FACE RECOGNITION SECURITY SYSTEM FOR ACCESS CONTROL NUR ADILA BINTI MOHD RAZALI UNIVERSITI TEKNOLOGI MALAYSIA

Transcription

1 FACE RECOGNITION SECURITY SYSTEM FOR ACCESS CONTROL NUR ADILA BINTI MOHD RAZALI UNIVERSITI TEKNOLOGI MALAYSIA

2 UNIVERSITI TEKNOLOGI MALAYSIA PSZ 19:16 (Pind. 1/07) DECLARATION OF THESIS / UNDERGRADUATE PROJECT REPORT AND COPYRIGHT Author s full name : NUR ADILA BINTI MOHD RAZALI Date of Birth : 22 ND MAY 1991 Title : FACE RECOGNITION SECURITY SYSTEM FOR ACCESS CONTROL I hereby declare that I have read this thesis and in my opinion this thesis is sufficient in terms of scope and quality for the Academic Session : 2013/2014 I declare that this thesis is classified as: award of the Bachelor s Degree of Electrical and Electronic Engineering (Instrumentation & Control) CONFIDENTIAL (Contains confidential information under the Official Secret Act 1972)* RESTRICTED (Contains restricted information as specified by the organization where research was done)* OPEN ACCESS I agree that my thesis to be published as online open access (full text) Signature :. I acknowledged that Universiti Teknologi Malaysia reserves the right as follows: Name of Supervisor : 1. The thesis is the property of Universiti Teknologi Malaysia. Date :. 2. The Library of Universiti Teknologi Malaysia has the right to make copies for the purpose of research only. 3. The Library has the right to make copies of the thesis for academic exchange. Certified by: SIGNATURE NUR ADILA BINTI MOHD RAZALI ( ) SIGNATURE OF SUPERVISOR ASSOC. PROF.IR.DR HAZLINA BINTI SELAMAT NOTES: * If the thesis is CONFIDENTAL or RESTRICTED, please attach with the letter from Date: 18 JUN 2014 Date: 18 JUN 2014 the organization with period and reasons for confidentiality or restriction.

3 I hereby declare that I have read this thesis and in my/our* opinion this thesis is sufficient in terms of scope and quality for the award of the degree of Bachelor of Engineering (Instrumentation and Control) Signature :... Name of Supervisor : Assoc Prof.Ir.Dr Hazlina Binti Selamat Date :..

4 FACE RECOGNITION SECURITY SYSTEM FOR ACCESS CONTROL NUR ADILA BINTI MOHD RAZALI A thesis submitted in fulfillment of the requirements for the award of the degree of Bachelor of Engineering (Electrical - Instrumentation and Control) Faculty of Electrical Engineering Universiti Teknologi Malaysia JUNE 2014

5 ii I declare that this thesis entitled Face Recognition Security System for Acceess Control is the result of my own research except as cited in the references. Signature Name of Candidate Date :... : NUR ADILA BINTI MOHD RAZALI :

6 iii Dedicated, in thankful appreciation for Support, encouragement and understanding to my beloved parents, Mohd Razali and Mazni my sibling, and my friends for their love and sacrifice

7 iv ACKNOWLEDGMENT First and foremost, praise be to Allah because of His love and strength that He has given to me to complete my final year project entitled Face Recognition Security System for Access Control. I do thank for His blessings in my daily life, good health, healthy mind and good ideas although I have to go through some difficulties along the way. I would like to express my gratitude and appreciation to those who had contributed to the completion of this project. First, I would like to dedicate a special thanks to my supervisor, Assoc Prof Ir Dr Hazlina Selamat for the support, advices, and experiences that I gained in completing my final year project. Next, I would like to express my gratitude to Dr Mohd Amri Bin Md Yunus, the final year project coordinator, for all the information and guidance given. Not to forget, I would also like to acknowledge with much appreciation to my family members for their concern, encouragement and understanding. Finally, thanks to those who have contributed directly or indirectly to the success of this project whom I have not mentioned their name specifically. Without them, this project would not successful. Thank You.

8 v ABSTRACT In recent years, security systems have become one of the most demanding systems to secure our assets and protect our privacy. A more reliable security system should be developed to avoid losses due to identity theft or fraud. Thus, a lot of researches have been done in order to improve established security system, especially systems that are based on human identification. Face recognition system is widely used in human identification process due to its capability to measure and subsequently identifies human identification especially for security purposes. The aim of this project is to develop a real-life application of a security lock system using a face recognition method. Principal Component Analysis (PCA) is selected for the face recognition algorithm due to its fast response of recognition process and less sensitive to noise and interference. A Graphical User Interface (GUI) is developed for this system and the Arduino microcontroller is used to switch off the magnetic door in response to positive face identification. USB serial communication is used to interface between the GUI in MATLAB and Arduino UNO microcontroller as it allows input data transmission from MATLAB to Arduino UNO microcontroller. First, the image of the individual is captured using an integrated webcam. The captured image is then transferred to the database developed in MATLAB. In this stage, the captured image compares to the training image in the database to determine the individual status. If the system recognizes the individual as an authorized person, the signal will be sent to the Arduino UNO microcontroller and used to grant access to an entry or not At the moment, the system gives 90% of accuracy.

9 vi ABSTRAK Dalam tahun-tahun kebelakangan ini, sistem keselamatan telah menjadi salah satu sistem yang paling dituntut untuk menjamin keselamatan aset sisamping melindungi privasi kita. Satu sistem keselamatan yang lebih dipercayai harus dibangunkan untuk mengelakkan kerugian akibat kecurian identiti atau penipuan. Oleh itu, banyak kajian telah dilakukan untuk memperbaiki sistem keselamatan yang sedia ada, terutama sistem yang berasaskan kepada pengenalan manusia. Sistem pengecaman muka digunakan secara meluas dalam proses pengenalan manusia kerana keupayaannya untuk mengukur dan kemudiannya mengenal pasti identiti manusia terutamanya untuk tujuan keselamatan. Tujuan projek ini adalah untuk membangunkan satu aplikasi kehidupan sebenar sistem kunci keselamatan dengan menggunakan kaedah pengecaman muka. Analisis Komponen Utama (PCA) dipilih untuk algoritma pengecaman muka kerana tindak balas yang cepat untuk proses pengecaman dan kurang sensitif kepada bunyi dan gangguan. Pengguna grafik antara muka (GUI) dibangunkan untuk sistem ini dan Arduino UNO mikropengawal digunakan untuk mematikan pintu magnet sebagai tindak balas kepada pengenalan muka yang positif. USB komunikasi bersiri digunakan untuk antara muka diantara GUI dalam MATLAB dan Arduino UNO mikropengawal kerana ia membolehkan penghantaran data input daripada MATLAB kepada Arduino UNO mikropengawal. Pertama, imej individu itu ditangkap denagn menggunakan kamera web bersepadu. Imej yang ditangkap kemudiannya dipindahkan ke pangkalan data yang dibangunkan dalam MATLAB. Pada peringkat ini, imej yang ditangkap dibandingkan dengan imej latihan di dalam pangkalan data untuk menentukan status individu. Jika sistem mengiktiraf individu sebagai orang diberi kuasa, isyarat akan dihantar ke mikropengawal Arduino dan digunakan untuk memberikan akses kepada entri atau tidak Pada masa ini, sistem ini memberikan 90% ketepatan.

