Li S.Z., Jain A.K. (eds.) Encyclopedia of Biometrics

Подождите немного. Документ загружается.

where P

and Q

are curvature values in the neigh-

borhood of the points being matched in the two cur-

vature feature maps, and



P and



Q are the mean

curvature values in those neighborhoods. Figures 4

and 5 illustrate the process of matching two curvature

maps of the same and different user respectively. Red

(dark) colored pixels in the correlation map represent

high values of correlation while blue (light) represents

low correlation. Final matching score is the average of

pixel values in the correlation map. It can be observed

from Figure 4 that genuine matching results in a

correlation map with large regions of red colored pix-

els, indicating high correlation between the two curva-

ture maps be ing matched.

Decision Module

At the decision stage, match score from the feature

matching process is used to make a decision as to

whether the claimant is a genuine user or an impostor,

(in the veriﬁcation scenario) or to decide the identity

Palmprint, 3D. Figure 3 Sample 3D palmprint images (top row) and their corresponding curvature feature maps

(bottom row).

Palmprint, 3D. Figure 4 Matching of two curvature maps from the same subject.

1040

Palmprint, 3D

Palmprint, 3D. Figure 5 Matching of two curvature maps from different subjects.

Palmprint, 3D. Figure 6 A framework for combing 2D and 3D palmprint matching scores.

Palmprint, 3D

1041

of the user in identiﬁcation applications. Decision

threshold is selected based on a threshold criterion,

which is often the equal error rate (EER) of the system.

Figure 6 presents a framework for combining 2D and

3D palmprint matc hers. It employs a two level decision

strategy. Decision module I accepts or rejects a claim

based only on the 3D matching score. This allows the

system to reject fake palm s based on their 3D matching

scores. At the second level, Decision module II operates

on the combined (2D and 3D) matching score to

achieve performance improvement.

Performance Evaluation

The 3D palmprint veriﬁcation system developed in

[10] presents promising performance on a database

of 108 users. Six images were collected from each

user, resulting in 648 3D palmprint images in the

database. Matching each palmprint image with all

other images in the database, 1,629 genuine and

208,008 impostor match scores were obtained.

Figure 2b shows the receiver operating characteristic

(ROC) curve for an authentication system based on 3D

palmprint features. The reported system achieves an

EER of 0.99% on the aforementioned database.

Summary

Palmprint based recognition systems are extensively

researche d for applications such as physical ac c ess con-

trol, attendanc e tracking, and other personal veriﬁcation

tasks. They have certain advantages such as high user

acceptance and ease of use, over other biometrics. While

2D image based palmprint based recognition systems

have achieved high performanc e, they can be sensitive

tofactorssuchasspuriouspatternsonthepalmand

variations in the illumination. 3D palmprint based r ec-

ognition systems, on the other hand, offers high degree

of robustness along with comparable performance,

making them a good choice for high security applica-

tions. In addition, since 2D and 3D palmprint images

can be simultaneously acquired using a single image

acquisition device, these features can easily be combined

to form a multi-biometric system that can potentially

achieve signiﬁcantly higher performanc e than either of

the tw o palmprint features. The referen ce [10] provides

details of a device that can simultaneously acquire 2D

and 3D palmprint images. It also provides experimen-

tal results on score level fusion of 2D and 3D palmprint

matchers. While 3D palmprint based biometric system

has several advantages over other biometrics, size of

these systems, due to large capture area, can be pro-

hibitive for its use in devices like PDA and laptop.

Related Entries

▶ Hand Geometry

▶ Palmprint Feature

▶ Palmprint Matching

References

1. Zhang, D.: Palmprint Authentication, Kluwer, Dordrecht (2004)

2. Jain, A.K., Ross, A., Prabhakar, S.: An introduction to bio-

metric recognition. Special Issue on Image- and Video-based

Biometrics. IEEE Trans. Cir. Syst. Video Technol. 14(1), 4–20

(2004)

3. Zhang, D., Kong, W.K., You, J., Wong, M.: On-Line palmprint

identiﬁcation. IEEE Trans. Pattern Anal. Mach. Intell. 25(9),

1041–1050 (2003)

4. Chen, F. (2006) 3D Fingerprint and palm print data model and

capture devices using multi structured lights and cameras. U.S.

