identification/verification, the subject (or the tester), presented the real fingerprint (or the artificial fingerprint) so that it suited the center of the scanning area of the fingerprint scanner. Enrollment The subjects right index finger, were enrolled into the systems. The enrollment was performed with the provided software for each fingerprint scanner, with their provided specifications, see section on page 64. The enrollment was also supervised carefully and each user s enrollment was verified by logging onto the system once, making sure the finger was enrolled properly. Identification/Verification In experiment type number 1 in table 7.1 on page 63, the subjects presented their own fingerprint under supervision as they were taught during the initial phase. The supervisor surveyed and counted the number of successful logins, the number of false rejections, and the number of erroneous identifications. In experiment type number 2 in table 7.1 on page 63, the tester presented the artificial fingerprints to the fingerprint scanners. This way, the time each subject had to spend on the experiments, was minimized. It also ensured that the identifications/verifications of the artificial fingerprints, were performed in the same way for all fingerprint scanners and subjects. In the first round, no special testing order between the scanners was used, but since the order might have affected the results, it is presented in table 7.2. In the second round, the Precise scanner was always tested first because of the results in the first round which showed that the right humidity of the gelatin print was very important to be able to fool the Precise scanner. Subject Round one Round two S1 Precise, Identix, Targus Precise, Identix, Targus S2 Targus, Identix, Precise Precise, Targus, Identix S3 Targus, Identix, Precise Precise, Identix, Targus Table 7.2. Testing order for the scanners in round one and two. The scanners are written in the order they were tested, with the scanner which was tested first written first. In the first round, no special order was chosen, but the order is simply a coincidence. In the second round, the Precise scanner was always tested first because of the results in the first round.

87 Chapter 8 Results The results from the experiments performed will be presented in this chapter. All the data in the diagrams, can be found in numbers in appendix F on page CeBIT The results from the experiments at CeBIT are shown in table 8.1 on page 68. Nine different sensors were tested. For each system tested, first subject S2 s fingerprint was enrolled and identified/verified. Then S2 s gelatin artificial fingerprint was tested for identification/verification. If the system could not be deceived with S2 s artificial fingerprint in a few tries, subject S1 s fingerprint was tested in the same manner. In other words, S1 s fingerprint was only used when S2 s artificial fingerprint was not able to circumvent the system. All the systems in table 8.1 on page 68, were deceived with either S2 s artificial fingerprint or S1 s artificial fingerprint. One electric field sweeping sensor, one capacitive sweeping sensor, and one thermal sweeping sensor, were tested and all three were circumvented. However, the tested sweeping sensors all required S1 s artificial fingerprint to be used, since S2 s artificial fingerprint was not able to deceive the sweeping sensors in a few tries. In all other cases, S2 s artificial fingerprint was enough to deceive the systems. 67

88 68 Results Type of sensor Defeated with subject Optical FTIR S2 Optical FTIR with a sheet prism S2 Solid-state Capacitive S2 Solid-state Capacitive (sweeping) S1 (not S2) Solid-state Thermal (sweeping) S1 (not S2) Solid-state Electric field S2 Solid-state Electric field S2 Solid-state Electric field S2 Solid-state Electric field (sweeping) S1 (not S2) Table 8.1. Results from attacks with artificial fingerprints at CeBIT. Nine fingerprint systems were tested and all systems tested were deceived either with subject S2 s artificial fingerprint or subject S1 s artificial fingerprint. 8.2 Extensive experiments The results from the the extensive experiments, experiment types 1 and 2 in table 7.1 on page 63, will be presented in this section, both in numbers and in percentage. The number of successful logins with the subjects real fingerprints, serves as a control when comparing with the number of false acceptances with the subjects artificial fingerprints. For each subject, 50 identifications/verifications were performed for each fingerprint scanner, both using their real fingerprints, and using two different copies of their artificial counterparts. The tests with the two different gelatin prints, are referred to as round one and round two Results in numbers In figure 8.1 on page 69, the number of successful logins with real fingerprints, is shown. The number of trials were 50 for each subject and fingerprint scanner. Since no false acceptances occurred during this type of experiment, the remaining number of trials were falsely rejected. The values for the verification/identification trials with real fingerprints, can be compared with the values from the trials with artificial fingerprints. The number of false acceptances using both rounds of artificial fingerprints, is shown in figure 8.2 on page 70. In the experiments with subject S3 s artificial fingerprint during round one, there was one occurrence of a false acceptance as another user. S3 s artificial fingerprint was falsely accepted as subject S1 one out of 50 times during the trials with the Targus DEFCON TM Authenticator TM fingerprint scanner, during the first round. Apart from this, there were no other false acceptances as other users, and thus the remaining number of trials were correctly rejected.

89 8.2 Extensive experiments 69 Notice that the number of false acceptances using artificial fingerprints, was increased for both subject S2 and S3 on all scanners from round one to round two. For S1, the number of false acceptances using artificial fingerprints, was decreased somewhat from round one to round two. Figure 8.1. The number of successful logins with real fingerprints for each of the subjects and for each fingerprint scanner tested. The number of trials for each subject and scanner was 50, and since no false acceptances occurred during this type of experiment, the remaining number of trials were falsely rejected.

90 70 Results (a) Round one. (b) Round two. Figure 8.2. The number of false acceptances with artificial fingerprints for each of the subjects and for each fingerprint scanner tested during round one and two. The number of trials for each subject and scanner was 50. One false acceptance as another user occurred during round one (subject S3 s artificial fingerprint was once logged in as subject S1 on the Targus fingerprint scanner), and the remaining number of trials were correctly rejected. No false acceptances as other users occurred during round two.

91 8.2 Extensive experiments Results in percent The number of successful logins shown in figure 8.1 on page 69, give the success rates shown in figure 8.3. Ideally, the success rate should be 100 %, the FRR 0 %, and the FAR 0 %. In a real situation, this is however impossible. Still, the FRR and FAR should be kept as low as possible. A high FRR is not a security threat, but can be very annoying for the user. A FRR around 10 % or lower is considered acceptable. A high FAR is a serious security threat. Figure 8.3. The success rate with real fingerprints for each fingerprint scanner tested. The mean values of all subjects are shown for each scanner. The FARs for all scanners were 0 %. The number of trials for each subject and scanner was 50. The FARs for artificial fingerprints during both rounds, are shown in figure 8.4 on page 72. Ideally, both the FAR with artificial fingerprints and the FAR that occurs when being logged in as another user, should be 0 %. The rejection rate should ideally be 100 % when trying to log in with artificial fingerprints. In both rounds, there were cases where the FARs with artificial fingerprints were as high as 100 %.

92 72 Results (a) Round one. (b) Round two. Figure 8.4. The FAR with artificial fingerprints, for each fingerprint scanner tested during round one and two. The number of trials for each subject and scanner was 50 during both rounds. During round one, the FAR as another user was 0 % for the Identix scanner and the Precise scanner, and about 0.7 % for the Targus scanner.

93 8.2 Extensive experiments 73 Round one In the first round of experiments performed with artificial fingerprints, the FAR as another user was 0 % for the Identix fingerprint scanner and the Precise fingerprint scanner. There was one occurrence of a false acceptance as another user during the first round, making the FAR as another user about 0.7 % for the Targus fingerprint scanner. The FAR with artificial fingerprints during round one, was greater than 0 % for all subjects on all fingerprint scanners except for subject S2 and S3 on the Precise fingerprint scanner. The other FARs with artificial fingerprints varied between 66 % and 100 %, depending on the subject and scanner. Round two In the second round of experiments with artificial fingerprints, no false acceptances as other users occurred, making the FAR as another user 0 % for all scanners and subjects. The FARs with artificial fingerprints were greater than 0 % for all subjects and scanners in the second round. Except for the FAR of 12 % for subject S2 on the Precise scanner, all FARs were greater than 82 % in the second round. Mean values The mean values in percent for all fingerprint scanners and subjects, for real and artificial fingerprints, for both rounds, are shown in figure 8.5 on page 74. The mean value of the success rate using real fingerprints was 90 %, and the mean values of the FAR with artificial fingerprints were 67 % in the first round and 86 % in the second round.

94 74 Results Figure 8.5. Mean values, in percent, for real and artificial fingerprints (both rounds).

95 Chapter 9 Discussion and analysis This chapter discusses and analyzes the method used and the results acquired in the experiments. Results acquired in previous studies in the same field, will be compared and put up against the results acquired in these experiments. 9.1 Experiment method Since the experiment method used, includes so many steps and a lot of material, there are plenty of ways to optimize the method. Some steps and material seem however more important or difficult to improve than others. For example, the dusting with soot powder mixture, can be improved. Small black particles got attached not only to the ridges where they should, but also at some points in the valleys. Even the world s best camera equipment can not improve a poor enhancement Enhancing the fingerprint The material used to enhance latent fingerprints can be easily found and bought on the Internet. By searching on the Internet for fingerprint kit, soot powder etc., you find many links to companies selling fingerprint enhancement products. Even though some companies might reject orders from private people, there are companies that do accept them. Powder A few other powders were tried before finding the soot powder mixture used in the experiments. Only one of the other powders, a graphite powder, did provide 75