10 vii TABLE OF CONTENTS CHAPTER TITLE PAGE TITLE DECLARATION ii DEDICATION iii ACKNOWLEDGMENT iv ABSTRACT v ABSTRAK vi TABLE OF CONTENTS vii LIST OF TABLES x LIST OF FIGURES xi LIST OF ABBREVIATIONS xii LIST OF APPENDICES xiii 1 INTRODUCTION 1.1 Background of.project Problem Statement Project Objective Project Scope Significance of Project Thesis Outline 4 2 LITERATURE REVIEW 2.1 Door Access Control System 6

11 viii Password/Keypad System Smart Card System Biometric Control System Biometric Technology Face Recognition Feature Extraction Algorithm in Face 11 Recognition Principal Component Analysis Linear Discriminant Analysis Conclusion 14 3 METHODOLOGY 3.1 Introduction Project Planning Project Overview Software Development Image Acquisition Pre Processing Feature Extraction using PCA 19 Algorithm Classification Hardware Development using Arduino UNO 22 Microcontroller 3.6 Conclusion 24 4 RESULT AND DISCUSSION 4.1 Introduction Background Parameter Multi Train Image Variation in Head Position Conclusion 33

12 ix 5 CONCLUSION AND FUTURE WORK 5.1 Conclusion Future Work 35 6 PROJECT MANAGEMENT 6.1 Introduction Project Schedule Cost Estimation 39 REFERENCES 41 APPENDICES A-B 43

13 x LIST OF TABLES TABLE TITLE PAGE 2 1 Comparison of various biometric technologies The result of uncontrolled parameters The result of controlled background parameter The result of using multi rain image in train 29 database 4.4 The result of the variation in head position The summary of the analysis Cost estimation of face recognition security 39 system 6.2 Cost estimation of RFID based security system 40

14 xi LIST OF FIGURES FIGURE TITLE PAGE 2.1 The example of magnetic door The example of smart card system Classification of biometric technology Flowchart of the project planning System overview Steps of face recognition process Working principle of PCA The example of Arduino Microcontroller The difference between uncontrolled and 27 controlled background 4.2 (a) Single image per person in training database (b) Five images per person in training database The variation in head position The example of negative recognition The example of positive recognition Relationship between the accuracy rate of the 34 system and the condition of parameter 6.1 Project schedule 38

15 xii LIST OF ABBREVIATIONS AND SYMBOLS - The mean of whole data in vector form - The image data after the mean is removed C - Covariance matrix Ω - Vector describing the face class DLM - Dynamic Link Matching EBGM - Elastic Bunch Graph Matching ED - Euclidean Distance GUI - Graphical User Interface ID - Identification LDA - Linear Discriminant Analysis LED - Light Emitting Diode MATLAB - Math Laboratory PCA - Principal Component Analysis RFID - Radio Frequency Identification TTL - Transistor Transistor Logic UART - Universal Asynchronous Receiver Transmitter USB - Universal Serial Bus W - Matrix composed of column vectors

16 xiii LIST OF APPENDICES APPENDIX TITLE PAGE A MATLAB source code 43 B Schematic diagram of circuit connection. 50

17 1 CHAPTER 1 INTRODUCTION 1.1 Background of Project In recent years, it is very important to have a reliable security system that can secure our assets as well as to protect our privacy. The conventional security system requires a person to use a key, identification (ID) card or password to access an area such as home and office. However, the existing security system has several weaknesses where it can be easily forged and stolen. Thus, the problems lead to an increased in interest in biometric technology to provide a higher degree of security to replace the conventional security system. Face recognition is one of the most popular authentication methods in biometric technology. It is the most natural means of biometric verification compared to the other biometric verification, such as fingerprint, iris and voice verification. Apart from that, the facial imaging has easy client acceptance and makes face recognition universal. The input of the system is a face image captured by the integrated webcam. Once the face image is captured, it is then transferred into MATLAB programming for the face recognition process. Next, the input face image will be compared to the training image in the training database. The recognition process is done by using Principal Component Analysis (PCA) algorithm.

18 2 The Arduino UNO microcontroller is used in the system to control the output of the system which is magnetic door. If the system recognizes the subject as an authorized person, the signal will be sent to the Arduino UNO microcontroller and used to unlock the magnetic door. Conversely, if the subject is unrecognized, the magnetic door will remain locked. Face recognition is not restricted to the application of access control. It has a broad range of potential application. The application is mainly in the fields of biometric such as intruder detection, surveillance system and law enforcement. 1.2 Problem Statement Security is an ultimate concern in our daily life. One of the most important fields in security system is the access control which controls the entranceways of a building or an area such as home and office. The security for access control is very important as a high number of burglary cases were reported every year. Based on statistics reported by the Royal Malaysia Police, 11,586 burglary cases were reported from January 2013 until June 2013 [1]. The huge amount of burglary cases lead to a huge amount of losses faced by the victims. The huge amount of losses emphasized that the security system should not be taken lightly. The conventional security system for access control is not reliable since it can be forged and stolen. For example, the password and ID card can be easily breached since the password can be divulged to an unauthorized user, and the ID card can be stolen by an impostor [2]. Apart from that, the traditional security methods like keys and ID card can be easily lost and misplaced. Therefore, security system for access control should be modernized to enhance the security purpose. A more reliable security system should be developed to avoid greater loss. Biometric technology can be implemented in the security system for access control as it offers a higher degree of security compared to the

19 3 conventional security system. From [3], biometrics is the most secure and convenient authentication tool since it cannot be borrowed, stolen, or forgotten, and forging one is practically impossible. The system is developed using the PCA algorithm for the recognition process. The PCA algorithm is selected as the main algorithm in face recognition process due to the simplicity of realization and the speed of recognition with respect to the other method. Besides, the efficiency of the system is increased as PCA reduced the data dimension and operate in smaller dimension. 1.3 Project Objective The objective of this project is to design a security system for access control using a face recognition system. The specific objectives that have to be achieved are as follows: i. To develop a security system based on face recognition. ii. To design a door access security system based on face recognition. iii. To design a face recognition security system using the PCA algorithm. 1.4 Project Scope This project is divided into three main parts. The first part focus on the research and literature review of the project. In this part, the theory, algorithm and application of a security system based on face recognition was studied. A small research on the previous face recognition project has been made in order to get an idea of the working principle of the system.

