1 Pattern Recognition 32 (1999) Two novel characteristics in palmprint verification: datum point invariance and line feature matching Dapeng Zhang*, Wei Shu Department of Computing, Hong Kong Polytechnic University, Hung Hom, Kowloon, Hong Kong Received 9 December 1997; in revised form 16 July 1998 Abstract As the first attempt of automatic personal identification by palmprint, in this paper, two novel characteristics, datum point invariance and line feature matching, are presented in palmprint verification. The datum points of palmprint, which have the remarkable advantage of invariable location, are defined and their determination using the directional projection algorithm is developed. Then, line feature extraction and line matching are proposed to detect whether a couple of palmprints are from the same palm. Various palmprint images have been tested to illustrate the effectiveness of the palmprint verification with both characteristics Pattern Recognition Society. Published by Elsevier Science Ltd. All rights reserved. Keywords: Biometrics; Personal identification; Palmprint verification; Feature extraction; Line matching 1. Introduction Palmprint, as well as fingerprint which has been utilized as a positive human identifier for more than 100 years, is still considered as one of the most reliable means distinguishing a man from his fellows because of its stability and uniqueness. Although law enforcement agencies use human prints (both fingerprints and palmprints) routinely for criminal identification, human prints can provide an unequalled method of personal identification for various other areas such as access control for high security installations, credit card usage verification and employee identification [1 2]. Fingerprint identification is the most widespread application of biometric *Corresponding author. Tel.: (852) ; Fax: (852) ; technology. However, it is a difficult task to extract some small unique features (known as minutiae) from the fingers of elderly people as well as manual labourers [3, 4]. As a result, it is necessary to develop a new technique of automatic personal identification by palmprint. Palm is the inner-surface of the hand between the wrist and the fingers (see Fig. 1). There are usually three principal lines made by flexing the hand and wrist in the palm, which are named as heart line, head line, and life line, respectively . Two endpoints, a and b, are obtained by the principal lines (life line and heart line) which intersect both sides of the palm. Owing to the stability of the principal lines, the endpoints and their midpoint o remain unchanged day in and day out. Some significant properties can be defined in the following: (1) the locations of the endpoints and their midpoint are rotation invariant in a palmprint; (2) a sole two-dimensional right angle coordinate system can be established, of which the /99/$ see front matter 1999 Pattern Recognition Society. Published by Elsevier Science Ltd. All rights reserved. PII: S ( 9 8 )
2 692 D. Zhang, W. Shu / Pattern Recognition 32 (1999) features could be listed as follows: Fig. 1. Definitions of a palmprint: principal lines (1 heart line, 2 head line and 3 life line), regions (I finger-root region, II inside region and III outside region) and datum points (a, b-endpoint, o-their midpoint). Fig. 2. Geometry features and delta point features of a palmprint, where c d is the perpendicular bisector of segment a b and points 1 5 are delta points. origin is the midpoint o and the main axis passes through the endpoints; (3) a palm can be divided into three regions, including finger-root region (I), inside region (II) and outside region (III) by some connections between the endpoints and their perpendicular bisector; (4) the size of a palm can be uniquely estimated by both the Euclidean distance between two endpoints and the length of their perpendicular bisector in the palm (segment c d in Fig. 2). In this way, the pair of endpoints and their midpoint will act as the important datum points in the palmprint identification because of their invariable locations. In addition to the datum points, there still exists rich and useful information in a palmprint. Some kinds of Geometry features: According to the palm s shape, we can easily get the corresponding geometry features, such as width, length and area. Principal line features: Both location and form of principal lines in a palmprint are very important physiological characteristics to identify individual because they vary little from time to time. ¼rinkle features: In a palmprint, there are many wrinkles which are different from the principal lines in that they are thinner and more irregular. They are classified as coarse wrinkles and fine wrinkles so that more features in detail can be acquired. Delta point features: Delta point is defined as the center of a delta-like region in the palmprint. It is shown that there are always some delta points located in fingerroot region and outside region (e.g. 1 5 in Fig. 2). This makes it possible to obtain these features and well establish the stability and uniqueness. Minutiae features: A palmprint is basically composed of the ridges, hence the minutiae features can