Patent No. 20060120576 (2006)

5. Koenderink, J.J., van Doorn, A.J.: Surface shape and curvature

scales. Image Vis. Comput. 10(8), 557–564 (1992)

6. Cantzler, H., Fisher, R.: Comparison of HK and SC curvature

description methods. Proceedings of the Third International

Conference on 3-D Digital Imaging and Modeling, QC, Canada,

pp. 285–291 (2001)

7. Besl, P.J., Jain, R.C.: Three-dimensional object recognition. ACM

Comput. Surv. 17(1), 75–145 (1985)

8. Woodard, D.L., Flynn, P. J.: Finger surface as a biometric identi-

ﬁer. Comput. Vis. Image Underst. 100(3), 357–384 (2005)

9. Besl, P., McKay, N.: A method for registration of 3D shapes. IEEE

Trans. Pattern Anal. Mach. Intell. 14(2), 239–256 (1992)

10. Kanhangad, V., Zhang, D., Luo, N.: A multimodal biometric

authentication system based on 2D and 3D palmprint features.

Proceedings of the SPIE, 6944: 69440C–69440C-9 (2008)

Palmprint Anatomy

▶ Anatomy of Friction Ridge Skin

1042

Palmprint Anatomy

Palmprint Authentication, 3D

▶ Palmprint, 3D

Palmprint Characteristics

▶ Palmprint Features

Palmprint Database

▶ Hand Databases and Evaluation

Palmprint Device

▶ Fingerprint, Palmprint, Handprint and Soleprint

Sensor

Palmprint Features

DAV I D ZHANG,LAURA L. LIU

Biometrics Research Centre, The Hong Kong

Polytechnic University, Kowloon, Hong Kong

Synonyms

Palmprint Representation; Palmprint Characteristics;

Handprint

Definition

Palmprint features refer to the representation of

palmprints which characterizes a palmprint in a stable

and unique way such that it has good discriminat-

ing ability for personal identiﬁcation. A palmprint

identiﬁcation system identiﬁes an indiv idual using

palmprint features which may or may not be observa-

ble to the naked eye. The selection of palmprint featu-

res is a fundamental problem in reliable palmprint

identiﬁcation.

Introduction

Palmprint is the skin patterns of the inner surface of

the human hand from the wrist to the root of the

ﬁngers. As a comparatively new biometric, palmprints

are rich in physical characteristics of skin patterns such

as lines, points and textures, which provide stable

and distinctive information sufﬁcient for separating

an individual from a large population. Compared

with other biometric traits, the advantages of palm-

print are user friendliness, environment ﬂexibility,

and discriminating ability.

A typical palmprint-based identiﬁcation system

involves

▶ pre-processing, feature extraction, feature

matching and decision-making. Normally, automatic

palmprint identiﬁcation systems can be classiﬁed into

two categories: ofﬂine and online. An ofﬂine system

usually processes previously captured palmprint

images which are often obtained from inked palmprints

or generated from

▶ forensic analysis, while an online

system captures palmprint images using a palmprint

scanner that is directly connected to a computer

for real-time processing. For both ty pes of palmprint

identiﬁcation systems, the selection of uni que features

to identify a person is a fundamental issue to be solved.

Many unique features of palmprint images can

be used for personal authentication, which include

▶ principal lines, wrinkles, ridges, minutiae points, sin-

gular points, and texture. Various features can be

extracted at different image resolutions. Features such

as minutiae, ridges and singular points can be obtained

in high-resolution palmprint images of at least 400 dpi

(dots per inch) [1–3] and are difﬁcult or even cannot

be observed in low-resolution images (<100 dpi).