20 4 The second part is based on the software programming. MATLAB has been used to develop the database to implement face recognition system. Face recognition system is developed in MATLAB using Computer Vision System Toolbox and Image Processing Toolbox. PCA algorithm has been used as the main algorithm in the process of face recognition. The last part is based on the hardware development. The output or application of this project is the magnetic door which is controlled by the Arduino UNO microcontroller. The output depends on the result from recognition process. A comparison is made between the input image and the training image in the training database during the recognition process. The magnetic door will unlock if the system recognizes the input image and vice versa. 1.5 Significance of Project The main advantage of this project is a higher degree of security system for access control system can be developed. The problem encounter with conventional existing security system such as stolen of ID card and keys can be solved. By implementing biometric technology based on face recognition system for access control, the losses due to burglary cases can be reduced. 1.6 Thesis Outline The first chapter is on the introduction of the project. A brief overview regarding on the project has been discussed. It consists of the problem statements, the objectives of the project, the scopes of the project, as well as the significance of the project.

21 5 Chapter 2 consists of a literature review and a brief explanation of the research. A comparison on the method and algorithm from various journals and other references were made to choose the most accurate method in order to be able to reach the objectives. This chapter also provides an overview of the algorithm and components used in this project. Chapter 3 discussed about the methodology and the approach used in the project. It defines and illustrates the steps involved in developing the system. The steps taken in software development as well as the hardware development are discussed in details in this chapter. In Chapter 4, it includes the results of the analysis that has been done. The analysis is done to determine the accuracy rate of the system. The analysis is done by adding challenging details such as background parameter, multi train image and variation in head position. Chapter 5 is for the discussion and the conclusion of the project. The conclusion is made based on the objective of the project. Future recommendations were also discussed for further improvement of the system. Project management is discussed in chapter 6. Project details and cost estimation are discussed in details. A simple comparison on the cost estimation between the system that are based on face recognition and a system that are based on RFID technology is done.

22 6 CHAPTER 2 LITERATURE REVIEW 2.1 Door Access Control System Access control system is a restriction of access to property or buildings. It is important to have a security system for access control in order to secure our assets and privacy. Magnetic door can be used to replace the traditional door knob to increase the level of security. Figure 2.1: The example of magnetic door The magnetic door lock which works according to the concept of electromagnetism is composed of an electromagnet and armature plate. Typically the electromagnet portion of the lock is attached to the door frame and a mating armature plate is attached to the door to allow the two mechanisms to work efficiently together. When the door is closed, the two components are in contact.

23 7 The design takes into account the concept where the current has allowed to flow through the wire for the production of magnetic flux or magnetic power. The door will keep locked as the produced power provides a necessary strength that keeps the door from being opened. The magnetic flux causes the armature plate to attract to the electromagnet. Thus, the locking action is created and securely locks the door. The operation methods of magnetic door can be divided into three basic operations. The first method involves the use of a keypad system such as passwords. The system will lock and unlock with the numeric code. A smart card such as Radio Frequency Identification (RFID) tag is used in the second operation method of magnetic door which is typically used for business and commercial buildings. For the last operation method, the magnetic door can be operated using biometrics technologies such as thumb print and face recognition Password/ Keypad System The most common form of system identification and authorization mechanism is a password or keypad system. For higher assurance, the user needs to change passwords frequently so that it cannot be guessed. The user can choose their password in the consideration of its practicality. The password should be generated properly and keep it as secret. This system can be considered as one of the weak security system as the password can be easily forgotten or it can be hacked easily. Thus, password or keypad based system is not really reliable for access control Smart Card System Smart card allows the card owner to access the facility. A smart card can be programmed to allow or deny access through specified doors or facility. It stores protected information and the person s privileges. There are two types of smart card

24 8 either contact or contactless. The contactless smart card usually used the electronic signal to transfer data while physical contact is used for communication in contact based card. However, it has several weaknesses as the card can be easily lost, stolen or damage if it is exposed to high electromagnetic field. Figure 2.2: The example of smart card system Biometric Control System Biometric control system which is used for personal authentication utilizes a person s features such as the fingerprint, voice and face which are unique for each individual. The system identifies and verifies a person by their unique characteristics. It offers a higher degree of security since the uses of a person s unique characteristics make it difficult to impersonate by a stranger. There are various types of biometric that can be used for access control system such as thumbprint, voice recognition and face recognition. Each method can be categorized in term of usability and the level of security as each method has their own particular characteristics.

25 9 2.2 Biometric Technology Biometric technology, which is widely used in the application of security, is a method of identification and verification based on human characteristics. Figure 2.3 shows the classification of biometric. Biometrics can be divided into two categories which are physiological and behavioral characteristics. Common physiological biometrics include face, fingerprint, hand, iris and DNA while behavioral characteristic depends on the pattern of behavior such as voice, keystroke pattern and signature. Figure 2.3: Classification of biometric technology Since each person has their own unique characteristics, security system based on biometric technology is highly recommended. In [3], it is stated that biometrics are the most secure and convenient authentication tool as it cannot be borrowed, stolen or forgotten. Besides, by using biometric authentication method, it is easy to check whether a person has several identities or not. In Table 2.1 below shows a comparison of various biometric technologies, according to the perception of [4]. The comparison is based on seven factors which are the universality, distinctiveness, permanence, collectability, performance, acceptability and circumvention. The rank is basically divided into three categories

26 Universality Distinctiveness Permanence Collectability Performance Acceptability Circumvention 10 as being low (L), medium (M) or high (H). A higher ranking indicates a good performance while a lower ranking indicates a poor performance. Table 2.1: Comparison of various biometric technologies [4] Identifier Face H L M H L H H Fingerprint M H H M H M M Signature L L L H L H H Voice M L L M L H H The requirements of the application domains will determine the applicability of a specific biometric technique. For an access control security system, the system should have a higher degree in universality, distinctiveness, permanence, and performance. The universality of an identifier means that every person should have the characteristics while the distinctiveness depends on the differences in characteristics to distinguish each person. Meanwhile, the permanence factor can be described as a stability of characteristics in a different time and environmental conditions and the performance of an identifier is basically depends on the identification accuracy and recognition time [5]. The face recognition method has a higher degree in universality and moderate degree in permanence factor. Although for the distinctiveness and performance factors of face recognition is relatively low, face recognition has higher usability and security where less than % failure to enroll and acquire rate [6]. Thus, based on this point of view, face recognition method was selected in developing the security system for access control.