96 76 Discussion and analysis a somewhat satisfiable enhancement. Graphite powder makes the fingerprint look more gray than using the soot powder mixture. The grayish look from the graphite is however a smaller problem than the blurriness and indistinctness, which the graphite powder produces. Using soot powder mixture makes the fingerprint much more indistinct. The reason is probably that the soot powder mixture was more finely ground than the graphite powder used. In [53], photocopier powder was used and the results (FAR with artificial fingerprints about 2 %) show that it is possible to create an artificial fingerprint from this. However, the authors in [53] suggest using another powder for improved results. This still shows that easily obtainable powders can be used to make a forged fingerprint. Brush As with the powders, a few different brushes were tested before finding the squirrel hair brush used in the experiments. The first brush that was tested cost 3.50 SEK and was made from an unknown material. The second brush that was tested cost 17 SEK and was made of imitated marten hair. The second of these did produce somewhat better enhancements than the first brush, but the difference was small. None of the above mentioned brushes did however produce the same good enhancement as the squirrel hair brush from KTM. A feather duster from KTM was also tried, but since it tended to carry too little powder, it was discarded. Note though that there are times when little powder is better [59]. The following two quotes from [1], show the importance of having a good brush and taking care of it: In powdering fingerprints, a fine quality brush is indispensable. Stiff hairs can damage papillary ridges and a very soft brush should therefore be used. Brushes need to be replaced from time to time, since in use, they inevitably absorb greasy substances and begin to pick up too much powder. In ordinary hobby shops, you can also find softer and better brushes than the cheap ones mentioned above. So, even if it was not possible to order forensic fingerprint enhancement products, it would be possible to get hold of a good brush. They are however more expensive than the forensic brushes. Lifting with tape The biggest problem when lifting the prints, were the bubbles which formed both when applying the tape to the latent print, and when applying the lifted print onto the paper. Two solutions were found to this problem. The first solution includes

97 9.1 Experiment method 77 using a good tape. The tape from KTM used in these experiments proved to be better than ordinary office tape. The second solution includes using a ruler to get even pressure, both when applying the tape to the micro slide, and when applying the lifted print to the paper. Unfortunately, this introduced another problem, namely scratches made by the ruler on the tape. To solve this problem, a soft piece of cloth was held between the tape and the ruler. Another problem with the tape are the folds that formed both when applying the tape to the micro slide and when applying the lifted print onto the paper. This problem was solved by sliding the ruler over the tape at a constant speed and never stopping before being finished (or stopping in between different fingerprints). The tape used can be ordered with a tape holder which might help against the bubbles. This was however not tried. Instead of using tape to lift the fingerprint, you can use Mikrosil TM (can be found at forensic Internet shops) which seems easier to use, but also more expensive. This was not tried either. Further ideas In [59], a way to enhance a fingerprint is described, which includes using a quality fiberglass brush and a gentle grinding of the powder using a rotating motion by twirling the brush handle slowly between the thumb and index finger. There are a number of ways to enhance a print with a brush and a powder, and this work often requires a lot of experience to make a good enhancement. One of the problems with the enhancement of the fingerprint, is the lifting part. To eliminate this problem, you could put a white paper or liquid behind the glass and take photos of it directly without lifting. This would only be possible for some surfaces however. For example, a colored glass plate introduces problems, and the glass having an irregular shape could make it more difficult to get a good picture when taking the photo. Instead of dusting the print with a powder and then lifting, the enhancement could be done chemically. In most forensic laboratories (at least in Sweden) today, they use mainly chemical methods to enhance latent prints. Cyanoacrylate fuming, described in section on page 20, is one of the methods that could be performed at home with easily available material. Cyanoacrylate fuming was therefore considered for these experiments, especially since it was used in the experiments by Matsumoto and colleagues. After a few initial tries, it was however discarded since it is a dangerous method and it seemed to require more time than available.

98 78 Discussion and analysis Photographing the fingerprint Instead of enhancing the latent fingerprint, which introduces many problems, the latent print could have been photographed directly. The following method was suggested by Ulf Söderholm, a Swedish criminal photographer [13]: In a room with little light, put a black plate about 50 cm behind the micro slide and photograph with a light beam against the print. A flash light might be used if no flash or halogen lamp is available. Despite the problem of making the latent print visible, this can be a bit tricky if the surface is curved or colored for example. When taking photos of the fingerprint, it is a good idea to have a millimeter scale by the print. Tape rulers with a millimeter scale can be bought at forensic internet shops. When measuring the size of the fingerprint, a good idea is to measure between two characteristic points [53]. Instead of taking photos of the print, a scanner can be used to scan the fingerprint. With a very good scanner, this could become almost as good as when using a camera. One advantage of this is the 1:1 scale you immediately get with the scanner. However, with the right camera equipment, this is also possible. If the forgery is worth a lot of money, the forger might be willing to spend a lot of money on the equipment used to make the forgery. A digital camera for latent fingerprints with fingerprint enhancement software and toolboxes, is available at Mason Vactron [40] Image processing This section discusses the choices made during the image processing phase and how these choices could have affected the results. The procedure used A difference of about 0.5 mm in size of the artificial fingerprint s height did not seem to matter to the sensors. This means that the measurement of the print s size with an ordinary office ruler should make no difference than measuring with a more exact measuring tool. Some parts of the images were so diffuse that the fingerprint pattern had to be guessed or left diffuse. If it was easy to guess the pattern, or if any other enhanced prints were available with that particular part of the print more distinct, the pattern was guessed. Otherwise, the diffuse part of the print was left untouched. If a correct guessing was made, the FAR for the artificial fingerprints was probably increased. If an incorrect guessing was made, the opposite was probably the case.

99 9.1 Experiment method 79 The image processing of the enhanced fingerprints could be done in a lot of ways, and the procedure used for these experiments is only one of them. Especially, the options levels, threshold and brightness/contrast could be used in different ways to improve the image even more. When laying your finger on a fingerprint sensor with a quite small area size, only a part of the fingerprint is scanned. If the enrollment has been performed in the right way, the most important part of the fingerprint, the middle, was positioned in the center of the sensor. Therefore, the image processing of the fingerprint should be more important closer to the middle of the print. This could have had the affect that a less thorough image processing of the edges of the fingerprint image, was made. Pores To the author s knowledge, none of the tested fingerprint systems check pores to identify/verify a user. If a fingerprint system was constructed to also include the position and size of the pores as an identification/verification tool, it would be much harder to construct an artificial fingerprint that would fool that system. Out of curiosity, an artificial fingerprint was created where the pores were tried to be kept. Neither the enhancement procedure, the photographing of the enhanced fingerprint, nor the image processing, seemed to be a problem. The problem occurred when producing the PCB. Even though the copper thickness of 35 µm was used, the pores would not be rendered as clearly on the PCB, as in the mask. A point should be made here that subject S2 s fingerprint was used for this procedure. Rendering the pores on the mold could be more difficult or easy depending on the specific fingerprint. Automatizing Since each fingerprint image is unique, it is difficult to make an image processing guide to be used on all prints. In fact, it would be more optimal to process each image individually depending on the characteristics of that particular image. In the image processing performed in these experiments, a combination of individually designed image processing and a standardized image processing, was used. Another way to look at the image processing could be from the automatization point of view. For example, why not put the image up for automatic binarization, thinning etc., and then widening the papillary lines again depending on either a standard value or by using a measured width of the papillary lines before thinning together with how close the papillary lines are situated. This way, the time consuming image processing could be used in a mass production.

100 80 Discussion and analysis Printing the image The resolution of the printer and the color of the toner in the printer are the two most critical points at this stage. The greater contrast between the fingerprint and the background, the better. Also, with a higher resolution of the printer, you get a more detailed and better print-out. The resolution of the image could of course also be a bottle-neck, but was not in this case PCB production Epoxy laminates, which created copper thicknesses of 35 µm, 70 µm, and 105 µm were available. The 105 µm thickness was discarded since the very small details of the fingerprint were difficult to etch that thick. A few tries were made with the remaining thicknesses and no clear difference could be found between the two. The quality of the etched fingerprint changed with different etchings. For subject S1, the 35 µm copper thickness was chosen, since the etching was somewhat better looking than the etching with the 70 µm copper thickness. For S2 and S3, no clear difference could be seen between the two, and the 70 µm copper thickness was therefore chosen, because the probability of the gelatin papillary lines to smear together on the sensor because of heat, should be less. The etching of the mold introduced some problems in the production of the artificial fingerprints. First of all, the etching time was critical. Under- or overdevelopment, as well as under- or overetching, made the mold look different from the fingerprint on the transparency. Another problem with the etching occurred when etching many prints at the same time. With a board of about cm, the edges were etched first and thus acquired a shorter etching time than the middle of the board which was etched last. The solution to this problem was cutting the board into smaller pieces and etching fewer prints at the same time. Etching only one print on a suitable board size is optimal from this point of view. Papillary lines that are equally wide and with an equal or comparably big distance between them, will simplify and make the etching less critical. In other words, some fingerprints are easier to etch than others. The etched mold can be stored and used over and over again. A few things can however lower the quality of the mold. First of all, the mold should be kept away from grease, dust, etc. Secondly, when using the mold to make a number of gelatin fingerprints, small parts of the gelatin can get stuck in between the copper edges. Two solutions were found to this problem. Firstly, the gelatin should be lifted from the mold as soon as the gelatin has stiffened, i.e. do not try to lift it too early, and do not let it set for too long time. Secondly, the gelatin solution should not contain too much water (not stating what amount that is).

101 9.1 Experiment method 81 If the mold would become damaged, it is still quite easy to make a new one reusing the same transparency Gelatin solution This section will discuss the problems with making a gelatin solution with the right water amount, and then being able to maintain the right humidity of the gelatin artificial fingerprint. Water amount The exact amount of water in the gelatin solution was difficult to check. Since the gelatin leaves were soaked in water in another container before pouring them into the glass jar, some water could have got stuck to the first container. The reason for using first one container, and then a glass jar, was that the glass jar did not have a shape to fit stiff square gelatin leaves. One solution to this problem is to simply skip the soaking of the gelatin leaves. Another solution could be to use a glass jar with a square shape to be used for both purposes. Square glass jars are however not as easily found as round glass jars. Water also evaporated during the process of heating and cooling the gelatin solution. This was done both when reducing bubbles, and later on when heating the gelatin solution a number of times to be used for different artificial fingerprints. Humidity After having worked with the gelatin solution and gelatin artificial fingerprints for a while, you might start to recognize when they are of good or bad quality. The most critical point here is the humidity of the gelatin print. When working with a thin (thickness of about 0 2 millimeter) gelatin fingerprint, it dries out very fast (both in room temperature and in the refrigerator). Storing the gelatin print in a humid film jar, as described in section on page 62, is a very good idea. When using the artificial fingerprint for one or a few times, it will not have time to dry out. However, when using the print for about 45 minutes as in the experiments, this problem occurred and could probably have lowered the FARs for artificial fingerprints. When you need to take the gelatin out of the film jar, and before using it on a fingerprint scanner, one good way to store it is in your hands because of the sweat pores creating a humid environment. However, if you have very warm hands, this could destroy the pattern.