be used as a significant measurement. In general, geometry features, principal line features and wrinkle features can be derived from the image with low resolution, even low quality. Delta points can be located in the directional image and their features will be procured from a fine resolution palmprint image in case of the presence of noise in it [6, 7]. Nevertheless, minutiae features are detected only from a high quality and fine resolution image. Palmprint verification, which is to determine whether two palmprints are from the same palm, can apply the physical features mentioned above in principle. However, both delta point and minutiae features only can be obtained in fine resolution image. In addition, geometry features are easily captured so that a fake palm could be created . As a new biometric technology, we will adopt some significant features called line feature (such as principal lines and coarse wrinkles) in palmprint verification. Compared with fingerprint identification, which is a point matching process , this palmprint verification by using line feature matching is effective in the following respects: (1) it can be used on the image with low spatial resolution so that the size of a palmprint image can be reduced, even it is comparable to that of a fingerprint image; (2) significant features can be determined in case of the presence of noise because of feature extraction relied on low spatial resolution; (3) palmprint verification is a line matching process, which has several advantages over point matching since a line has more information than a point . In addition, the singular points, namely the core and the delta points, are used as points of registration for comparing minutiae in fingerprints . However, the effective method to detect these
3 structural features makes use of directional histograms in the neighborhood of these points [3, 6, 7], hence it will be difficult to provide accurate estimates of the singular points while fingerprint image is rotated and deformed. Therefore, the datum points of palmprint are more stable than those of fingerprint. The palmprint verification approach based on both datum point invariance and line feature matching is the important supplement to biometric technology. This will be discussed in the following sections. How to determine the datum points which act as the important registrations in palmprint verification will be demonstrated in the next section. Section 3 presents line feature extraction and matching. In Section 4, experimental results of the proposed approach are obtained. Section 5 contains the summary and discussion. D. Zhang, W. Shu / Pattern Recognition 32 (1999) Datum point determination The goal of datum point determination is to achieve the registrations in palmprint feature representation and matching. As a prime but important process in the palmprint verification, it is demanded as simple and effective as possible. The basic idea of datum point determination is to locate the endpoint of each principal line. According to the given regularity, the principal lines and their endpoints are accurately detected by using the directional projection algorithm Directional projection algorithm It is widely known that projection is a simple yet effective method of line segment detection along some orientation . Let F be an M N gray scale image and f (i, j) bethe gray level of pixel (i, j) (i"0, 1,2, M!1; j"0, 1,2, N!1). Without loss of generality, we consider that the projective angle α is measured clockwise from the i-axis and pixel (i, j ) is a pixel in F. Then, an x y right angle coordinate system can be established, of which pixel (i, j ) is the origin and the orientation of the x-axis is that of the projection (see in Fig. 3). In the x y coordinate system, a (2m#1) (2n#1) pixel in size subimage, F 1, can be obtained, and f(x, y) is the gray level of pixel (x, y) (x"!m,!m#1, 2, 0, 2, m!1, m; y"!n,!n#1, 2,0,2, n!1, n). As a result, the correspondence between this pair of coordinate systems is denoted as follows: i"i #cos(α#β) x #y, j"j #sin(α#β) x #y, (1) where β"tan (y/x). Fig. 3. The correspondence of two coordinate systems in directional projection algorithm. In F 1, obviously, the directional projection of this subimage is p(y)" f (x, y). (2) The smoothed set q(y) is calculated as q(y)" 1 p (y#k). (3) 2w#1 Then, pixel (0, y ), where y " k q(k)"max q(y) (4) is detected and the corresponding pixel in F is defined as pixel (i, j ), where i "i #y cos(α# ), j "j #y sin(α# ). (5) Here, this basic algorithm is classified as four forms by the different projective orientations: horizontal projection (α"0 ), projection with 45 (α"45 ), vertical projection (α"90 ) and projection with 135 (α"135 ) Properties for principal lines From the study of the principal lines, their properties can be obtained as follows: (1) each principal line meets the side of palm at approximate right angle when it flows out the palm; (2) the life line is located at the inside part of palm, which gradually inclines to the inside of the palm in parallel at the beginning; (3) most of the life line and head line flow out of the palm at the same point; (4) the endpoints are closer to fingers than wrist.