27 Face Recognition Face recognition system is a biometric identification and verification that use a person s face as their corresponding input. The face recognition system is similar to other biometric system such as fingerprint or voice recognition as one s face has many unique structures and features. Face recognition is selected for the system based on the considerations that facial imaging, being non-intrusive, has easy client acceptance, apart from the fact that face recognition is the most natural means of biometric identification of human beings [7]. The face recognition process consists of four main steps which are image acquisition, preprocessing, features extraction and classification. Feature extraction is the key step of any face recognition process. Feature extraction extracts the feature vector and information which represent the face in the face image. Thus, feature extraction is the most important step in face recognition. The selection of features that represent the face image is done in feature extraction step. There are two basic methods in feature selection of face recognition as discussed in [8]. The first method used global features approach or holistic approach while the second method used local features approach. Global features use the pixel level information from the input face image as the main features while local features use geometric relationship among features such as the distance between two eyes and the size of the eyes itself [9]. 2.3 Feature Extraction Algorithm in Face Recognition There are numerous algorithms develop for feature extraction in face recognition system. An algorithm for feature extraction is basically based on feature selection method. Two typical global features algorithms are PCA and Linear

28 12 Discriminant Analysis (LDA) while Elastic Bunch Graph Matching (EBGM) and Dynamic Link Matching (DLM) is based on local features method [10]. PCA is an overall statistical method on the face image where the objective is to retain the most information about the original image by finding the set of most representative projection vectors [11]. Meanwhile, LDA aims to find the subspace that best discriminates different face classes between intra-personal classes and extra-personal classes. Both PCA and LDA algorithms contribute in the reduction of data Principal Component Analysis (PCA) The Principal Component Analysis (PCA) is a method of projection to a subspace and is widely used in pattern recognition [12]. PCA is used to re-express the original data in lower dimension basis vector [13]. Therefore, the noise and redundancy of the data are kept to a minimum and the data is described economically. Pattern recognition based on the Karhunen-Loeve expansion, Kirby and Sirovich [14, 15] have shown that any particular face can be represented in terms of a best coordinate system termed as eigenfaces. In face recognition,. PCA is used to calculate the eigenfaces and find the vectors that best accounts for the distribution of face images within the entire image space [16]. Typically, two phases are included in the PCA algorithm which are the training phase and the classification phase. In the training phase, the eigenspace was established from training samples and the training images are mapped to eigenspace for classification. During the classification phase, an appropriate classifier is used to classify when input is projected to the same eigenspace.

29 Linear Discriminant Analysis (LDA) Linear Discriminant Analysis (LDA) is an algorithm used for decreasing data dimension. The objective of LDA is to preserve as much of the class discriminating information as possible. It is appearance-based technique which encodes discriminatory information in a linear separable space which bases are not necessarily orthogonal [17]. Rather than searching for information that best describe the data, LDA searches for vectors in underlying space that best discriminate among classes. In other words, LDA generates a linear combination of independent features that yields the largest mean differences between the desired classes [13]. There are two class scatter matrix in LDA. The image within-class scatter matrix or known as intrapersonal class and the image between-class scatter matrix. Within-class scatter matrix or intra-personal class can be described as different in variations of expression based on same individual while between-class scatter matrix or extra-personal class can be described as differences in variations of appearance due to different identity. The ratio of the data distance between intrapersonal classes must keep as small as possible. Conversely, the ratio of the data distance between extra-personal classes must be maximized. Thus, at least two training samples are required to calculate between-class scatter matrix, and withinclass scatter matrix [16]. In [18], it is stated that PCA can outperform LDA if the training data set is small. PCA is a more suitable algorithm for access control since PCA gives faster response of recognition compared to LDA. Besides, PCA also has the advantage of simple computerization.

30 Conclusion As a conclusion, after reviewing several thesis and paper work from previous research, it can be concluded that biometrics based security system offers a higher degree of security compared to the conventional security system. Face recognition has an advantages of easy client acceptance and universality. For features extraction algorithm in face recognition process, PCA is chosen due to its fast response in recognition process as well as the low sensitivity to noise. Therefore, the system will be developed based on face recognition method by using the PCA algorithm.

31 15 CHAPTER 3 METHODOLOGY 3.1 Introduction The aim of this project is to develop a security system for access control based on face recognition. This includes the study of development on biometric technology, face recognition system and also the magnetic door theory. The development of this project consists of three main parts. The first part is basically on literature review followed by the software and hardware development. 3.2 Project Planning As referred to Figure 3.1, the project development is started on literature review where a small research on the theory, applications and algorithm on anything that related to the face recognition system for security purpose had been done. It is an important part in order to collect the information from published materials and resources. The next part focuses on the software development where the face recognition takes place in MATLAB programming followed by hardware development of the project. The application of the system is the access control, thus a magnetic door system is used as the output for the project.

32 16 Start Literature review Security system for access control Face recognition system Development of database for face recognition using PCA Hardware development for magnetic door using Arduino Devices analysis based on recognition rate Acceptable? No Design improvement Yes Stop Figure 3.1: Flowchart of the project planning

33 Project Overview Figure 3.2: System overview Figure 3.2 shows an overview of the system. The input image which is captured using integrated webcam is transferred to MATLAB programming for the recognition process. The data from recognition process is used as the input for the Arduino UNO microcontroller. If the system recognized the person or claim that the person as an authorized person, the magnetic door will unlock. In contrast, if the system did not recognize the person, the magnetic door will remain locked. 3.4 Software Development In software development, the face recognition process is done in MATLAB. The system is developed using Computer Vision Toolbox, Statistical and Image Acquisition Toolbox. A Graphical User Interface (GUI) is developed to provide an interface for interaction of electronics devices through graphical icons.

34 18 Figure 3.3 shows the main steps of face recognition process. The face recognition process goes under image acquisition, preprocessing, feature extraction, and classification stages. Image Acquisition Pre Processing Feature Extraction Classification Figure 3.3: Steps of face recognition process Image Acquisition First, in the image acquisition process, the input face image is captured via integrated webcam. Once the input image is captured, the features information will be extracted. The purpose of image acquisition is to seek and extract a region which contains only the face Preprocessing In preprocessing, the acquired image is resized to a specific size and resolution. The image is resized to 180x200 pixels. Dimensionally reduction is done by compressing the original features without destroying the important information from the image. Noises are removed by filtering techniques.