Nevertheless, features like principal lines and wrinkles

can be extracted from low-resolution palmprint images

and play an important role in palmprint identiﬁca-

tion [4–6]. In the real-time palmprint identiﬁcation

system developed by the Biometric Research Centre at

Palmprint Features

1043

The Hong Kong Polytechnic University [7], a very

low resolution (75 dpi) palmprint image can be captured

by a CCD (charge-coupled device) camera-based palm-

print acquisition device (Fig. 1). In a typical palmprint

image captured by this system, the main patterns can

be generalized to principal lines, w rinkles, and creases

(also called ridges) (Fig. 2).

Regardless of image resolution and applications,

the selected feature should have good discriminating

ability to exhibit large variations between individuals

and small variations between samples from the same

palm. Thus, image feature extraction/representation

play an essential role in palmprint-based biometrics.

In general, the palmprint features can be classiﬁed into

three categories [8]: texture-, line-, and appeara nce-

based features.

Texture Features

Texture representation of coarse level palm print classi-

ﬁcation provides an effective approach to palmprint

recognition. In the spatial domain, considering a palm-

print as a texture image, a statistical approach, e.g.,

Laws’ convolution masks, can be used to compute the

texture energy of palmprints [2]. In the frequency

domain, a polar coordination system ðr; yÞ can be

established to represent the Fourier transform of palm-

print images where less compact information indicates

stronger line features in the spatial domain [9]. In

addition, if a palmprint image has a strong line, there

is more information along the line’s perpendicular

direction in the frequency domain. In this polar coor-

dination system, the energy change tendency along

r corresponds to the intensity of a palmprint’s creases

and that along y to their directions. Another frequency

domain approach to the palmprint feature extraction

relies on the discrete cosine transform (DCT) of the

image pixels [10].

Two-dimensional Gabor ﬁlter, an effective tool for

texture analysis, is widely used to extract texture fea-

tures in palmprint-based biometrics [7, 11, 12]. A

circular 2-D Gabor ﬁlter in the spatial domain has

the following general form:

Gðx; y; y; u; s; bÞ¼

2psb

exp p



exp 2iux

ðÞ;

ð1Þ

where i ¼

ﬃﬃﬃﬃﬃﬃﬃ

1

; x

¼ðx  x

Þcos y þðy  y

Þsin y;

¼ðx  x

Þsin y þðy  y

Þcos y; ðx

; y

Þ is the

Palmprint Features. Figure 1 The Design Principle of the Palmprint Acquisition Device.

1044

Palmprint Features

center of the function, u is the radial frequency in

radians per unit length, and y is the orientation of

the Gabor function in radians. s and b are the standard

deviations of the Gassian envelop along x and y axes,

respectively, and s ¼ b for the circular 2-D Gabor

ﬁlter. In order to provide more robustness to illumina-

tion, the Gabor ﬁlter is tuned to 0 DC (average value)

with the application of the following formula:

G½x; y; y; u; s; b¼G½x; y; y; u; s; b



i¼n

j¼n

G½i; j; y; u; s; b

ð2n þ 1Þ

;

ð2Þ

where ð2n þ 1Þ

is the size of the ﬁl ter and

G½i; j; y; u; s; b denotes the corresponding discrete

Gabor ﬁlter. The adjusted Gabor ﬁlter is ﬁrst used to

convolute with the pre-processed central part palm-

print sub-image and then each sample point in the

ﬁltered image is coded to two bits according to the

signs of the real and imaginary parts of the convolution

results. Using this coding method, only the phase in-

formation in palmprint images is stored in the feature

vector, which is called the PalmCode.

The PalmCode has been improved in two aspects.