4 694 D. Zhang, W. Shu / Pattern Recognition 32 (1999) Fig. 4. Computation of the directional projection: (a) original image; (b) directional projection along principal line; and (c) directional projection after median filter. Based on the projective correspondence (see Fig. 4), it is clear that the pixel (x, y) calculated by the basic algorithm belongs to the principal line if the orientation of directional projection follows the principal line and the parameter w in Eq. (3) equals the half width of a principal line. However, the pixel on the other different kinds of lines cannot be determined by means of the conditions given above. This is because the projection of line which is not in parallel with the projective orientation or is shorter than the principal line would be less than that of principal line. In addition, a thin line might map to the maximum value in the directional projection, p(y), but it could be reduced after smoothing. So, the basic algorithm can be used to locate the pixel only on the principal line while the directional projection along the principal line is maximized in a palmprint subimage Datum point determination To apply the above rules to get the endpoints, two processing stages, i.e. principal line detection stage and principal line tracing stage, should be defined. Without loss of generality, the images of a person s right palm are employed and all palmprints are considered as fingers upturned so as to describe the method of datum point determination simply. The track of the datum point determination is shown in Fig Principal line detection The purpose of this stage is to obtain the points, those belong to heart line or life line. The heart line detection is easier than that of life line because there is only one principal line in the outside part of palm. After the edges of outside and topside are detected, a pixel which has the suitable offsets to two edges is adopted in palmprint. Point b, which belongs to heart line, is located by the horizontal projection algorithm. However, the detection of life line cannot use the horizontal projection in the inside part of palm because the head line is determined sometimes instead of the life line while their flowing out palm at the different points. Therefore, another subimage Fig. 5. Track of datum point determination by using the directional projection algorithm. close to wrist is processed by the vertical projection algorithm and point a on the life line is calculated Principal line tracing The main idea is to estimate a pair of endpoints by tracing principal lines. According to the peculiarities of the heart line, the horizontal projection algorithm is easy to locate endpoint b to a subimage where there is a pixel situated at the outside edge with the same level as pixel b. However, the detection of endpoint a is different from that of endpoint b because the life line is a curve from pixel a to point a. As a result, the details of point a determination are shown below: Step 1: Point a is defined by the straight line, which is a connection between endpoint b and b, intersecting the inside edge of the palm. Step 2: Pixels a and a, which trisect the vertical distance between pixel a and point a, are obtained in proper order along the life line by the vertical projection algorithm. Step 3: Pixel a belonging to the life line is denoted. The horizontal distance from pixel a to the inside edge of the palm is calculated and a new point can be located, from which both horizontal and vertical distances are half of above distance to pixel a. A subimage around this new point is processed by the directional projection algorithm with 135. Step 4: Point a is detected by the straight line, which flows out from pixel a, intersecting the inside edge of the palm at Step 5: Endpoint a is estimated in the subimage with point a by the horizontal projection algorithm. Based on the endpoint a and b, it is simple to determine the midpoint o. For the left palmprint s datum point
5 D. Zhang, W. Shu / Pattern Recognition 32 (1999) detection, the similar steps are used except that the directional projection algorithm in Step 3 is replaced by that with 45. Obviously, the proposed method is also suitable for the palmprint under rotation while the outside edge of the palm is detected and rotated to approximate vertical. 3. Verification by line feature There are mainly two operations in the following verification: feature extraction and feature matching. Owing to line features adopted to identify the individual, line feature extraction and line matching are proposed. Obviously, line feature includes curves and straight lines. In fact, most line features can approximate straight-line segments except for the principal lines. Based on our observation, the curvature of the principal lines is small enough to be represented by several straight-line segments. In this way, the representation and matching of straight lines are simple. As a result, line feature is considered as straight-line segments in this section. The measurement of palmprints matching will be also described Line feature extraction Line feature extraction is always an important yet difficult step in image verification and many line feature detection methods have been proposed [10 12]. Most of these methods cannot generate a precise line map of stripe images such as palmprint images. Although the algorithm of pyramid edge detection based on stack filter performs well for these types of lines , it only extracts the principal lines from a palmprint image. In addition, a 3 3 operator, h, which convolves the image with! 1!! 