102 82 Discussion and analysis 9.2 Experiments at CeBIT After having finished the experiments at CeBIT, a few aspects about the actual experiment method can be discussed. It would have been interesting to test both subject S1 s and subject S2 s artificial fingerprints on all systems tested. This would however had taken up even more time from the exhibitors, and some exhibitors could have found this inconvenient. Therefore, only S2 s artificial fingerprint was chosen to test the systems at first, and if not successful, S1 s artificial fingerprint came into use. The results would probably have looked different if S1 s artificial fingerprint had been tested first. The artificial fingerprints tested at CeBIT were only tried a few times on each sensor. This is not enough to draw any statistical conclusion of the security of a special sensor. However, it is enough to show that it is possible to defeat the sensor using an artificial fingerprint. If a protection against attacks with artificial fingerprints is to be included in a system, the most obvious place is at the sensor level. It should not be forgotten however, that it could be possible to produce a software which gives enough protection. It is therefore not totally fair to only state what sensors that were tested and leave out the softwares. Not a single system was found at CeBIT that could not be circumvented, even though some companies claim that their products are safe against this type of attack. Using a sweeping sensor, is not safe either, as will be evident in the following subsection Sweeping sensors The sweeping sensors produced a somewhat more difficult situation than the ordinary touch sensors. More friction is produced when sweeping a gelatin artificial fingerprint, than sweeping a live finger on a sweeping sensor. After having tried a few times, the sweeping technique needed for the gelatin was however learned and a good quality image was achieved. The first time a sweeping sensor is used with a live finger, it also requires some training to learn the sweeping technique. Interesting to note is that the three tested sweeping sensors (one electric field, one capacitive, and one thermal), all required subject S1 s artificial fingerprint to be used. The reason why subject S2 s artificial fingerprint could not deceive the sweeping sensors tested, could be a few different ones: The storing of the prints for about two days. Both S1 s and S2 s artificial fingerprints were however stored and should have been exposed to the same kind of conditions. Still, no identical environment was used for the two prints. Since S2 s fingerprint always was tested first, S2 s artificial fingerprint was subject to more dirt and could have been worn out more than S1 s artificial

103 9.3 Extensive experiments 83 fingerprint. S2 s fingerprint has very wide ridges as compared to the valleys, and will therefore produce a print which is more difficult to recognize regardless of whether the live finger is used or whether an artificial fingerprint is used. This also made it more difficult to perform the image processing of the print if not a very exact enhancing method was used (and dusting cannot be considered as one of the most exact enhancing methods). When S2 s fingerprint was dusted with soot powder mixture, some parts of the fingerprint got blurry. It was thus almost impossible to guess how the lines formed at those points, and especially at one point, it was extra difficult. Those places of the print that were somewhat blurry and especially those that were placed close to the middle of the print, could have affected the outcome of the acceptance/rejection decision by the scanner. It is however somewhat strange that only the sweeping sensors would have problems with this. It is more difficult to get a good picture with a sweeping sensor than using a touch sensor. Combining that difficulty with using an artificial fingerprint with somewhat bad quality as S2 s artificial fingerprint was (at least compared to S1 s artificial fingerprint), this could be the reason of the results. The surroundings of S2 s artificial fingerprint were not black on the transparency used for etching, but S1 s were. This resulted in a surrounding of copper in S1 s print, and thus lower parts on the gelatin (the same height as valleys in the print). A combination of all of the above. 9.3 Extensive experiments This section discusses and analyzes the method used in and the results acquired from the extensive experiments. The fooling of the fingerprint systems often occurred in the beginning of the trials. This indicates that the quality of the artificial fingerprints is important Experience Both subjects S1 and S2 got experience from using fingerprint scanners on the CeBIT trade fair. S2 also had experience from using the three tested fingerprint systems before the actual tests began. Both these things could have affected the results by increasing the success rate with their real fingerprints. The tester also got a lot of experience from the fingerprint scanners because while developing the artificial fingerprints, they were tested on the same scanners to be

104 84 Discussion and analysis used during the extensive experiments. Also, when you try the same gelatin fingerprint on a scanner a number of times after each other, like in these experiments, you can get used to the way you should place the fingerprint in order for the system to accept it. In a real-world case, you might not have access to the scanner and can therefore not get this experience. The tester had seen about 150 trials done by each subject before the tester performed the experiments with the subjects artificial fingerprints. Being able to watch the way each subject presented his/her finger could have influenced the tester and thus affected the results by giving a higher FAR with artificial fingerprints Subjects The subjects were not obligated to lift their hands from the table between the trials. Lifting only the finger could make the finger placement on the sensor more similar between different trials, than it would if the whole hand was lifted. Lifting the entire hand would probably have imitated a real life situation more and would probably have given a somewhat lower success rate. The subjects were not obligated to wash their hands before the testing began. This way it resembled a real life situation, both when it comes to presenting a finger to a fingerprint scanner, and when it comes to leaving a latent print on a surface. The subjects right index fingers were checked for scars and dirt before the testing began. If any scars or dirt would have been found, the right middle finger would have been chosen instead. This was however not the case for any of the subjects. Avoiding scars and dirt on fingers used in the experiments, eliminated one of the uncertainty and erroneous factors. S3 s print had many minutiae points, especially in the upper part of the print. This could affect how easy/difficult it is to forge a fingerprint. S3 s enhanced print was however very diffuse in the middle, thus making it very difficult to get a good mold. Also, S3 s print was a lot bigger than S1 s and S2 s prints, thus making the center of the fingerprint even more important on the artificial fingerprint (since the sensing areas were comparably small) Initial test results Since the extensive experiments were performed after the CeBIT experiments, both subject S1 and subject S2 had got used to a lot of different fingerprint scanners and the way their fingers should be presented in order for the scanners to successfully identify/verify them. This has most certainly affected the results. In fact, S2 also tried two of the scanners 50 times each, before going to CeBIT. These tests were only performed as a check while developing the gelatin artificial fingerprints and were initially not intended to be used for any real experiments. The results are

105 9.3 Extensive experiments 85 however interesting so they cannot be left out at this point. The results are shown in figure 9.1. The same enrollment was used as in the real extensive tests performed later. For the Precise TM Biometrics 100 MC scanner, the exact same method, as described in section on page 65, was used. For the Identix scanner, only verification was tried by locking the computer, and then logging in again. Figure 9.1. Results of unofficial tests with subject S2 prior to the CeBIT trade fair. S2 tried verifying 50 times each for the Identix scanner and the Precise TM Biometrics 100 MC scanner. The same enrollments as in the real tests were used. The results in figure 9.1, can be compared with the results in figure 8.1 on page 69. The improvement from the initial results to the official results, could be a result of increased familiarity with the systems and how to present the finger in the best way for the scanners to successfully identify/verify him/her. The improvement could also depend on the humidity condition of S2 s finger. If you can learn how to successfully log in to a system with a live finger, it should also be possible to learn how to log in to a system with an artificial fingerprint. Since the tester got experience with presenting artificial fingerprints to the scanners, this could have affected the results The A/R value When comparing the results from the extensive experiments, it is interesting to look at the ratio between the FAR for the artificial fingerprints and the success rate for the real fingerprints. For example, if a low success rate for real fingerprints is

106 86 Discussion and analysis achieved, the FAR with artificial fingerprints will probably also be low. Therefore, the new term Artificial/Real value, or simply A/R value, is introduced and defined in definition 9.1. Definition 9.1 A/R value = FAR for artificial fingerprint Success rate for real fingerprint This assumes that the success rate is greater than 0 %, or the A/R value will be undefined. In most cases, the success rate for real fingerprints will be greater than the FAR for artificial fingerprints, thus giving an A/R value between 0 and 1. A value close to (or equal to) 1 means that the artificial fingerprint was falsely accepted almost (or the same) number of times as the real fingerprint was correctly accepted. A value close to 0 indicates that the artificial fingerprint was of bad quality. A value greater than 1 means that the FAR for the artificial fingerprint was greater than the success rate for the real fingerprint. If the success rate for the real fingerprint for some reason happens to be 0 %, the A/R value is undefined. The A/R value for the different subjects and scanners are shown in table 9.1 and table 9.2 for round one and two respectively. Subject S1 Subject S2 Subject S3 Identix Targus Precise Table 9.1. The A/R value for all subjects and all fingerprint scanners tested in the extensive experiments. Values from round one have been used. Subject S1 Subject S2 Subject S3 Identix Targus Precise Table 9.2. The A/R value for all subjects and all fingerprint scanners tested in the extensive experiments. Values from round two have been used. In the first round, two A/R values were zero due to no success with the artificial fingerprints. Except for those cases, all A/R values were above In the second round, all A/R values except for one, were above This clearly shows that results were improved from round one to round two. The reasons for this improvement are mainly two: The quality of the gelatin solution: After having worked with gelatin artificial fingerprints for a while, you start recognizing when the solution is of good