35 Feature Extraction using PCA Algorithm This system used global features approached for feature selection. Global features approach weights each pixel equally regardless it is the face pixel or the background pixel. This approach will encode the entire face and represent face as a code point in higher dimensional image space [9]. Feature extraction algorithm extracts features of the data and creates new features based on the transformation or a combination of the original data. In this system, PCA algorithm is implemented for the feature extraction algorithm. The PCA algorithm used eigenfaces to find a vector that best distribution of face images. PCA is used for feature extraction and data reductions. The basic working principle of PCA is to extract the characteristic of feature on the face and represent the face in the linear combination. The principle component of face in the training set is then calculated. The principle component or eigenfaces are a set of eigenvector associated with a particular eigenvalue. The vector called as eigenfaces are eigenvector of the covariance matrix is corresponding to the original face images. PCA aims to find the vector that best accounts for the distribution of face images within the entire image space. PCA algorithm is used to calculate the eigenfaces of the image. It includes the calculation of the mean image in face space and the each face difference is further computed from the mean. The difference is used to compute a covariance matrix to reveal how much sets are correlated.

36 20 [12]. Figure 3.4 shows the working principle of the PCA algorithm as discussed in Figure 3.4: Working principle of PCA [12]. In mathematical terms, the 2D facial image is converted from 2D array to 1D dimensional vector. The image is centered by subtracting the mean image from image vector.

37 21 (3.1) (3.2) From Equation 3.1 and Equation 3.2 above, Ψ is the mean of whole data in vector form while is the image data after the mean is removed. Next, the image data are combined into data matrix and to create covariance matrix, by multiplying to its transpose. The equation to create a covariance matrix is shown in the Equation 3.3. (3.3) In Equation 3.3, W is a matrix composed of column vectors. The eigenvector of the covariance matrix is calculated. The greatest variance in the image is represented by the eigenvector with the largest eigenvalue. The image data are then projected into the eigenfaces space. For the distance measure, the Euclidean Distance between the image data and eigenfaces are calculated using Equation 3.4 where Ω is a vector describing the face class. An image with a minimum value of ED is recognized as the equivalent image. (3.4) To express the data noise and data redundancy quantitatively, a covariance matrix should be introduced where it includes all the correlation information on noise and redundancy of the data. The eigenvectors are derived from the covariance matrix. The eigenspace is calculated by identifying the eigenvectors of the covariance matrix derived from a set of facial images (vectors) [19].

38 22 The main advantages of the PCA are its low sensitivity to noise, the reduction of the requirements of the memory and the capacity, and the increase in the efficiency due to the operation in a space of smaller dimensions [12]. Even though the PCA algorithm is sensitive to changes in illumination and facial expression, however PCA algorithm has an advantage on real-time recognition Classification The system used the Euclidean Distance (ED) method as the classifier. ED is the nearest mean classifier which is commonly used for decision making rules. The ED is obtained by calculating the distance between the test image and the training image in the database. A minimum ED must be obtained in order to recognize the expression of the input image. The maximum value of ED between the test image and the training image is set to 4.00 x Thus, if the ED between two images is smaller than 4.00ex10 15, the system will recognize the identity of the person and gives an entry permission which means the magnetic door will unlock. However, if the ED is higher than the value that we have set earlier, the identity of the person will not recognize by the system and the magnetic door will remain locked. Refer to Appendix A for the source code of the project in software development. 3.5 Hardware Development using Arduino UNO Microcontroller For hardware development, it will focus on the microcontroller used in the system. The microcontroller used in this project is an Arduino UNO microcontroller. The Arduino UNO microcontroller was used to control the lock and unlocking process of magnetic door depends on the output of face recognition phase.

39 23 Arduino UNO mirocontroller is an open source hardware board that contains everything to support the microcontroller such as 14 digital input/output pins, a Universal Serial Bus (USB) connection, a power jack and reset button. Arduino UNO microcontroller is based on ATmega328 and can be simply connected to a computer with a USB cable provided [20]. The Arduino UNO microcontroller can be powered up either by using USB connection or external power supply. To communicate with the computer, the ATmega328 provides Universal Asynchronous Receiver Transmitter (UART) using transistor-transistor logic circuit (TTL) (5V) serial communication which is available on digital pins 0 and pins 1. The Arduino UNO microcontroller provides 14 digital input/output pins which operate at 5 volts. Basically, the digital pins default to inputs or the pins configured to be in high impedance state. Each pin has an internal pull-up resistor where it is disconnected by default of k Ohms. Figure 3.5: The example of Arduino UNO microcontroller A simple circuit is designed for the output of the system. Electrical components such as relay, transistor, resistor and Light Emitting Diode (LED) are used in the circuit. Refer to Appendix B for the schematic diagram of the circuit connection.

40 Conclusion In a nutshell, PCA algorithm is used in feature extraction for face recognition as it is not depending on facial expression, faster in computational time as well as lower sensitivity to noise. Meanwhile, for the microcontroller, Arduino Uno microcontroller is selected as it is easy to program, and quite cheap.

41 25 CHAPTER 4 RESULT AND DISCUSSION 4.1 Introduction The objective of the analysis is to determine the rate of accuracy of the system. The rate of accuracy can be defined as the ratio between the number of successful recognition and the total number of recognition, which derives from the comparison of the distance between two images [13]. The accuracy rate of the system can be calculated from Equation 4.1 below. (4.1) The Euclidean Distance is set to 4.00x10 15 as the highest value of distance for the system to recognize the image. If two images of the same person have the smallest distance, it is assumed as a successful recognition. The total number of the recognition in this analysis has been fixed at 20 times. Based on previous works, there are three main parameters that have been considered in order to determine the rate of accuracy of the system. Three main parameters that were analyzed are the background of the face image, the effect of using multi train image in training database and the variation in head position. The parameters were varied between uncontrolled condition and controlled conditions to

42 26 determine the effect on the accuracy rate of the system. The analysis is done using an offline approach where the face image is used from available digital images. Initially, the system is set up without controlling any parameter. The result of the initial analysis is used as a reference. The objective is to determine the accuracy rate of the system at the most uncontrolled condition. The result of the analysis is tabulated in Table 4.1 and will be used as a reference. Table 4.1: The result of uncontrolled parameters Test Image Euclidean Distance Status x10 15 Unmatched x10 14 Matched x10 15 Unmatched x10 15 Unmatched x10 15 Matched x10 15 Matched x10 15 Unmatched x10 14 Matched x10 15 Unmatched x10 15 Unmatched x10 15 Unmatched x10 13 Matched x10 15 Unmatched x10 15 Unmatched x10 14 Matched x10 14 Matched x10 15 Unmatched x10 15 Unmatched x10 14 Matched x10 15 Unmatched