To reduce the correlation between PalmCodes, a fusion

rule has been developed to produce a single feature,

called the Fusion Code. According to this fusion rule,

multiple ellipt ical Gabor ﬁlters (s 6¼ b) with different

orientations are utilized to extract the magnitude

and the phase information on a palmprint image and

then employ the magnitude for fusion and the phase

for the ﬁnal feature (Fig. 3). To utilize the orientation

information of palmprints which is absent in a Palm-

Code, a competitive rule is further designed to extract

the orientation characteristics of palm lines by using a

set of neurophysiology-based Gabor ﬁlters. This com-

petitive rule is a winner-take-all rule where the orien-

tation of the sample point is determined by the Gabor

ﬁlter which gives the minimum ﬁlter response. The

corresponding feature codes, referred to as the Com-

petitive Code (Fig. 4), are more robust to different

capturing environments and devices.

The ordinal feature representation is another power-

ful method to capture the texture features from low-

resolution palmprint images [13]. In low-resolution

palmprint images, the main palmprint patterns are

negative line segments and can be characterized using

▶ ordinal measures. This measure qualitatively com-

pares two elongated, line-like image regions which are

orthogonal in orientation and generates one bit feature

code. The 2-D Gaussian ﬁlter which is used to obtain

the weighted average intensity of a line-like region is

formulated as follows:

f ðx; y; y Þ¼exp 

x cos y þ y sin y





x sin y þ y cos y



;

ð3Þ

where y denotes the orientation of the 2-D Gaussian

ﬁlter, s

and s

denote the ﬁlter’s horizontal and

vertical scales, respectively and s



should be higher

Palmprint Features. Figure 2 The Main Patterns in a Low-resolution Palmprint Image.

Palmprint Features

1045

than 3 to guarantee that the region is shaped like a line.

The orthogonal line ordinal ﬁlter is deﬁned as the com-

parison/difference of two orthogonal line-like palm-

print image regions. For each local region, three

oriented ordinal ﬁlters with 0, p

6 and p

3 are per-

formed on it to obtain three bit ordinal codes based on

the sign of ﬁltering results. Finally, three ordinal tem-

plates called the Ordinal Co de are obtained as the

feature of the input palmprint image.

Line Features

In palmprint images, lines and textures are the most

observable features and lines are more appealing than

texture to the human eye. The line feature of palmprints

can be generalized to principal lines, wrinkles, and

creases. Normally, there are three principal lines in a

palmprint: the heart line, the head line, and the life

line. These lines vary little over time, and their shapes

and locations on t he palm are the most important

physiological features for individual recognition.

Most wrinkles are thinner and more irregular than

the principal lines; some wrinkles are not only as

strong as the principal lines, but are also stable and

reliable for identiﬁcation. Creases cover the entire

palm just like ridges in a ﬁngerprint and cannot be

observed in low-resolution images.

The principal lines and wrinkles, also called

palm lines, are stable and reliable for individual

identiﬁcation and can be exploited and derived from

a low-resolution palmprint image. Many algorithms

have been developed to extract palm line features

for personal authentication. Regarding palm lines as

Palmprint Features. Figure 3 Block Diagram of Fusion Code-based Palmprint Identification.

1046

Palmprint Features

several straight line segments, line-like features can

be extrac ted by employing some directional masks,

as well as Sobel and morphological operations [1, 14].

Applying a wavelet transform followed by directio-

nal context modeling, a set of statistical signatures

of palm lines can be obtained to characterize palm-

prints [4].

The structural features of palm lines are a natural

choice for palmprint recognition in that they can

describe a palmprint clearly and are robust against

illumination and noises. According to the properties

of palm lines, a set of directional line detectors are

devised to detect palm lines in different directions

[5]. To ensure that the details of the palm line structure

are not lost, the detected lines are ﬁnally represented

by a chain code which is a pixel-by-pixel direction

code of a line.