107 9.3 Extensive experiments 87 quality, i.e. has the right concentration of water etc. Different batches of gelatin solutions were used in the different rounds. In the first round, the end of one batch was used, and in the second round, the beginning of another batch was used. Originally, both batches had the same ratio of gelatin and water, but as the gelatin solution is heated and cooled a number of times and then used for making prints, the gelatin-water ratio changes. The quality of the gelatin solution used during round two, was therefore better than the quality of the gelatin solution used during round one. The experience of the tester: The tester got experience to use specific subjects artificial fingerprints on specific fingerprint scanners. For example, subject S3 s artificial fingerprint had to be presented in a certain way to get accepted. This was however already learned during the first round, but could still have affected the results. The tester also learnt from the first round in which order to test the fingerprint scanners to get the best results. The Precise TM Biometrics 100 MC scanner required a more exact concentration of water in the gelatin solution than the other two scanners seemed to do. Therefore, from having had no specific order between testing the scanners in the first round, the second round always started with testing the Precise TM Biometrics 100 MC scanner. Except for these two differences, the circumstances were comparable between the two rounds. The A/R value of 1.02 in round one and the A/R value of 1.06 in round two, both from the Identix fingerprint scanner, show that artificial fingerprints can in fact be better than using real fingerprints. The probable reason is the experience of the tester from using both artificial fingerprints and using fingerprint scanners, and the lack of experience from the subjects of using fingerprint scanners. Also, it confirms the ideas presented earlier about the possibility to learn how to get high acceptance rates, both when it comes to artificial fingerprints and real fingerprints. None of the tested fingerprint scanners was impossible to fool with the subjects used, even though one system could not be fooled by two of the subjects in the first round. The overall results improved to round two, except for subject S1, which lowered somewhat. Except for the one A/R value of 0.35 in round two, all other A/R values in round two were above This is extremely high values and clearly shows that artificial fingerprints made from latent prints are a serious security threat to fingerprint recognition systems. The reason for the comparably low A/R value of 0.35 could be discussed. Subject S2 s fingerprint is quite small and has quite wide ridges. This made the process of making an artificial copy quite complicated compared to the other subjects. As mentioned in section on page 58, S2 s valleys (appearing as black lines in the image) had to be widened to be able to make a good etching of the print. Also, it must be noted that S2 was the user who had problems logging in on the Precise TM Biometrics 100 MC scanner with the real fingerprint (even after having

108 88 Discussion and analysis reenrolled the finger). Therefore, it is not that surprising that it was difficult to log in with the artificial fingerprints on the same system. Remembering how the initial test results looked, shown in figure 9.1 on page 85, it would be interesting to use these values to calculate the A/R value instead. The A/R value for subject S2 on the Precise scanner, round two, would be 0.75 if the initial test results would have been used to calculate it. Taking this into account, all A/R values would be above 0.75 in the second round. Subject S3 had a A/R value of 0 during the first round when testing the Precise TM Biometrics 100 MC scanner. Since the A/R value for S3 was as high as 0.93 for the same system in round two, the low value in round one was probably due to a bad quality of the gelatin print. Since the Precise scanner was tested last in round one for subject S2 and S3, the gelatin print had probably dried out too much, which was not the case in round two where the Precise scanner was tested first Comparison with results from previous studies During the first round, the FAR with artificial fingerprints was 0 % for two subjects on the Precise scanner. Except for that, all FARs were above 65 % in the first round. In round two, the FAR for subject S2 on the Precise scanner was 12 %. Except for that, all FARs were above 81 %. This can be compared to all FARs for artificial fingerprints above 67 % in [42]. In [42], no subjects nor fingerprint scanners were used that showed the same low success rate as S2 did with real fingerprints on the Precise scanner in these experiments. Taking that into consideration, and therefore mainly looking at the other FARs in these experiments, the results are similar to those in [42]. It should however be noted that the highest security settings were used in [42], while the default security settings were used in the experiments prior to this report. Interesting to note is the results presented in [53], which also tested one of Precise TM Biometrics scanners. The FAR with artificial fingerprints acquired in those experiments, was 2 %. This value can be compared with the mean values acquired in these experiments on the Precise TM Biometrics 100 MC scanner. The mean FAR value in the first round was 33 % and in the second round 62 %. The low results acquired in [53] could be due to lack of good equipment (powder and brush). In [3], artificial fingerprints were made with the legitimate user s cooperation, and they were accepted about % of the time. The mean value of FARs in [3] for all artificial fingerprints and tested scanners, was 38 %. This can be compared with the mean values of FARs with artificial fingerprints of 67 % and 86 % for each round respectively. This is quite interesting to notice since the exact same fingerprint scanners have been tested, and the method used in these experiments is a lot more complicated than the method used in [3]. Using different subjects has of course affected the results and makes it a bit more difficult to compare them. Still, the conclusion can be drawn that it is easier to affect the result to a greater

109 9.4 Additional comments about artificial fingerprints 89 extent when using a latent fingerprint than making a mold directly. For example, if a bad picture has been taken of the enhanced print, it is possible to make it a lot better using an image processing software. 9.4 Additional comments about artificial fingerprints This section discusses some of the additional problems connected to creating artificial fingerprints from latent fingerprints Finding a quality latent fingerprint It is not a trivial task to find a good quality latent fingerprint. First of all, the intruder has to know the whereabouts of the legitimate user. Secondly, the legitimate user must leave a latent with a suitable pressure, without the finger slipping, and with most of the fingerprint touching the surface. The flatness of the surface is also an issue. [41] The latent prints that were the starting point of the experiments, were clearly visible and of good quality. In real life, all latent prints are not of such good quality. Still, we leave about 25 latent fingerprints every day that are of good quality, so the assumption is not totally off track Alternative acquisition of fingerprint image There could be other ways to get hold of the image of someone s fingerprint. For example, in a fingerprint system, the fingerprint image information could be intercepted on the communication channel, or requested by a Trojan horse server. The possibility of synthesizing a two-dimensional fingerprint image from a fingerprint minutiae template could however be questioned. Chris Hill showed in his Bachelor of science thesis that some commercial fingerprint recognition systems can be fooled with synthesized fingerprint images [28]. It is very likely that the same method as described in [28], also could be applied to create physical artificial fingerprints that would fool fingerprint scanners. This method does however require knowledge of the legitimate user. [41] Economies of scale In [41] it is argued that there is no economies of scale to create artificial fingerprints. This is however not completely true. Buying the required equipment costs quite some money the first time an artificial fingerprint is to be created. Also, learning

110 90 Discussion and analysis the creation procedure does take some time. After the initial successful artificial fingerprint has been created, it will however cost much less and take less time to redo the procedure Forging fingerprints The work of Albert Wehde (see section 6.1 on page 49), who created similar molds to the ones used in these experiments, was performed in the 1920 s, about 80 years ago. Doing the work he did, with the cameras at that time, is quite extraordinary, and gives a hint about what should be possible to do today with advanced digital cameras, computers and image processing softwares. It has been questioned whether or not fingerprint examiners and experts can distinguish between a real and a forged fingerprint. Albert Wehde s co-workers at the identification bureau were unable to distinguish genuine latent prints from his forgeries. One reason proposed was that the forgeries could be identified by the absence of sweat pores. Moreover, the sharp outline, due to copper plate engraving, would instantly cause suspicion on the part of the expert. [7] Today s fingerprint examiners insist that no fingerprint forgery is ultimately undetectable since there will always be traces of the fabrication process to detect, e.g. background noise from the surface upon which the fingerprint was deposited [7]. No real tests have however been performed that supports this statement. If it was possible to render the small details such as sweat pores and background noise on the PCB, the gelatin print made from this mold, with some fat etc. on it, could be used to place fingerprints on a crime scene. This latent fingerprint could definitely be identified as belonging to the real person. To get an alibi, the forger could have his/her finger logged in at another place far from the crime scene. How would the court judge in a case like this? Cooperation using latent print If the same method as described in this report is used, but with the cooperation of the owner of the fingerprint, the artificial finger can become even better. The fingerprint to be duplicated can then be studied in detail, finding important singularity points and minutiae points. This way you can make sure that some of the most important information in the fingerprint is also found in the picture of the fingerprint. And if it is not there, you can simply add it with an image processing software. Even without the cooperation of the owner of the fingerprint, the position of the minutiae and the type of minutiae can be guessed to a great extent to enhance the hardly visible details even more in the picture.

111 9.4 Additional comments about artificial fingerprints Using the artificial fingerprint With the experiment method used, a wafer-thin artificial fingerprint can be created. This tiny, almost transparent fingerprint, will easily fool guards if used appropriate. Since gelatin is used, the fingerprint, i.e. the evidence, can be eaten after usage. A laptop with a (built in) fingerprint scanner, can be stolen, fingerprints will definitely be found on the laptop and can thus be used to create an artificial fingerprint. This way, the thief can steal the laptop and get access to all the files on it.

112 92 Discussion and analysis

113 Chapter 10 Conclusion This chapter contains a final conclusion and recommendations and speculations about the future regarding liveness detection, fingerprint systems, and artificial fingerprints Final conclusion All tested fingerprint systems were defeated with artificial fingerprints. Some systems were easier to fool than others, and some artificial fingerprints were more successful than others. Interesting to notice is that a capacitive, an electric-field, and a thermal sweeping sensor were all circumvented with artificial fingerprints. Still, fingerprint recognition systems can be very useful if used in the right applications under the right circumstances. Depending on the specific application, the users of the system might be able to accept a possibility of intrusion with artificial fingerprints instead of paying a lot extra in terms of money, inconvenience, etc. for a liveness detection/extra means of security that still will not be 100 % secure. In another application, the users might demand a very high level of security with high costs, larger acquisition times, more user inconvenience, a higher FRR, etc. It is very important to consider all these factors before starting to use a fingerprint recognition system. Even though it is possible to circumvent a fingerprint scanner with help of an artificial fingerprint, the question can be asked how often this will happen and what the consequences will be. It is very difficult to know how often the attack would take place, but how severe the consequences would be is easier to find out. Liveness detection is definitely a good way to increase the security if it does not increase the costs, FRRs, acquisition time, and user inconvenience to a great extent. 93

114 94 Conclusion Other means of increasing the security were discussed in section 5.6 on page 45. Much research is currently being performed in the area of multi-modal biometrics, and it is something that could be more widely used in the future. Otherwise, combining two or more identification/verification methods, is a security-increasing method that is widely used in commercial applications. One of the most simple and cheap means of protecting against attacks with artificial fingerprints, is by using a verification system with personal smart cards where each user s fingerprint template is stored. An intruder would have to get hold of both the user s smart card and the latent fingerprint. Furthermore, by storing the fingerprint information on the smart card instead of storing it in a central database, another possible attack is removed. If an even higher security level is required, this type of system could also be integrated with a password check Future work A lot of studies have been performed in the area of attacks with artificial fingerprints on fingerprint scanners. Still, as fingerprint scanners develop, more testing and development of artificial fingerprints, is also needed. There have been discussions about integrating fingerprints into passports and identification cards, using fingerprint recognition systems in border controls, and for airport travel. With regards to how relatively easy it is to fool a fingerprint recognition system with artificial fingerprints, further research is needed before this becomes reality. The following subsections will describe the future work needed in the fields of liveness detection, artificial fingerprints, and fingerprint scanners Liveness detection Many liveness detection methods have been suggested, but few have been tested and evaluated, especially not by third parties. Further development and testing of the perspiration method is currently being performed at the Biomedical Signal Analysis Laboratory at West Virginia University, USA. If this further development and testing is successful, evaluation and testing by a third party is also necessary before the method is used commercially. A big issue with the perspiration method is the acquisition time, which has to be less than today s five seconds Artificial fingerprints Automating parts of the process of creating artificial fingerprints, and simplifying other parts, could be a step towards a mass production of artificial fingerprints. Will it be possible to buy an artificial fingerprint in the future?