1! (6)! 1! is commonly used as a detector for thin vertical lines and it can be extended to detect both lines in directions other than the vertical and thick lines . A straight line that passes through the center of a 3 3 neighborhood must intersect that neighborhood in one of the following 12 patterns (the center of the neighborhood has been underlined): 0 θ 0 θ θ θ 0 0 θ 0 0 θ 0 0 θ 0 θ 0 (a) (b) (c) (d) 0 θ θ θ 0 θ θ 0 0 θ 0 θ 0 0 θ (e) (f ) (7) 0 θ 0 θ 0 0 θ θ 0 θ θ θ θ θ 0 0 θ θ 0 0 (g) (h) (i) (j) θ θ θ 0 0 θ θ θ 0 0 θ θ θ 0 (k) (l) To obtain a complete set of thick-line operators, we would need an analogue of h for each of these patterns. For the five near-vertical patterns, then h operator would be defined by comparing the three θ points with their horizontal neighbors; for the five near-horizontal patterns, we would compare the points with their vertical neighbors; and for the two diagonal patterns, we could do either type of comparison. A similar set of operators could be used for streak detection, with the θ s representing adjacent nonoverlapping averages rather than individual points. Nevertheless, many ridges and fine wrinkles have the same width as coarse wrinkles except that they are shorter. As a result, the algorithm mentioned above is also difficult to acquire the line features from a palmprint because a mass of ridges and fine wrinkles could dirty the line features. We improve on this algorithm, which is called a improved template algorithm, so as to extract and post-process line segments of each orientation respectively and then combine them. The improved template algorithm consists of the following main steps: (1) Determine vertical line segments by using the five near-vertical templates, and then thin and post-process these line segments. The rule of post-processing is to clear up the short segments. (2) Similarly, detect lines in the other three directions. (3) Combine the results of four different directions, and then post-process once more. The purpose of postprocessing is to eliminate the segments overlapped. Both the process and result of line feature extraction by the improved template algorithm are shown in Fig Line matching In general, there are many ways to represent a line. One way, which is always possible, is to store the
6 696 D. Zhang, W. Shu / Pattern Recognition 32 (1999) endpoints of every straight line segment . In a twodimensional right angle coordinate system which is uniquely established by the datum points (see Fig. 2), line segments can be described by endpoints: (X (i), ½ (i)), (X (i), ½ (i)), i"1, 2, I, where I is the number of line segments. Without loss of generality, exchange the endpoints of each line segment so that X (i))x (i), i"1, 2, I. IfX (i)" X (i), exchange the endpoints so that ½ (i))½ (i). Next, three parameters of each line segment, including slope, intercept and angle of inclination, can be calculated as follows: slope(i)"(½ (i)!½ (i))/(x (i)!x (i)), (8) intercept(i)"½ (i)!x (i) slope(i) (9) and α(i)"tan (slope(i)). (10) The object of matching is to tell whether two line segments from a couple of palmprint images are the same in a palmprint. The two-dimensional right angle coordinate system established by the datum points does act as the important registration in line feature matching. For example, two line segments from two images can be represented as (X (i), ½ (i)), (X (i), ½ (i)) and (X ( j), ½ ( j)), (X ( j), ½ ( j)), respectively. The Euclidean distances between the endpoints of two line segments are denoted as " ((X (i)!x ( j)) #((½ (i)!½ ( j)), (11) " ((X (i)!x ( j)) #((½ (i)!½ ( j)). (12) Without question, the following conditions for line segment matching can be proposed: (1) If both and are less than some threshold D, then it clearly indicates that two line segments are the same. (2) If the difference of angle of inclination (difference between two line segments) is less than some threshold β and that of the intercepts is less than some threshold B, then it shows that two line segments have the equal angle of inclination and intercept. Within class of equal angle of inclination and equal intercept, if one of and is less than D, then two line segments clearly belong to the same one. (3) While two line segments overlap, they are regarded as one line segment if the midpoint of one line segment is between two endpoints of another line Verification function By applying the above rules of line feature matching to a couple of palmprint images, we can achieve the corresponding pairs of lines (shown as Fig. 7). The verification Fig. 6. Process and result of linear feature extraction: (a) input image; (b) map with linear features extracted; (c), (e), (g) and (i) line segments detection in four different directions by template algorithm, respectively; (d), (f ), (h) and (j) their results thinned and post-processed.