43 27 From the analysis, the system gives 40% of accuracy. The accuracy rate of the system with uncontrolled parameters is not suitable for the application of access control. The application of access control needs high rate of accuracy as the system need to differentiate between authorized and unauthorized person to grant the access. Thus, further analysis should be made to improve the accuracy rate of the system Background parameters For further analysis, the background parameter is changed from uncontrolled conditions to controlled conditions. The number of training images in train database is set for a single image per person and the variation in head position is set in uncontrolled conditions. For uncontrolled background, the face image is taken with some noises in the background, such as more than one face appeared in the image while to set the background in controlled conditions, the image is taken in dark and homogeneous background. The difference between uncontrolled and controlled background is shown in Table 4.2. Figure 4.1: The Difference between uncontrolled and controlled background. Table 4.2: The results of controlled background parameter Test Image Euclidean Distance Status x10 14 Matched x10 14 Matched

44 x10 15 Unmatched x10 15 Unmatched x10 15 Unmatched x10 14 Matched x10 15 Unmatched x10 15 Unmatched x10 15 Matched x10 14 Matched x10 15 Matched x10 16 Unmatched x10 15 Unmatched x10 15 Matched x10 14 Matched x10 16 Unmatched x10 16 Unmatched x10 14 Matched x10 14 Matched x10 13 Matched With a controlled background parameter, the system has successfully recognized 11 identities. It is shown that with controlled background parameters, the accuracy rate of the system is increased from 40% to 55%.The accuracy rate of the system is increased with controlled background parameter as the noise in the background is reduced. Based on the result, we can conclude that the accuracy rate of the system is higher with controlled parameter compared to the uncontrolled parameter. 4.3 Multi Train Image Next, the number of training image per person is increased from a single image per person to 5 images per person in order to increase the accuracy rate of the

45 29 system. However, the parameter of variation in head position is kept in uncontrolled condition. Figure 4.2 shows the difference in the number of training image per person and Table 4.3 shows the results of the analysis on multi train image per person in the training database with controlled background parameter. Figure 4.2 (a): Single image per person in training database Figure 4.2 (b): Five images per person in training database Table 4.3: The result of using multi train image in training database Test Image Euclidean Distance Status x10 15 Unmatched

46 x10 13 Matched x10 13 Matched x10 13 Matched x10 14 Matched x10 14 Matched x10 13 Matched x10 14 Matched x10 16 Unmatched x10 14 Matched x10 13 Matched x10 13 Matched x10 13 Matched x10 15 Unmatched x10 15 Unmatched x10 12 Matched x10 13 Matched x10 13 Matched x10 13 Matched x10 14 Unmatched The system shows an increased in the rate of accuracy from 55% to 75%. Thus, it is proven that the rate of accuracy of the system is proportional to the number of training image per person in train database. The accuracy rate of the system increased as the number of training image increased. Higher number of training image provides more basis vector in the feature space and produce less reconstruction errors. 4.4 Variations in Head Position In order to increase the accuracy rate, the variation in position of the subject should be minimized. For uncontrolled position, the head position of the subjects is

47 31 varied while in controlled conditions, the position of head of subjects is standardized in an upright, frontal position with some tolerance of side movement. Figure 4.3 shows the uncontrolled and controlled position of the subject s head and the result of the analysis is tabulated in Table 4.4. Figure 4.3: The variation in head position Table 4.4: The result of the variation in head position Test Image Euclidean Distance Status x10 13 Matched x10 13 Matched x10 13 Matched x10 14 Matched x10 14 Matched x10 15 Unmatched x10 13 Matched x10 13 Matched x10 13 Matched x10 13 Matched x10 13 Matched x10 13 Matched x10 12 Matched x10 14 Matched x10 15 Unmatched x10 13 Matched

48 x10 13 Matched x10 13 Matched x10 13 Matched x10 13 Matched The accuracy rate of the system improved from 75% to 90% based on the analysis. With a minor tolerance of side movement of the head position, the system successfully recognized 18 identities. It shows that the head position of the subject is one of the challenges in face recognition. The example of negative recognition and positive recognition is shown in Figure 4.4 and Figure 4.5. After recognition process, the data will be sent to Arduino Uno microcontroller. Arduino Uno microcontroller will decide whether to unlock or remain locked based on the recognition data. Figure 4.4: The example of negative recognition Figure 4.5: The example of positive recognition

49 Conclusion For the conclusion, the system gives the lowest accuracy rate when all the three parameters are in uncontrolled condition. By controlling the background parameters, the accuracy rate increased from 40% to 55%. Besides, the rate of accuracy of the system with the use of multi train image. Lastly, by reducing the variation in head position also increased the accuracy rate of the system. The system gives the highest accuracy which is 90% with all the parameters in controlled conditions. The summary of the analysis is tabulated in Table 4.5. Table 4.5: The summary of the analysis Analysis Parameter Accuracy Rate Background Multi Train Head Position (%) Image 1 Uncontrolled Uncontrolled Uncontrolled 40 2 Controlled Uncontrolled Uncontrolled 55 3 Controlled Controlled Uncontrolled 75 4 Controlled Controlled Controlled 90 Based on Table 4.5, the graph of the relationship between the accuracy rate of the system and the condition of the parameters is plotted. The graph of the relationship between the accuracy rate of the system and the condition of the parameters is plotted in Figure 4.6.

50 Accuracy Rate of The System Accuracy Rate Figure 4.6: Relationship between the accuracy rate of the system and the condition of parameter

51 35 CHAPTER 5 CONCLUSION 5.1 Conclusion The analysis shows that the system gives a higher accuracy rate with controlled background parameter compared to the uncontrolled background parameter. Besides, the accuracy rate of the system is directly proportional to the number of training image and as the variation in head position decrease, the accuracy rate increase. Therefore, in order to get maximum accuracy rate of the system, all parameters should be set in controlled conditions. The system gives the highest accuracy rate with all parameters is in controlled condition which is 90%. As a conclusion, PCA is a reliable algorithm to be used in face recognition security system if the parameters are set in controlled conditions. The objective in developing a face recognition security system for access control using PCA algorithm is achieved. 5.2 Future Work There are several problems occurs during the development of the system. The main problem is the serial connection between MATLAB and Arduino UNO microcontroller. Based on previous reading, the communication error is a common

52 36 problem when interfacing MATLAB and Arduino UNO microcontroller. The Arduino UNO microcontroller is used in this project due to its simplicity, easy to program and low cost controller. However, Arduino UNO microcontroller is not suitable for continuous system since it is not capable to run for a long time and only suits with a low capacity processor. Thus, it is recommended to use higher capacity of Arduino microcontroller such as Arduino Duemilanove and Arduino Mega. Next, to have a larger size of a database, an improvement of the algorithm should be made. A combination of algorithm such as the combination of PCA and LDA can be made to keep the accuracy rate high. A combination of algorithm also makes the system more user-friendly since the improved algorithm makes the system less sensitive to the challenging parameter.