115 10.2 Future work 95 Coming up with a way to store gelatin artificial fingerprints for a longer time than a week, e.g. by adding a preservative, would increase the use of them. Further research is needed to investigate if it is possible to create artificial fingerprints with pores and a simulation of the perspiration process in fingertips Fingerprint scanners The extensive experiments in this report showed that electric field sensors could be defeated with artificial fingerprints. However, no ultrasound scanners have yet been tested by an independent actor regarding attacks with artificial fingerprints. As more sweeping sensors using different technologies, are being developed, these also need to be tested with regards to attacks with artificial fingerprints Alternative biometrics An alternative approach to further investigation in the biometric area is to investigate which biometric that is really suited to be used in high security applications, e.g. passports, identification cards, and border controls. While fingerprints might be best suited for low security applications, other biometrics might be better to use in other applications.

116 96 Conclusion

117 Bibliography [1] KTM Krim. Teknisk Materiel AB. Available at index.html [accessed 12/05/04]. [2] Precise TM Biometrics. Available at [accessed 12/05/04]. [3] J. Blommé. Evaluation of biometric security systems against artificial fingers. Master s thesis LITH-ISY-EX , Department of Electrical Engineering, Linköping University, Linköping, Sweden, October [4] D. Bursky. Biometric sensor leverages e-fields to improve fingerprint capture. Electronic Design, [5] E. D. Campbell. Fingerprints & palmar dermatoglyphics, Available at [accessed 12/05/04]. [6] Fingerprint Cards. Available at [accessed 12/05/04]. [7] S. A. Cole. Suspect Identitites A History of Fingerprinting and Criminal Identification. Harvard University Press, Cambridge, Massachusetts, London, England, [8] Symantec Corporation. What is the difference between viruses, worms, and trojans? Available at docid/ [accessed 17/05/04]. [9] R. Derakhshani, S. A. C. Schuckers, L. Hornak, and L. O Gorman. Determination of vitality from a non-invasive biomedical measurement for use in fingerprint scanners. Pattern Recognition, 36(2): , [10] The Free Dictionary. Available at dermatoglyphics [accessed 12/05/04]. [11] Dr. E. Hill and Dr. M. D. Stoneham. Practical applications of pulse oximetry, Available at 01.htm [accessed 12/05/04]. 97

118 98 Bibliography [12] Personal from L. Hallberg (Head of the Fingerprint Group at the Swedish National Laboratory of Forensic Science) to M. Sandström, May [13] Personal from U. Söderholm (Swedish forensic photographer) to M. Sandström, February [14] ELFA. Available at [accessed 12/05/04]. [15] ELFA. Factsheet - PCB production, Available at en/fakta.pdf [accessed 12/05/04]. [16] S. A. C. Schuckers et al. Liveness phase ii: Solidifying citer s core competency, May Available at spring/Solidifying%20Liveness.ppt [accessed 12/05/04]. [17] National Kidney Federation. Glossary. Available at uk/medical-info/glossary/glossary.html [accessed 18/05/04]. [18] findbiometrics. Available at glossary.html [accessed 12/05/04]. [19] Center for Nonproliferation Studies at the Monterey Institute of International Studies. Glossary. Available at research/e6 glossary.html [accessed 18/05/04]. [20] E. German. Cyanoacrylate (superglue) fuming tips, Available at http: //onin.com/fp/cyanoho.html [accessed 12/05/04]. [21] Bergdata Biometrics GmbH. Fingerprint touchless sensors, Available at [accessed 12/05/04]. [22] International Biometric Group. The henry classification system, Available at 20Classification.pdf [accessed 12/05/04]. [23] International Biometric Group. Liveness detection in biometric systems, White paper. Available at reports/liveness.html [accessed 12/05/04]. [24] International Biometric Group. Optical silicon ultrasound, White paper. Available at finger-scan optsilult.html [accessed 12/05/04]. [25] T. Harris. Howstuffworks: How fingerprint scanners work. Available at computer.howstuffworks.com/fingerprint-scanner2.htm [accessed 28/05/04]. [26] Visual Health and Surgical Centers. Glossary of vision terms. Available at [accessed 18/05/04].

119 Bibliography 99 [27] M. Henriksson. Analys av fingeravtryck. Master s thesis LITH-ISY-EX- ET , Department of Electrical Engineering, Linköping University, Linköping, Sweden, June [28] C. J. Hill. Risk of masquerade arising from the storage of biometrics. Bachelor of science thesis, The Department of Computer Science, Australian National University, Australia, November [29] Biometric ID. Finger-scan (fingerprints). Available at biometricid.org/finger.html [accessed 12/05/04]. [30] Identix Inc. BioLogon TM Product Guide, first edition, May [31] AuthenTec Inc. Available at cfm [accessed 12/05/04]. [32] Identix Inc. Available at [accessed 12/05/04]. [33] Kingston Technology Company Inc. The glossary. Available at kingston.com/tools/umg/umg10.asp [accessed 18/05/04]. [34] Jr. J. D. Woodward, N. M. Orlands, and P. T. Higgins. Biometrics: Identity assurance in the information age. McGraw-Hill/Osborne, Berkeley, California, USA, [35] A. K. Jain, S. Prabhakar, and S. Pankanti. Can identical twins be discriminated based on fingerprints? In Pattern Recognition, volume 35, pages , [36] P. Kallo, I. Kiss, A. Podmaniczky, and J. Talosi. Detector for recognizing the living character of a finger in a fingerprint recognizing apparatus. Dermo Corporation, Ltd., US Patent #6,175,641, January [37] P. D. Lapsley, J. A. Lee, Jr. D. F. Pare, and N. Hoffman. Anti-fraud biometric sensor that accurately detects blood flow. SmartTouch, LLC., US Patent #5,737,439, April [38] V. Levesque. Measurement of skin deformation using fingerprint feature tracking. Master s thesis, Department of Electrical and Computer Engineering, McGill University, Montréal, France, November [39] A. Ligon. An investigation into the vulnerability of the siemens id mouse professional version 4, September Available at com/knowhow/idm4vul.htm [accessed 12/05/04]. [40] Mason Vactron Ltd. Dcs-3. Available at PDF/DCS-3.pdf [accessed 12/05/04]. [41] D. Maltoni, D. Maio, A. K. Jain, and S. Prabhakar. Handbook of Fingerprint Recognition. Springer Verlag, New York, NY, USA, June 2003.

120 100 Bibliography [42] T. Matsumoto, H. Matsumoto, K. Yamada, and S. Hoshino. Impact of artificial gummy fingers on fingerprint systems. In Proceedings of SPIE Vol. #4677, Optical Security and Counterfeit Deterrence Techniques IV, Yokohama, Japan, January Yokohama National University. [43] P. McFedries. The word spy biometrics. Available at com/words/biometrics.asp [accessed 12/05/04]. [44] G. Moore. The history of fingerprints, February Available at http: //onin.com/fp/fphistory.html [accessed 12/05/04]. [45] Nationalencyklopedin och Språkdata (ordboksartiklar), Available at [accessed 26/05/04]. [46] D. Osten, H. M. Carim, M. R. Arneson, and B. L. Blan. Biometric, personal authentication system. Minnesota Mining and Manufacturing Company, US Patent #5,719,950, February [47] Wi-Fi Planet. Available at equal error rate.html [accessed 18/05/04]. [48] Manukau police. Fingerprints, Available at countiesmanukaupolice.govt.nz/specialists/forensic Services/fingerprints. htm [accessed 12/05/04]. [49] Ridges and Furrows. Friction skin. Available at homestead.com/friction skin.html [accessed 12/05/04]. [50] Schlumberger. Oilfield glossary. Available at oilfield.slb.com/display.cfm?term=relative%20dielectric%20permittivity [accessed 12/05/04]. [51] S. A. C. Schuckers. Spoofing and anti-spoofing measures. Information Security Technical Report, 7(4):56 62, December [52] Softex Inc., Austin, TX, USA. OmniPass User s Guide, Version 1.1. [53] A. Stén, A. Kaseva, and T. Virtanen. Fooling fingerprint scanners - biometric vulnerabilities of the precise TM biometrics 100 sc scanner. In 4th Australian Information Warfare and IT Security Conference 2003, Helsinki, Finland, Telecommunication Software and Multimedia Laboratory, Helsinki University of Technology. [54] Targus. Available at details.asp?sku= PA460E [accessed 12/05/04]. [55] L. Thalheim, J. Krissler, and P-M. Ziegler. Body check biometric access protection devices and their programs put to the test. c t magazine, 1(11):114, May 2002.

121 Bibliography 101 [56] T. Trozzi, R. Schwartz, and M. Hollars. Processing Guide for Developing Latent Prints. Latent Print Unit of the Federal Bureau of Investigation, Washington DC, USA, [57] T. van der Putte and J. Keuning. Biometrical fingerprint recognition: don t get your fingers burned. In Proceedings of IFIP TC8/WG8.8 Fourth Working Conference on Smart Card Research and Advanced Applications, pages Kluwer Academic Publishers, September [58] BVDA Bureau voor Dactyloscopische Artikelen. Available at bvda.com/en/index.html [accessed 12/05/04]. [59] P. A. Wertheim. Black powder processing. Minutiae, The Lightening Powder Co. Newsletter, [60] D. Willis and M. Lee. Six biometric devices point the finger at security. Network Computing, June 1998.