Except the structural features, the width feature

generally reﬂects the strength information of palm

lines ( Fig. 5) and is also important for describing the

characteristics of a palmprint especially when different

palmprints have similar palm line structures. The

Palmprint Features. Figure 4 An Example of Competitive Code. (a) Pre-processed Image. (c)–(h) The Winning Code 0, 1,

2, 3, 4 and 5, respectively. (b) Combination of (c)–(h).

Palmprint Features

1047

width features of palm lines as well as the structural

features can be extracted by using a wide line dete ctor

[6]. This detector can extract a line completely by

inverting the kernel mass which is obtained by an

isotropic nonlinear ﬁlter.

Appearance Features

In palmprint recognition, the performance of subspace

learning methods has been promising [15–17]. These

methods efﬁciently characterize the overall space of

raw images by a low-dimensional subspace where stan-

dard statistical methods can be used to determine the

range of appearance of palmprints.

The Principal Component Analysis (PCA) has been

widely used in face recognition and it offers good char-

acterization for palmprint recognition. Based on the

Karhunen-Loeve (K-L) transform which is an optimal

transform for eliminating statistical correlation, the

original palmprint images used in training are trans-

formed into a small set of characteristic feature images,

called ‘‘eigenpalms,’’ which are the eigenvectors of the

training set. Feature extraction is then performed by

projecting a new palmprint image into the subspace

spanned by the ‘‘eigenpalms.’’

Features of palmprints extracted by the PCA are

actually ‘‘global’’ features of all palmprint images im-

plying that they are not necessarily good discriminati ve

representations. The LDA (linear discriminant analy-

sis), based on linear projections, seeks a linear trans-

formation by maximizing between-class variance and

minimizing within-class variance and thereby has

strong discriminability. Applying the LDA for palm-

print recognition, palmprints can be projected from a

high-dimensional space to a signiﬁcantly lower dimen-

sional feature space spanned by Fisherpalms. In this

low-dimensional feature space, palmprints from differ-

ent palms can be discriminated more efﬁciently.

Both the PCA and LDA methods attempt to ﬁnd the

holistic features of the whole enrollment palmprints and

consequently miss the crucial details. Applying the Lo-

cality Preserving Projection (LPP) to palmprint images,

‘‘Laplacianpalms’’ are obtained by ﬁnding the optimal

linear approximation to the eigenfunctions of the Lapla-

cian Beltrami operator on the manifold [17]. They are

linear projective maps that arise by solv ing a variation-

al problem that optimally preserves the neighborhood

structure of the dataset. While the Eigenpalm method

aims to preserve the global structure of the palmprint

image space and the Fisherpalm method preserves the

global discriminating information, the Laplacianpalm

method strives to preserve the local structure of the

palmprint image space.

Summary

Several palmprint features have been investigated for

personal authentication and have yielded promising

results. According to the performance analysis of

various palmprint features presented in the litera-

ture, texture-based features, such as ordinal features

and competitive coding, revealed the best perfor-

mance in a large palmprint population. The study

of how to effectively combine different palmprint

Palmprint Features. Figure 5 Structural Feature vs. Width Feature. (a) Pre-processed Palmprint Image. (b) Its Structural

Features. (c) Width Features.

1048

Palmprint Features

representations to achieve higher performance has

attracted the attention of an increasing number of

researchers.

Related Entries

▶ Anatomy of Hand

▶ Feature Extraction

▶ Local Feature Filters

▶ Palmprint Matching

▶ Palmprint, 3D

References

1. Zhang, D., Shu, W.: Two novel characteristics in palmprint

veriﬁcation: Datum point invariance and line feature matching.

Pattern Recognit. 32, 691–702 (1999)

2. You, J., Li, W.X., Zhang, D.: Hierarchical palmprint identiﬁca-

tion via multiple feature extraction. Pattern Recognit. 35,

847–859 (2002)

3. Chen, J., Zhang, C., Rong, G.: Palmprint recognition using

crease. In: Proceedings of the 7th International Conference on

Image Processing (ICIP), Thessaloniki, Greece, 234–237 (2001)

4. Zhang, L., Zhang, D.: Characterization of palmprints by wavelet

signatures via directional context modeling. IEEE Trans. Syst.