7 D. Zhang, W. Shu / Pattern Recognition 32 (1999) Fig. 6. (Continued.) function, r, is defined as r"2n/(n #N ), (13) where N is the number of these corresponding pairs; N and N are the numbers of the line segments determined from two palmprint images, respectively. In principle, it shows that two images are from one palm if r is more than some threshold ¹ which is between 0 and Experimental results In our experiments, gray-scale inked palmprint images are employed. The resolution is 100 dpi, 256 gray levels. The algorithm of datum point determination has been tested by many palmprint images. Of the datum points calculated in the 100 palmprint images, 95 are found to be in excellent agreement with the manual estimate. In
8 698 D. Zhang, W. Shu / Pattern Recognition 32 (1999) Fig. 6. (Continued.) particular, this algorithm is used to estimate the datum points with 60 special palmprint images, of which 20 belong to under rotation, 35 to incomplete and 5 to those life line and head line flow out the palm at different points. Some typical palmprint images are shown in Fig. 8 and the results of the datum points detection from special palmprint images are given in Table 1. Next, the palmprint verification with both datum point invariance and line feature matching also has been tested by 20 couples of palmprint images from 20 right palms. The measure of experimental result is represented by a slippery pair of statistics known as false rejection rate (FRR) and false acceptance rate (FAR). In this experiment, some thresholds are adopted as follows: (a) D"5, β"0.157 and B"10; (b) D"5, β"0.314 and
9 D. Zhang, W. Shu / Pattern Recognition 32 (1999) Fig. 7. Results of a pair of palmprint matching by using line features: (a) template linear feature set; (b) input linear feature set; and (c) matching result. B"10, or β"0.157 and B"50. The results are shown in Fig. 9 with the various threshold ¹. In Case (a), the palmprint verification can obtain an ideal result while ¹" Conclusion In this paper, we have implemented an automated biometric-based identification by using both datum points and line features extracted from the palmprints to verify the identity of a live person. As the primary palmprint identification, the datum points are defined as the points of registration in palmprint matching because of their stability. The basic algorithm using directional projection is designed to locate the principal line as well as to determine the endpoints accurately. A number of palmprint images have been tested and the experimental results show that most datum points are in excellent agreement with the manual
10 700 D. Zhang, W. Shu / Pattern Recognition 32 (1999) Fig. 8. Examples of datum points determination: (a) normal palmprint; (b) rotated palmprint; (c) incomplete palmprint; and (d) life line and head line unintersection. Table 1 Experimental results of datum points determination in special palmprint images Classification R I U Experiment images Accurate determination images The rate of accuracy (%) Note: R"rotated image; I" incomplete image; U"life line and head line unintersection. estimate. It is also reliable and effective to locate the datum points in the special palmprint images. The computation of the basic algorithm is simple and logical. An improved template algorithm of line feature extraction is then presented due to its simplicity. The line features are considered as straight line segments and are represented by their endpoints. Several rules are applied to match the line features so as to detect whether a couple of palmprints are from the same palm. Twenty couples of palmprint images are determined by both datum point
11 D. Zhang, W. Shu / Pattern Recognition 32 (1999) Fig. 9. Experimental results for FAR and FRR: (a) D"5, β"0.157 and B"10; (b) D"5, β"0.314 and B"10, or β"0.157 and B"50. invariance and line feature matching and the experimental result shows that the verification accuracy is acceptable. The palmprint verification proposed in this paper is also foolproof because these significant physiological characteristics are unique, unchanging, and cannot be forged and transferred. In addition, it has the remarkable advantage of requiring lower spatial resolution so as to limit the size of palmprint image and also to be insensitive to noise. As an important complement of automated biometric-based identification, the palmprint verification with both characteristics can be effectively used to identify an individual. 