122 102 Bibliography

123 Appendix A Dictionary An alphabetized explanatory list of abbreviations, technical terms, and medical terms used in this report follows. AFIS Abbreviation for Automated Fingerprint Identification System. The system compares a single fingerprint with a database of fingerprint images. AFIS systems are used both in forensics and in the security field. [18] Area sensor See Touch sensor. Artificial fingerprint A fingerprint made to imitate a real fingerprint. In the experiments performed prior to this report, the artificial fingerprints were made out of gelatin, but they can also be made out of silicone or other materials. The term artificial fingerprint should not be confused with the term fake fingerprint, which may also include fingerprints which are modified from live fingers. Authentication See Verification. Bifurcation The point on a papillary line where it divides into two. BVDA Abbreviation for Bureau Voor Dactyloscopische Artikelen. BVDA is a manufacturer and distributor of materials and equipment for crime scene officers and forensic laboratories since CCD Abbreviation for Charge Coupled Device. An array of light-sensitive diodes, generate an electrical signal in response to light photons. Each light-sensitive diode records a pixel representing the light that hit that spot. Together, the array of diodes, form an image of the scanned object. CCDs are used in digital cameras and camcorders, as well as in optical fingerprint scanners. Many of today s optical sensors use a CMOS technology however. [25] Challenge-response Challenge-response is used to determine the presence of a 103

124 104 Dictionary person. Challenge-response can be either voluntary (behavioral) or involuntary (reflexive) responses. In a voluntary challenge-response system, the user will hear, see, or feel something and do something in response. In an involuntary challenge-response system, the user s body automatically responds to a stimulus. Examples include muscles responding to electrical stimulation, the dynamic change in the color of skin when pressure is applied, and the reflex of a knee when struck. [34] Conductivity The ability of a material to allow electrons to flow, measured by the current per unit of voltage applied. Electrical conductivity is proportional to the inverse of resistance. Core point This point is the center of the print. In a whorl pattern, the core point is found in the middle of the spiral/circles. In a loop pattern, the core point is found in the top region of the innermost loop. CMOS Abbreviation for Complementary Metal Oxide Semiconductor. CMOS circuits are widely used in building small low speed and low power electronic systems. [3] Cyanoacrylate fuming Cyanoacrylate fuming is a method used to enhance latent fingerprints on nonporous specimens. See [56] and [20] for more information. Delta point Part of a fingerprint pattern which looks similar to the Greek letter delta. Dermatoglyphics The study of whorls, loops, and arches in fingertips and on palms of the hand and soles of the feet. [10] Dermis An inner layer of the skin found deeper than the outmost layer (epidermis). DNA Abbreviation for deoxyribonucleic acid. DNA molecules contain the genetic information used for the organization and functioning of most living cells. [19] ECG Abbreviation for electrocardiogram. Measurement of the electrical activity within the heart. EKG See ECG. EER Abbreviation for Equal Error Rate. The value of the FAR and FRR when the FAR equals the FRR. This is the value where both the FAR and FRR are kept as low as possible at the same time. [47] Enrollment The process of storing a profile (template) containing the biometrical properties of a person. Epidermis The outermost skin layer.

125 105 FA Abbreviation for False Acceptance. A false acceptance occurs when a system falsely accepts a nonregistered (or another registered) fingerprint as a registered one. [3] FAR Abbreviation for False Acceptance Rate. The rate at which the system falsely accepts a nonregistered (or another registered) fingerprint as a registered one compared to the total number of trials. [3] FBI Abbreviation for Federal Bureau of Investigation. FR Abbreviation for False Rejection. falsely rejects a register user. [3] A false rejection occurs when a system FRR Abbreviation for False Rejection Rate. The rate at which the system falsely rejects a registered user compared to the total number of trials. [3] FTIR Abbreviation for Frustrated Total Internal Reflection. See section on page 23 for more information. Gelatin Gelatin is made by dissolving collagen (a protein found in bone and connective tissues) in a hot solution [45]. Since gelatin is made out of collagen, it resembles the surface of human skin in ways of moisture, electric resistance, and texture [3, 42]. Gelatin is used e.g. in candies and cooking as a thickening agent. Haemoglobin The part of red blood cells that carry oxygen to tissues. [17] Identification In an identification system, an individual is recognized by comparing with an entire database of templates to find a match. The system conducts one-to-many comparisons to establish the identity of the individual. The individual to be identified does not have to claim an identity (Who am I?). [41] KTM Abbreviation for Krim. Teknisk Material AB. KTM is a Swedish manufacturer of forensic laboratory equipment. LED Abbreviation for Light Emitting Diode. Latent fingerprint A fingerprint consists of a combination of different chemicals that originate from natural secretions, blood, and contaminants. When a fingerprint touches a surface it leaves traces of these chemicals. These traces make up a latent fingerprint (sometimes called residual fingerprint). [56] Lipid A lipid is an organic substance which is not dissolvable in water, but dissolves itself in other organic solvents. Fat and waxes are two examples of lipids. [45] Live finger A human finger which is still living, i.e. has blood pumping through it etc. The terms real finger and live finger are used synonymous in this report. Liveness detection Liveness detection (sometimes called vitality detection) in a biometric system means the capability for the system to detect, during en-

126 106 Dictionary rollment and identification/verification, whether or not the biometric sample presented is alive or not. Furthermore, if the system is designed to protect against attacks with artificial fingerprints, it must also check that the presented biometric sample belongs to the live human being who was originally enrolled in the system and not just any live human being. Minutiae details The characteristics by which fingerprints can be identified. Minutiae details are sometimes referred to as Galton s details, ridge characteristics, or ridge details. Multi-modal biometrics A multi-modal biometric system, combines two or more biometric techniques. Papillary lines The lines on a fingerprint that are visible to the human eye. The parts of the lines that are raised are called ridges and the lines that are lower are called valleys. PCB Abbreviation for Printed Circuit Board. A PCB is a board with copper traces which are normally used to provide electrical connections for chips and other components [33]. Two examples of PCBs are mother-boards and credit card memory [33]. In this report, PCBs are used to get a three-dimensional fingerprint as a mold. Perspiration Human skin contain about 600 sweat glands per square inch. Sweat (a dilute sodium chloride solution) diffuses from the sweat glands on to the surface of the skin through small pores. In live fingers, perspiration starts from the pores. The sweat then diffuses along the ridges during time, making the semi-dry regions between the pores moister or darker in the image. The perspiration process does not occur in cadaver fingers or artificial fingerprints. [9] PIN Abbreviation for Personal Identification Number. Pore Sweat pores on the human skin are used for the perspiration process. Pulse oximetry Pulse oximetry is used in the medical field to measure the oxygen saturation of haemoglobin in a patient s arterial blood. [11] Relative Dielectric Permittivity The degree to which a medium resists the flow of electric charge divided by the degree to which free space resists such charge. The degree, or dielectric permittivity, is defined as the ratio of the electric displacement to the electric field strength. The term is also known as the relative dielectric constant (RDC). However, at high frequencies, the value is no longer constant, but decreases with frequency. The relative dielectric permittivity is unitless. [50] Real finger/fingerprint A real finger is a human finger which is still living, i.e. has blood pumping through it etc. The terms real finger and live finger are used synonymous in this report. A real fingerprint is a fingerprint on a real/live finger.

127 107 Residual fingerprint See Latent fingerprint. Retina The nerve layer that lines the inside wall of the eye. With a similar function as the film in a camera, the retina captures images, transforms the images into electrical signals, and sends the signals to the brain. [26] Ridge A part of the fingerprint that like a ridge in a landscape raises itself above the rest of the area. Ridges show up as dark parts in the image of the scanned fingerprint. [3] Ridge termination A ridge termination is the end point of a ridge, and is therefore sometimes referred to as an ending point. Smart card A plastic card which has a microprocessor chip embedded inside the card. [6] Success rate The rate at which the system successfully identifies/verifies a registered user s fingerprint compared to the total number of trials. [3] Sweeping sensor A new type of fingerprint sensor, the sweeping sensor, is as wide as a finger, but only a few pixels high. Therefore, the main advantage of sweeping sensors, especially in silicon sensors, is reduced cost. The sweeping consists of a vertical movement only. At the end of the swipe or on-the-fly, the fingerprint image is reconstructed from all the images acquired earlier. [41] Touch sensor Most fingerprint sensors used today are touch sensors (area sensors). When using a touch sensor, you simply put your finger on the sensor and hold it for a few seconds without moving it. Very little user training is required to use a touch sensor. Trojan horse Trojan horses are files that claim to be something desirable, but are in fact malicious. Trojans contain malicious code that when triggered cause loss or theft of data. Trojan horse programs do not replicate themselves but spread e.g. by attachments that are opened or by downloading and running a file from the Internet. [8] Valley The part of the fingerprint that like a valley in a landscape lies lower than the rest of the area [3]. Valleys show up as light parts in the image of the scanned fingerprint [3]. In some literature, the term furrow is used instead of valley. Verification In a verification system, the individual to be identified has to claim his/her identity (Am I whom I claim to be?) and this template is then compared to the individual s biometric characteristics. The system conducts oneto-one comparisons to establish the identity of the individual. Two synonyms to verification are authentication and identity verification. [41] Vitality detection See Liveness detection.