Man Cybern. Part B. 34, 1335–1347 (2004)

5. Wu, X.Q., Wang, K.Q., Zhang, D.: Palm-line extraction and

matching for personal authentication. IEEE Trans. Syst. Man

Cybern. Part A. 36, 978–987 (2006)

6. Liu, L., Zhang, D.: A Novel palm-line detector. In: Proceedings

of 5th International Conference on Audio- and Video-Based

Biometric Person Authentication (AVBPA), New York, 563–571

(2005)

7. Zhang, D., Kong, W.K., You, J., Wong, M.: Online palmprint

identiﬁcation. IEEE Trans. Pattern Anal. Mach. Intell. 25,

1041–1050 (2003)

8. Kumar, A., Zhang, D.: Personal authentication using multiple

palmprint representation, Pattern Recognit. 38, 1695–1704

(2005)

9. Li, W., Zhang, D., Xu, Z.: Palmprint identiﬁcation by fourier

transform. Intern. J. Pattern Recognit. Artif. Intell. 16 , 417–432

(2002)

10. Kumar, A., Zhang, D.: Personal recognition using hand-shape

and texture. IEEE Trans. Image Processing 15, 2454–2461 (2006)

11. Kong, A., Zhang, D., Kamel, M.: Palmprint identiﬁcation using

feature-level fusion. Pattern Recognit. 39, 478–487 (2006)

12. Kong, A.W.K., Zhang, D.: Competitive coding scheme for palm-

print veriﬁcation. In: Proceedings of 17th International Confer-

ence of Pattern Recognition (ICPR), Cambridge, UK, 520–523

(2004)

13. Sun, Z.N., Tan, T.N., Wang Y.H., Li, S.Z.: Ordinal palmprint

representation for personal identiﬁcation. In: Proceedings of

IEEE Computer Society International Conference on Computer

Vision and Pattern Recognition (CVPR), San Diego, USA,

279–284 (2005)

14. Han, C.C., Cheng, H.L., Fan, K.C., Lin, C.L.: Personal authenti-

cation using palmprint features. Pattern Recognit. 36, 371–381

(2003)

15. Lu, G., Zhang, D., Wang, K.Q.: Palmprint recognition using

eigenpalm-like features. Pattern Recognit. Lett. 24, 1473–1477

(2003)

16. Wu, X.Q., Zhang, D., Wang, K.Q.: Fisherpalms based palmprint

recognition. Pattern Recognit. Lett. 24, 2829–2838 (2003)

17. Wang, J.G., Yau, W.Y., Suwandy, A., Sung, E.: Person recognition

by fusing palmprint and palm vein images based on ‘‘Lapla-

cianpalm’’ representatio n. Pattern Recognit. 41, 1514–1527

(2008)

Palmprint Matching

ANDREW BEN G JIN TEOH

Biometrics Engin eering Research Center (BERC),

School of Electrical and Electronic Engineering, Yonsei

University, Seoul, South Korea

Synonyms

Comparison; Dissimilarity; Similarity

Definition

Palmprint matching is a comparison process of two

given palmprint and returns either a dichotomy deci-

sion (yes or no) or a degree of similarity. Due to the

rich features in a palm, including geometrical features

(e.g., width, length, area etc. of a palm), principle lines,

ridges, singular points, minutiae points, and texture,

the matching algorithms require an intermediate

palmprint representation to be extracted through a

▶ feature extraction stage. Based on these palmprint

features, several approaches to palmprint matching

have been devised and they can be broadly classiﬁed

into two major categories: geometry-based matching

and feature-based matching. The integration of two

approaches can be done in hierarchical manner to

improve the palmprint recognition systems in terms

of performance and speed.

Palmprint Matching

1049