53 37 CHAPTER 6 PROJECT MANAGEMENT 6.1 Introduction Project management is one of the important elements in order to achieve all project objectives and goals. Project management is the application of knowledge where it helps in project planning, organizing and controlling resources. It is important to ensure that the project is executed efficiently and effectively within the prescribed time with the best use of all resources. In this project, research scope, research time and research budget are the main constraints to meet the project requirement. Thus, the project schedule has been tabulated in a Gantt chart for a better guideline in time management as well as to track the project development. Next, project budgeting is a process of cost estimation that approximate the cost of the project. Cost estimation must be accurate as underestimation of cost and resources can lead to project failure. Therefore, cost estimation is done to ensure that the cost required is minimal without neglecting the project requirement. Therefore, a market survey is carried out and the final cost is tabulated.

54 September October November December January February March April May June Project Schedule Two semesters are given for the development of the project, including the analysis and testing of the system. A Gantt chart on the details of work contribution for the whole one year is shown in Figure 6.1 below: Activities Progress Evaluation Literature Review Software Development Hardware Development Model Building Analysis and Troubleshooting Thesis Write-up Thesis submission Figure 6.1: Project Schedule Figure 6.1 shows that the progress discussion and literature review are done throughout the two semesters. The literature review is very important as it justifies the proposed methodology and the progress discussion is important to track the project development so that it can be done within the prescribed time. Software development is done from October to March while hardware development is done from January to March. Next, it is followed by the building of the model, analysis and troubleshooting of the system, writing the thesis as well as the thesis submission.

55 Cost Estimation Table 6.1 shows the cost estimation for the hardware development of the system. As the system requires a simple circuit for the hardware, thus the system is developed within the budget given. Since only the prototype of the system is being developed, cheaper components are used in this project without neglecting the functions and criteria needed. Table 6.1: Cost estimation of face recognition security system Component Name Quantity Unit Price Total Arduino Uno-R3 1 RM58.00 RM mm Magnetic Door Lock 1 RM70.00 RM70.00 Single SPDT Relay SRD 1 RM2.50 RM2.50 LED Super Bright 5mm Green 1 RM0.20 RM0.20 LED Super Bright 5mm Red 1 RM0.20 RM0.20 Transistor 2N RM0.40 RM0.40 Resistor 0.25W 5% (22OR) 2 RM0.05 RM0.10 Resistor 0.25W 5% (1K) 1 RM0.05 RM0.05 Single Core Cable 2 RM0.40 RM0.80 Donut Board Small 6x15cm 1 RM1.60 RM1.60 TOTAL RM The cost estimation of the hardware development of the system is compared to the cost estimation in developing the security system for access control using another method. The cost estimation of security system for access control that are based on RFID is discussed. The comparison is important to determine whether the system developed is cost effective or not.

56 40 in Table 6.2. The cost estimation of a system that are based on RFID technology is shown Table 6.2: Cost estimation of RFID based security system Component Name Quantity Unit Price Total RFID Card 1 RM3.50 RM3.50 RFID Reader RS232 1 RM RM Arduino Uno-R3 1 RM58.00 RM mm Magnetic Door Lock 1 RM70.00 RM70.00 Single SPDT Relay SRD 1 RM2.50 RM2.50 LED Super Bright 5mm Green 1 RM0.20 RM0.20 LED Super Bright 5mm Red 1 RM0.20 RM0.20 Transistor 2N RM0.40 RM0.40 Resistor 0.25W 5% (22OR) 2 RM0.05 RM0.10 Resistor 0.25W 5% (1K) 1 RM0.05 RM0.05 Single Core Cable 2 RM0.40 RM0.80 Donut Board Small 6x15cm 1 RM1.60 RM1.60 TOTAL RM Economic factors are an important consideration when choosing among potential projects to be developed. From the result of the comparison, it is shown that the system developed require a lower cost compared to the system that are based on RFID technology. Therefore, a low cost security system for access control is successfully developed using a biometric technology based on face recognition method.

57 41 REFERENCES 1. Official Portal of Royal Malaysia Police, Statistik Jenayah Pecah Rumah Jan- Jun 2013, Retrieved November 20, 2013 from 2. Nazeer, S.A., Omar, N., and Khalid, M. (2007). Face Recognition System using Artificial Neural Networks Approach. International Conference on Signal Processing, Communications and Networking February ICSCN , 3. Liu, S., and Silverman, M. A Practical Guide to Biometric Security Technology, IT Professional, (2001). 3(1): Jain, A. K., Ross, A., and Prabhakar, S. An Introduction to Biometric Recognition. IEEE Transaction on Circuits and System for Video Technology. (2004) 14(1): Faundez-Zanuy, M., Biometric Security Technology. IEEE Transaction on Aerospace and Electronic System. (2006) 21(6): Committee of Hanvon. (n.d.). Face ID :FAQ. Retrieved October 27, 2013, from 7. Kar, S., Hiremath, S., Joshi, D.G., Chadda, V.K, and Bajpai, A. A Multi- Algorithmic Face Recognition System. International Conference on Advanced Computing and Communication. December 20-23, 2006 ADCOM Agarwal, M., Agarwal, H., Jain, N., and Kumar, M., Face Recognition Using Principle Component Analysis, Eigenface and Neural Network. International Conference on Signal Acquisition and Processing. February 9-10, ICSAP Riddhi, C., and Neha, P., Details Study On 2D Face Recognition Technique Using Local And Global Features. Indian Streams Research Journal (2):1-17.