128 108 Dictionary

129 Appendix B Material This appendix contains a detailed description of the material used in the experiments. B.1 Enhancing the fingerprint The material used to enhance the fingerprints is listed below and is available from KTM - Krim. Teknisk Material AB, Box 171, S BÅLSTA. Telephone number: +46(0) See [1] for more details. All material listed below is also found internationally from BVDA (Bureau voor Dactyloscopische Artikelen) but with different catalog numbers. Mail address: BVDA International b.v., Postbus 2323, 2002 CH HAARLEM - The Netherlands. Telephone number: See [58] for more details. Soot powder mixture, 100/500 ml. Catalog number 13400/ Squirrel hair brush SKA, round brush with black lacquered shaft, length: 14 cm. Catalog number B.2 Photographing the fingerprint A Minolta DiMAGE 5 from Minolta Co., Ltd., with the following detailed information, was used during the experiments: Number of effective pixels: 3.17 million ( ). Camera sensitivity (ISO): Auto and 100, 200, 400, and 800 ISO equivalents. 109

130 110 Material Focal length: mm. Focusing range: 0.5 m infinity (from the CCD), m (from the CCD) macro mode. B.3 Image processing Adobe r Photoshop r CS, version 8.0, from Adobe Systems Inc., was used for the graphics processing. B.4 Printing The printer used was a HP LaserJet 5Si/5Si MX PS. The highest resolution the printer could manage was 600 dpi. B.5 PCB production The material needed for exposing, developing, and etching of the mold can be bought from ELFA AB, S Järfälla. Telephone number: +46(0) See [14] for more details. All prices given in this section are exclusive vat. The following material (see figure B.1 on page 111), all available at ELFA, was used during the experiments described in this report [14]: UV exposure, single-sided. Stock number , price 4451 SEK. UV light box for exposing on copper-clad board coated with photoresist. The lid has a snap lock and is fitted with 20 mm thick foam for securing the object. Fitted with timer and supplied with fluorescent lamps. Loading surface: mm No. of fluor. lamps: 4 pcs. Output, fluor. lamps: 15 W/unit UV wavelength: 365 nm Epoxy laminate, 1.55 mm/70 µ Cu. Stock number , price 54 SEK. With photoresist, type FR-4 Copper-clad board, 1.55 mm thick, coated with positive photoresist. Copper layer: 70 µm. Single-sided, size: mm. Epoxy laminate, 1.55 mm/35 µ Cu. Stock number , price 111 SEK. With photoresist, type FR-4 Copper-clad board, 1.55 mm thick, coated with positive photoresist. Copper layer: 35 µm. Single-sided, size: mm.

131 B.5 PCB production 111 Developer for photo resist 20 g. Stock number , price 43 SEK. Powder developer for positive photo resist. Dissolves in 2 liter 30 C water. Contains sodium hydroxide (NaOH). Etching powder 1000 g. Stock number , price 137 SEK. Etching powder for copper laminate. The powder (sodium peroxide sulphate) is dissolved in boiling hot water, thus giving the solution a temperature of 50 C which gives the best result. 140 g of powder etches 6 dm 2 in five minutes at 50 C while stirring. Premixed solution can be stored in a storage box without a tight-fitting lid and the box cannot be made out of plastic. Mixture relationship 1:5. (a) UV light box. (b) Epoxy laminate. (c) Developer for photo resist. (d) Etching powder. Figure B.1. Materials used during production of PCB. [14]

132 112 Material B.6 Gelatin solution Gelatin used for making artificial fingerprints: Favorit Gelatin extra guld (17 g), see figure B.2. Produced by: AB Törsleff & Co, Box 4017, S Solna. Telephone number: +46(0) Costs around 10 SEK. Figure B.2. Gelatin used for making artificial fingerprints. [3]

133 Appendix C Experiment details This appendix describes some of the steps in the experiment method in more detail. C.1 Photographing A tripod and a Minolta DiMAGE 5 digital camera was used with the following settings while taking photos of the lifted fingerprints: Macro mode. No flash. Sharpness: Hard(+). Manual focus. ISO 100. The lowest possible ISO value was used to get the least noise in the image. Quality Fine. The image was saved formatted as a JPEG file. Preferably, the super fine quality should have been used and the image would then have been saved in raw format. This setting was not used however, because many pictures of the prints were taken and with a higher quality it required a lot of space to save the image. Size pixels (horizontally vertically). This size was used since it was equal or greater to the resolution the printer could cope with, and it was big enough to make the image processing easy to perform. Aperture value f/8. The quite small aperture (i.e. quite big aperture value) was used to get the picture sharper. 113

134 114 Experiment details White balance Auto. Contrast compensation +3. After the unsharp mask in Photoshop had been applied, you could however hardly see the difference between using no contrast compensation and using +3 contrast compensation. Self-timer. Shake warning appeared otherwise. Taking photos of the tape and powder was not a problem when considering reflects from the tape. C.2 Image processing Adobe r Photoshop r CS from Adobe Systems Inc., was used for the image processing performed. To start off with, a picture in jpeg format with the size of pixels, was used. The steps of the image processing for the fingerprints used in the experiments in this report will follow. These steps can sometimes be done in a different order, depending on the quality of the photograph, the darkness and sharpness of the picture, scars on the fingertip from which the fingerprint is taken etc. Sometimes you might also want to perform things not mentioned here to make the picture look better. 1. First of all, the picture was cropped with the crop tool to a smaller square containing only the fingerprint (and some parts of the surroundings that can not be cropped using a square). 2. The picture was zoomed in with the zoom tool, and then sharpened with the following filter and values: Filter Sharpen Unsharp Mask... Amount: 500 % Radius: 7.0 pixels Threshold: 0 levels 3. To get it right when performing the etching of the picture later on, the image was reversed by the following selection: Image Rotate Canvas Flip Canvas Horizontal Depending on which way the image is positioned, the canvas should be flipped vertically instead. 4. If necessary, those parts of the print that were more diffuse/had less contrast, were improved by first selecting them with the lasso tool, and then using the levels option:

135 C.2 Image processing 115 Image Adjustments Levels... Input levels changed to appropriate values. 5. The papillary lines were sharpened even more and pores were removed by using the pen tool and the brush tool (black color) with the following values: Master Diameter: 2 9 pixels Hardness: 100 % 6. The colors of the fingerprint image were inverted to make it right when producing the PCB. This should only be done though if the print has been powdered with a dark powder (not white). The reason being that the dark parts of the fingerprint will appear as copper on the PCB and thus valleys on the artificial fingerprint. The white parts will thus appear as empty space on the PCB and thus ridges on the artificial fingerprint. The invert selection is found here: Image Adjustments Invert 7. Noise in the image was removed by the following filter and settings: Filer Noise Dust & Scratches... Radius: 1 pixels Threshold: 0 levels 8. If there were any traces left from the soot powder mixture at this point (showing up as white dots on the black areas), these were erased by either using the Dust & Scratches filter again, or by erasing them by hand with the pen tool and the brush tool. 9. For subject S2 s fingerprint image, the black lines had to be widened a bit for the image to become easier to etch. This was done by using the following filter: Filter Brush Strokes Accented Edges... Edge Width: 2 Edge Brightness: 16 Smoothness: 5 For the other subjects fingerprint images, this step was discarded. 10. The length of the powdered/fumed fingerprint must be measured to be able to get the right size of the image. The fingerprint was measured with an ordinary office ruler with a millimeter scale. The image size was set to the real size of the fingerprint by the following selection: Image Image Size... While changing the image size, the resample option was deselected. After that was done, and before pressing ok, the resolution was changed to

136 116 Experiment details the highest resolution the printer could handle (here 600 dpi). The highest resolution of the fingerprint sensors tested in the extensive experiments, was about 500 dpi, so 600 dpi images should not have been a problem. 11. The image should be black and white and not grayscaled in order for the etching to work better. Therefore, the threshold option was chosen (without changing anything) to make the image black and white. Image Adjustments Threshold... Threshold Level: If there were any traces left now from the soot powder mixture (appearing as white dots on the black parts), those were erased. The image processing of subject S2 s fingerprint followed the above description, except for a few things. First of all, the image did not have a black surrounding of the print (instead it was white). This should however not have affected the result to a great extent, since often this part of the gelatin is not even interpreted by the sensor. It was mainly of laziness reasons why the surrounding was kept black for the other subjects when it did not seem to have any impact on the outcome. Also, in step number four above, the brush size five was used most of the time for the image processing of S2 s fingerprint, thus making the somewhat thinner lines a bit thicker. C.3 Fingerprint images before and after image processing Figures C.1, C.2, and C.3 show a part of the images of all subjects fingerprints before and after image processing. (a) S1 before image processing. (b) S1 after image processing. Figure C.1. A part of subject S1 s fingerprint image before and after image processing.

137 C.4 PCB production 117 (a) S2 before image processing. (b) S2 after image processing. Figure C.2. A part of subject S2 s fingerprint image before and after image processing. (a) S3 before image processing. (b) S3 after image processing. Figure C.3. A part of subject S3 s fingerprint image before and after image processing. C.4 PCB production To expose, develop, and etch the mold, the following equipment and material were used: Ultra-violet radiation (UV light) box. An UV lamp can also be used, but is a bit more tricky to use. Exposing container. Etching bowl of glass. An etching bowl of suitable plastic can also be used. Epoxy laminate with positive photo resist. Developer for photo resist. Etching powder for copper laminate.

138 118 Experiment details Stirrer. All the above is available at ELFA (except the etching bowl and the stirrer), see appendix B.5 on page 110. The following procedure for the production of the PCB was used: 1. The developer for photo resist was mixed. One bag (20 g) of photo resist powder (sodium hydroxide, NaOH) was mixed with 1 liter of water. 2. The etching solution was mixed. 2 dl of etching powder (sodium peroxide sulphate) was mixed with 1 liter of boiling hot water, giving an etching solution of about +50 C. 3. The protective film was removed from the epoxy laminate. 4. The transparency film with the fingerprint image, was placed on the UV light box glass. The blackness side of the transparency should be turned upwards (away from the glass side of the UV light box and towards the emulsion side of the resist-coated board to be placed on top of it). The epoxy laminate was then placed on top of the transparency. To get a more even pressure, a pile of papers were placed on top of the epoxy laminate before closing the lid of the UV light box. 5. The exposure time varies with the height of the lamp above the illuminated board and any possible glass sheets between the lamp and film laminate. The best time to use with the tools available here, was experimentally determined to 3 minutes. 6. The exposure was followed by developing with the developer mixture prepared earlier. The developing time varies between 30 seconds and 4 minutes depending on the type of resist. The laminate was first developed once, and left in the developer mixture until the pattern appeared, and then a second time to get rid of any remaining photo resist. When positive resist is used, those parts that have not been exposed to light (i.e. are covered with black color on the transparency), will be protected during etching. Before continuing, the board was washed thoroughly with running water. 7. The laminate was put in the etching solution in the etching bowl. To keep a temperature of about +50 C (for the best result), the etching bowl was placed in a hot water bath in the sink. The etching solution was then stirred to make sure the active parts of the solution always was in contact with the copper surface. The etching took about minutes depending on how many fingerprints that were etched at the same time. 8. The laminate was then washed with water again. Note that ELFA recommends to use protective gloves and safety glasses when working with UV light and chemicals. The etching solution should not be poured out in the sink.