6. Summary Palmprint has been used as a positive human identifier for many years. As the first attempt of automatic personal identification by palmprint, in this paper, we have developed such a biometric technology to verify the identity of a live person by using both datum points and line features extracted from the palmprint. In Sect. 1, the properties of palmprints are introduced and some significant features in the palmprint are defined. Compared with fingerprint identification, two new characteristics, datum point invariance and line feature matching, are presented in palmprint verification. Sect. 2 mainly demonstrates how to determine the datum points which act as the important registrations in palmprint verification. The basic algorithm, including directional projection, is developed. The proposed algorithm is effective to locate the principal line as well as to accurately detect its endpoint based on the approximate position. The computation of the basic algorithm is simple and logical. Line feature extraction and matching are presented in Sect. 3. An improved template algorithm for line feature extraction is also designed due to its simplicity and efficiency. The line features are considered as straight-line segments and represented by their endpoints. Furthermore, several rules of line feature matching are applied to decide whether a couple of palmprints are from the same palm. Experimental results of the palmprint verification are described in Sect. 4. The algorithm of datum point determination has been tested on a number of palmprint images. It shows that most of datum points are in excellent agreement with the manual estimate and this algorithm is also effective to locate the datum points in the special palmprint images. Twenty couples of palmprint images have been determined by using both datum point invariance and line feature matching and the result shows the effectiveness of the palmprint verification. As summary and discussion in Sect. 5, in brief, the palmprint verification is foolproof because these significant physiological characteristics are unique, unchanging, and cannot be forged and transferred. This palmprint verification has the remarkable advantage of requiring lower spatial resolution, hence it can limit the size of palmprint image and also be insensitive to noise. As an important complement of automated biometric-based identification, the palmprint verification can be effectively used to identify an individual. References  M. Eleccion, Automatic fingerprint identification, IEEE Spectrum 10 (9) (1973)  B. Miller, Vital signs of identity, IEEE Spectrum 32 (2) (1994)  A. Jain, L. Hong, R. Bolle, On-line fingerprint verification, IEEE Trans. Pattern Anal. Mach. Intell., 19 (4) (1997)  L. Coetzee, E.C. Botha, Fingerprint recognition in low quality images, Pattern Recognition, 26 (10) (1993)
12 702 D. Zhang, W. Shu / Pattern Recognition 32 (1999)  Grolier Incorporated, The Encyclopedia American. Grolier, U.S.A. (1995).  V.S. Srinivasan, N.N. Murthy, Detection of singular points in fingerprint images, Pattern Recognition 25 (2) (1992)  Kalle Karu, Anil K. Jain, Fingerprint classification, Pattern Recognition 29 (3) (1996)  James H. Mcintish, Kathleen M. Mutch, Matching straight lines, Comput. Vision Graphics Image Process. 43 (1988)  Thoe Pavlidis, Algorithms for Graphics and Image Processing, Computer Science Press, Rockville, Md. (1982).  Paul S. Wu, Ming Li, Pyramid edge detection based on stack filter, Pattern Recognition Lett. 18 (4) (1997)  T.S. Chan, Raymond K.K. Yip, Line detection algorithm, Proc. 13th ICPR (1996) pp  J. Brian Burns, Allen R. Hanson, Edward M. Riseman, Extracting straight lines, IEEE Trans. Pattern Anal. Mach. Intell. 8 (4) (1986)  Azriel Rosenfeld, Avinash C. Kak, Digital Picture Processing. Academic Press, London (1982).  J.F. Keegan, How can you tell if two line drawings are the same?, Comput. Graphics Image Process. 6 (1) (1977) About the Author DAPENG ZHANG graduated in computer science from Peking University in 1974 and received his M.Sc and Ph.D degrees in computer science and engineering from Harbin Institute of Technology (HIT) in 1983 and 1985, respectively. From 1986 to 1988 he was a postdoctoral fellow at Tsinghua University and became an associate professor at Academia Sinica, Beijing, China. In 1988, he joined the University of Windsor, Ontario, Canada, as a Visiting Research Fellow in electrical engineering. He received his second Ph.D in electrical and computer engineering at University of Waterloo, Ontario, Canada, in Currently, he is an associate professor at the Hong Kong Polytechnic University. His research areas include automated biometric-based identification, neural systems and applications, VLSI system design methodology and parallel computer architecturetion. He has authored and co-authored over 100 papers including three books, as well as received several recognisable project awards. Dr Zhang is a senior member of the IEEE. About the Author WEI SHU received the B.E. Degree in automation control and the M.E. Degree in pattern recognition and intelligence control from Tsinghua University, Beijing, in 1990 and 1994, respectively. He is currently a Research Assistant at Department of Computer Science at City University of Hong Kong as well as a Ph.D student in Department of Automation at Tsinghua University. His research interests include pattern recognition, image analysis and biometrics.