58 Juwei, L., Plataniotis, K.N, and Venetsanopoulos, A.N. Regularization Studies of Linear Discriminant Analysis in Small Sample Size Scenarion with Application to Face Recognition. Pattern Recognition Letter (2): Xueguang, W., and Xiaowei, D. Study on Algorithm of Access Control System Based on Face Recognition. International Colloquium on Control and Management. August 8-9, ISECS Çarıkçı, M., and Özen, F., A Face Recognition System Based on Eigenfaces Method, Procedia Technology ; Chaoyang, Z., Zhaoxian, Z., Hua, S., and Fan, D. Comparison of Three Face Recognition Algorithms. International Conference on Systems and Informatics. May 19-20, ICSAI Kirby, M., and Sirovich, L. Application of The Karhunen-Loeve Procedure for The Characterization of Human Faces. IEEE Transaction on Pattern Analysis and Machine Intelligence (1): Sirovich, L. & Kirby, M. Low-Dimensional Procedure for the Characterization of Human Faces, Journal of the Optical Society of America A. (1987). 4(3): Lih-Heng, C., Aslleh, S.H., and Chee-Ming, T., PCA,LDA and Neural Network for Face Identification IEEE Conference on Industrial Electronics and Applications. May 25-27, ICIEA Eleyan, A., and Demirel, H. PCA and LDA based Neural Networks for Human Face Recognition. In: Delac, K., and Grgic, M. Face Recognition System. Austria: I-Tech Education and Publishing ;2007Martinez, Aleix M.; Kak, A.C., PCA versus LDA. IEEE Transactions on Pattern Analysis and Machine Intelligence (2): Martinez, Aleix M.; Kak, A.C., PCA versus LDA. IEEE Transactions on Pattern Analysis and Machine Intelligence (2): Kim K (2003) Face recognition using principal component analysis. Department of Computer Science, University of Maryland,College Park 20. Arduino, Arduino UNO Board, Retrieved Disember 21, 2013 from

59 43 APPENDIX A Source Code of Project MATLAB Programming source code Original version by Amir Hossein Omidvarnia, October [email protected]. % Real Time Face Recognition System for Access Control % Date : 10/4/2014 clear all; close all; clc; a=arduino('com19'); a.pinmode(10,'output'); %% LED for positive recognition a.pinmode(12,'output'); %% LED for negative recognition a.pinmode(8,'output'); %% magnetic door lock a.digitalwrite (10,0); a.digitalwrite (11,0); a.digitalwrite (12,0); a.digitalwrite (8,1); % HIGH (1) = lock, LOW(0)=unlock while (1==1) choice=menu('face RECOGNITION FOR ACCESS CONTROL',. 'Recognize from Webcam',... 'Exit');

60 44 if (choice == 1) % 'Recognize from webcam' option from main menu vid = videoinput('winvideo', 1, 'MJPG_320x240'); preview(vid); choice1=menu('capture Frame',... 'Capture & Compare ', %%% (a) 'Capture & Save ',... %%% (b) 'Exit '); %%% (c) Image = []; if (choice1 == 1) % 'Capture' option from (a) Image = getsnapshot(vid); try Image = imresize(image,[200,180]); testimage=fullfile('c:\users\missdidie\desktop\complete\test database\','1.jpg'); imwrite(image,testimage); % %%%%%%%%Paths for TrainDatabase and TestDatabase %%%%%%%%% TrainDatabasePath ='C:\Users\MissDidie\Desktop\complete\train database'; TestDatabasePath ='C:\Users\MissDidie\Desktop\complete\test database'; %%%%%%%%%%%%%%% recall and compare %%%%%%%%%%%%%%% TestImage = strcat(testdatabasepath,'\','1.jpg'); im = imread(testimage); %%%%%%%%%%%%%%% Create Database (1) %%%%%%%%%%%%%%% TrainFiles = dir(traindatabasepath); Train_Number = 0; for i = 1:size(TrainFiles,1)

61 45 if not(strcmp(trainfiles(i).name,'.') strcmp(trainfiles(i).name,'..') strcmp(trainfiles(i). name,'thumbs.db')) Train_Number = Train_Number + 1; % Nol images in the training database end end T = []; for i = 1 : Train_Number str = int2str(i); str = strcat('\',str,'.jpg'); str = strcat(traindatabasepath,str); img = imread(str); img = rgb2gray(img); [irow icol] = size(img); end temp = reshape(img',irow*icol,1);% Reshaping 2D -> 1D image T = [T temp]; % 'T' grows after each turn %%%%%%%%%%%%%%% Eigenface Core (2) %%%%%%%%%%%%%%%% m = mean(t,2); % Computing the average face image Train_Number = size(t,2); end A = []; for i = 1 : Train_Number temp = double(t(:,i)) - m; A = [A temp]; % Merging all centered images L = A'*A; % L is the surrogate of covariance matrix C=A*A'.

62 46 [V D] = eig(l); L_eig_vec = []; for i = 1 : size(v,2) if( D(i,i)>1 ) L_eig_vec = [L_eig_vec V(:,i)]; end end Eigenfaces = A * L_eig_vec; % A: centered image vectors %%%%%%%%%%%%%% Recognition (3) %%%%%%%%%%%%%% global face; face = 0; ProjectedImages = []; Train_Number = size(eigenfaces,2); for i = 1 : Train_Number temp = Eigenfaces'*A(:,i); % Projection of centered images ProjectedImages = [ProjectedImages temp]; end InputImage = imread(testimage); temp = InputImage(:,:,1); [irow icol] = size(temp); InImage = reshape(temp',irow*icol,1); Difference = double(inimage)-m; % Centered test image ProjectedTestImage = Eigenfaces'*Difference; % Test image feature vector Euc_dist = []; % ED for i = 1 : Train_Number

63 47 q = ProjectedImages(:,i); temp = ( norm( ProjectedTestImage - q ) )^2; Euc_dist = [Euc_dist temp]; end min(euc_dist) if(min(euc_dist) <= e+015) face = 1; % match a.digitalwrite (10,1); a.digitalwrite (11,1); a.digitalwrite (12,0); a.digitalwrite (8,0); pause(5); a.digitalwrite (8,1); end if (min(euc_dist) > e+015) end face = 2; % unmatch a.digitalwrite (10,0); a.digitalwrite (11,0); a.digitalwrite (12,1); a.digitalwrite (8,1); face [Euc_dist_min, Recognized_index] = min(euc_dist); OutputName = strcat(int2str(recognized_index),'.jpg'); %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%% SelectedImage = strcat(traindatabasepath,'\',outputname); SelectedImage = imread(selectedimage);

64 48 CaptureImage = strcat(testdatabasepath,'\','1.jpg'); ia = imread(captureimage); figure,imshow(ia); title('capture Image'); imshow(im); title('test Image'); figure,imshow(selectedimage); title('equivalent Image'); str = strcat('matched image is : ',OutputName); disp(str) end end closepreview(vid); if (choice1 == 2) %%%% 'Capture & Save' option from (a) Image1 = getsnapshot(vid); Image1 = imresize(image1,[200,180]); imshow(image1); title ('Save Image'); testimage1=fullfile('c:\users\missdidie\desktop\analysis\train database\','1.jpg'); imwrite(image1,testimage1); end closepreview(vid); if (choice1 == 3) %%% 'Exit' from main menu clear choice1

65 49 end end end if (choice == 2) clear choice return; end;

66 50 APPENDIX B Schematic Diagram of Circuit Connection