139 Appendix D Scanners used in extensive experiments Identix fingerprint scanner (see figure D.1) from Compaq Computer Cooperation (nowadays Hewlett Packard) uses an DFR-300 optical sensor (FTIR with a sheet prism, CMOS-camera). The resolution is 380 dpi and the sensing area size is mm. The operating temperature is 0 50 C. [3, 32] Figure D.1. Identix fingerprint scanner used in the extensive experiments. [30] Both Targus DEFCON TM Authenticator TM (see figure D.2 on page 120) and Precise TM Biometrics 100 MC (see figure D.3 on page 120) use an EntréPad r AES4000 silicon sensor from AuthenTec, Inc. AES4000 is an electric field sensor and has a resolution of 250 dpi. The sensing area size is mm and the operating temperature is 0 70 C. [31] Precise TM Biometrics 100 MC fingerprint scanner also has a built in smart card reader, which makes it possible to store the fingerprint on a smart card and thus get a higher degree of security [2]. 119

140 120 Scanners used in extensive experiments Figure D.2. Targus DEFCON TM Authenticator TM fingerprint scanner used in the extensive experiments. [54] Figure D.3. Precise TM Biometrics 100 MC fingerprint scanner used in the extensive experiments. [2]

141 Appendix E Software used in extensive experiments Screenshots of software used in the extensive experiments can be found in figures E.1, E.2 on page 122, and E.3 on page 122. Figure E.1. Screenshot of BioLogon TM for Windows. BioLogon TM is the software that was used together with the Identix fingerprint scanner. [3] 121

142 122 Software used in extensive experiments Figure E.2. Screenshot of Softex Omnipass. Softex Omnipass is the software that was used together with the Targus DEFCON TM Authenticator TM fingerprint scanner. [3] Figure E.3. Screenshot of Precise BioManager TM included in Precise Logon software. Precise Logon is the software that was used together with the Precise TM Biometrics 100 MC fingerprint scanner. [3]

143 Appendix F Test results The detailed test results for each fingerprint scanner and each subject follow below. F.1 Results per fingerprint scanner The detailed results, for each fingerprint scanner, both for real fingerprints and artificial fingerprints, are presented in this section. F.1.1 Identix The detailed test results of the Identix fingerprint scanner for real fingerprints can be found in table F.1. The detailed test results of the Identix fingerprint scanner for artificial fingerprints, round one and two, can be found in table F.2 and table F.3 on page 124. Subject Successful logins False rejections False acceptances S S S Sum Percent 97.3 % 2.7 % 0.0 % Table F.1. Results of the Identix fingerprint scanner for real fingerprints. 123

144 124 Test results Subject FAs (artificial print) Rejected logins FAs (other user) S S S Sum Percent 81.3 % 18.7 % 0.0 % Table F.2. one. Results of the Identix fingerprint scanner for artificial fingerprints, round Subject FAs (artificial print) Rejected logins FAs (other user) S S S Sum Percent 96.7 % 3.3 % 0.0 % Table F.3. two. Results of the Identix fingerprint scanner for artificial fingerprints, round F.1.2 Targus DEFCON TM Authenticator TM The detailed test results of the Targus DEFCON TM Authenticator TM fingerprint scanner for real fingerprints can be found in table F.4. The detailed test results of the Targus DEFCON TM Authenticator TM fingerprint scanner for artificial fingerprints, round one and two, can be found in table F.5 and table F.6 on page 125. Subject Successful logins False rejections False acceptances S S S Sum Percent % 0.0 % 0.0 % Table F.4. Results of the Targus DEFCON TM Authenticator TM fingerprint scanner for real fingerprints.

145 F.1 Results per fingerprint scanner 125 Subject FAs (artificial print) Rejected logins FAs (other user) S S S Sum Percent 87.3 % 12.0 % 0.7 % Table F.5. Results of the Targus DEFCON TM Authenticator TM fingerprint scanner for artificial fingerprints, round one. Subject FAs (artificial print) Rejected logins FAs (other user) S S S Sum Percent 98.0 % 2.0 % 0.0 % Table F.6. Results of the Targus DEFCON TM Authenticator TM fingerprint scanner for artificial fingerprints, round two. F.1.3 Precise TM Biometrics 100 MC The detailed test results of the Precise TM Biometrics 100 MC fingerprint scanner for real fingerprints can be found in table F.7. The detailed test results of the Precise TM Biometrics 100 MC fingerprint scanner for artificial fingerprints can be found in table F.8 and table F.9 on page 126. Subject Successful logins False rejections False acceptances S S S Sum Percent 73.3 % 26.7 % 0.0 % Table F.7. fingerprints. Results of the Precise TM Biometrics 100 MC fingerprint scanner for real

146 126 Test results Subject FAs (artificial print) Rejected logins FAs (other user) S S S Sum Percent 32.7 % 67.3 % 0.0 % Table F.8. Results of the Precise TM Biometrics 100 MC fingerprint scanner for artificial fingerprints, round one. Subject FAs (artificial print) Rejected logins FAs (other user) S S S Sum Percent 62.0 % 38.0 % 0.0 % Table F.9. Results of the Precise TM Biometrics 100 MC fingerprint scanner for artificial fingerprints, round two. F.2 Results per subject The detailed results per subject, in numbers and percent, for both real fingerprints and artificial fingerprints, are presented in this section. F.2.1 Real fingerprints The detailed results per subject, in numbers, for real fingerprints are shown in table F.10. The success rate, FRR, and FAR, per subject, for real fingerprints, as well as the mean values for all subjects, are shown in table F.11 on page 127. Subject Successful logins False rejections False acceptances S S S Sum Table F.10. The sum of successful logins, false rejections, and false acceptances, per user, for all fingerprint scanners tested, for real fingerprints.

147 F.2 Results per subject 127 Subject Success rate FRR FAR S % 1.3 % 0.0 % S % 22.0 % 0.0 % S % 6.0 % 0.0 % Mean value 90.2 % 9.8 % 0.0 % Table F.11. The success rate, FRR, and FAR, per subject, for all fingerprint scanners tested, for real fingerprints. The mean values for all subjects are also shown. F.2.2 Artificial fingerprints The detailed results per subject, in numbers, for artificial fingerprints, are shown in table F.12 and table F.13 for round one and two respectively. The FAR (with artificial fingerprints), rejection rate, and FAR (other user), per subject, for artificial fingerprints, as well as the mean values for all subjects, are shown in table F.14 and table F.15 on page 128. Subject FAs (artificial print) Rejected logins FAs (other user) S S S Sum Table F.12. The sum of false acceptances (with artificial fingerprints), rejected logins, and false acceptances (other user), per subject, for all fingerprint scanners tested, for artificial fingerprints during round one. Subject FAs (artificial print) Rejected logins FAs (other user) S S S Sum Table F.13. The sum of false acceptances (with artificial fingerprints), rejected logins, and false acceptances (other user), per subject, for all fingerprint scanners tested, for artificial fingerprints during round two.

148 128 Test results Subject FAR (artificial print) Rejection rate FAR (other user) S % 0.7 % 0.0 % S % 54.7 % 0.0 % S % 42.7 % 0.7 % Mean value 67.1 % 32.7 % 0.2 % Table F.14. The FAR (with artificial fingerprint), the rejection rate, and the FAR (other user), per subject, for artificial fingerprints during round one. The mean values for all subjects are also shown. Subject FAR (artificial print) Rejection rate FAR (other user) S % 4.0 % 0.0 % S % 32.0 % 0.0 % S % 7.3 % 0.0 % Mean value 85.6 % 14.4 % 0.0 % Table F.15. The FAR (with artificial fingerprint), the rejection rate, and the FAR (other user), per subject, for artificial fingerprints during round two. The mean values for all subjects are also shown.

149 På svenska Detta dokument hålls tillgängligt på Internet eller dess framtida ersättare under en längre tid från publiceringsdatum under förutsättning att inga extraordinära omständigheter uppstår. Tillgång till dokumentet innebär tillstånd för var och en att läsa, ladda ner, skriva ut enstaka kopior för enskilt bruk och att använda det oförändrat för ickekommersiell forskning och för undervisning. Överföring av upphovsrätten vid en senare tidpunkt kan inte upphäva detta tillstånd. All annan användning av dokumentet kräver upphovsmannens medgivande. För att garantera äktheten, säkerheten och tillgängligheten finns det lösningar av teknisk och administrativ art. Upphovsmannens ideella rätt innefattar rätt att bli nämnd som upphovsman i den omfattning som god sed kräver vid användning av dokumentet på ovan beskrivna sätt samt skydd mot att dokumentet ändras eller presenteras i sådan form eller i sådant sammanhang som är kränkande för upphovsmannens litterära eller konstnärliga anseende eller egenart. För ytterligare information om Linköping University Electronic Press se förlagets hemsida In English The publishers will keep this document online on the Internet - or its possible replacement - for a considerable time from the date of publication barring exceptional circumstances. The online availability of the document implies a permanent permission for anyone to read, to download, to print out single copies for your own use and to use it unchanged for any non-commercial research and educational purpose. Subsequent transfers of copyright cannot revoke this permission. All other uses of the document are conditional on the consent of the copyright owner. The publisher has taken technical and administrative measures to assure authenticity, security and accessibility. According to intellectual property law the author has the right to be mentioned when his/her work is accessed as described above and to be protected against infringement. For additional information about the Linköping University Electronic Press and its procedures for publication and for assurance of document integrity, please refer to its WWW home page: Marie Sandström