Biometric Authentication System based on Hand Geometry and

Palmprint Features

Laura Gulyás Oldal and András Kovács

John von Neumann Faculty of Informatics, Óbuda University, Budapest, Hungary

Keywords: Palmprint, Hand Geometry, Biometric Authentication.

Abstract: In today’s society, biometric authentication has gained more significance, since it uses physical characteristics

of a person for identification. Physical features provide greater security compared to ownership or knowledge

based factors. More and more physiological measures prove to be great characteristics for personal

authentication. A multimodal biometric authentication system has the advantage of using multiple physical

characteristics for authentication achieving greater accuracy. If one modality fails to identify a person with

high accuracy, other modalities are employed. However, in these kind of systems, every modality has a

different imagery data requirement, which provides multiple captured images for evaluation. The method

described in the article uses the same input data for processing multiple physiological features at once.

Biometric characteristics used by the system are hand geometry and palmprint features. The imagery data

requirement is a high-resolution image of a well-lit hand with dark background. Capturing the image in good

sanitary conditions has become an important requirement in the past few years. Advantage of a high-resolution

image compared to images captured with dedicated hardware devices like fingerprint or palmprint scanners,

is contactless capturing of the image. Another benefit of a high-resolution camera usage is lower cost claims

compared to the other systems using dedicated hardware for image capturing.

1 INTRODUCTION

Biometric authentication has become an important

topic in security systems, as it is used for automated

identification, criminal or forensic applications and

access control systems. Some of the physiological

features of a person suitable for authentication are

fingerprint, palmprint, palm vein features, face

characteristics, hand geometry features, etc.

(Golfarelli, 1997).

Figure 1: The convex hull of the point set after the auxiliary

point removal.

A palmprint consists of multiple characteristics

suitable for authentication, such as: wrinkles,

epidermal ridges and principal lines. The principal

lines of a palm provide quality measures for

identification with high confidence. Some of the hand

geometry features, appropriate for identification are

finger lengths, and widths, maximal inscribed circle

radius, palm area, width and height, finger

proportions, etc. Features of the hand and palm can be

extracted from a single high-resolution image. The

aggregation of hand geometry and palmprint features

in the system provides a more reliable authentication

system using only one imagery data. The proposed

method is an improved version of our previously

published work (Gulyás Oldal, 2020). The main

improvements of the current method are simpler

keypoint detecting algorithms and testing the method

on a much larger data set. The simplified algorithms

provide the same results as the original solution, but

with a much smaller margin of error. Another

improvement made in the current version of the

method is the performance improvement of key point

detection. As the key point detecting algorithms do

not require a high resolution image, the algorithms

Oldal, L. and Kovács, A.

Biometric Authentication System based on Hand Geometry and Palmprint Features.

DOI: 10.5220/0010408900580065

In Proceedings of the International Conference on Image Processing and Vision Engineering (IMPROVE 2021), pages 58-65

ISBN: 978-989-758-511-1

use a downsized version of the original image, which

results in much shorter run time.

Figure 2: Two categories of the hull.

The imagery data requirement for the proposed

method is a high-resolution image containing the

entire hand, the foreground has to be well lit, while

the background has to be as dark as possible. Another

prerequisite of the image is the vertical orientation of

the hand, as the image is not rotated during keypoint

detection.

2 RELATED WORKS

In this section we give a brief overview of some

methods using hand geometry or palmprint features

for personal authentication.

In (Badrinath, 2012) the authors propose a

method for palmprint based recognition where the

palmprint is divided into several blocks, which are

further processed using principal component analysis

(PCA). The features used for matching are the phase-

differences of the vertical and horizontal phase. The

feature matching algorithm computes the Hamming

distance of features and uses the given values for the

calculation of matching score. The method has an

EER (equal error rate) value of 0.25%.

Method described in (Li, 2012) uses a unique

approach for palmprint matching. The authors of the

paper consider the improper alignment of ROI-s of

two palmprints to be the biggest flaw of most existing

palmprint matching methods. The correct alignment

of the ROI-s is achieved by extracting the palmprint

principal lines and computing the adequate rotation

and translation values with ICP (Iterative Closest

Point) algorithm. Given the properly aligned

palmprint images, an arbitrary method for palmprint

feature extraction and recognition method is used.

The authors achieved the best results using the

CompCode (Huang, 2008) method, which uses Gabor

filters in six different orientations for the direction

estimation of palmprint image points. The

CompCode method alone has an EER value of

0.0388%, while in joint with the proposed method

had an EER value of 0.0201%. The additional step

results in increased feature extraction and matching

time as well.

The method proposed in (Badrinath, 2012) shows

great results, however, the authors assume the proper

alignment of the ROI-s. In (Li, 2012) method the

authors propose a method for proper feature map

alignment using the palmprint principal lines. The

method is used as an additional step of existing

palmprint recognition methods. The features

extracted for ROI alignment and palmprint matching

may differ. Our method uses a similar approach,

however, the palmprint principal lines are used for

ROI alignment and palmprint matching as well,

which results in a simpler implementation of a

palmprint recognition system.

Figure 3: 4 subcategories of the second group and classified

points.

3 HAND GEOMETRY

CHARACTERISTIC POINT

DETECTION

3.1 Preprocessing

The preprocessing of the imagery data consists of

gray scaling, downsizing, binarization, noise removal

and morphological edge detection. The key point

detection does not require a high resolution image,

therefore, the original image is downsized. This step

provides greater efficiency for keypoint detecting

algorithms and hand geometry feature extraction. The

palm feature extracting algorithms are executed on

the original image, as the palm lines require high

resolution image for extracting quality principal line

features. Subsequently, the downsized image is

binarized with a threshold determined by Otsu’s

method. Having very different intensities of the

Biometric Authentication System based on Hand Geometry and Palmprint Features

foreground and background insures satisfactory

results using Otsu’s algorithm (Xu, 2011). As the

binarized image contains significant amount of noise,

the most effective method for noise removal is blob

detection followed by filtering the largest blob

(Grycuk, 2014). The extracted blob is the object

defined by the shape of hand. This is an improvement

of the previously published method, it is less sensitive

to lighter areas of the background. The original

method used median-filter for reducing noise, which

proved to be effective only on inputs having almost

completely black background. Any reflected light

from the background causes non salt and pepper

noise. The median-filter is not an effective method for

the removal of bigger noise objects, consequently,

this step had to be introduced.

Figure 4: The reduced point set for finger valley detection.

Figure 5: The output of the finger valley detection.

Figure 6: Calculated finger lengths.

Figure 7: Calculated maximal inscribed circle.

The next step of preprocessing the image is

morphological edge detection. The morphological

operations used for one-pixel thick edge extraction

are erosion and subtraction (Hsiao, 2005)

(Maksimović, 2020).

3.2 Fingertip Detection

Given the image containing one-pixel thick edges, the

algorithm determines the fingertips. The set of edge

points is reduced by retrieving the topmost points of

every column on the image. The hull displayed on

Fig. 1 is determined by multiple operations including

Graham scan (Graham, 1972). With the analysis of

the distances between successive points, the right or

left orientation of the hand can be determined.

Subsequently, the hull points are classified to the

following two groups:

• group of thumb points

• group of remaining four fingertip points.

The two categories are displayed on Fig. 2, the

points of thumb are marked with red color, while the

other category is marked with a blue color. The

middle point from the first group is selected and

represents the fingertip of the thumb. The second

category is divided to four additional groups

displayed on Fig. 3, and the remaining four fingertips

are determined by selecting the local maxima from

every group.

3.3 Wrist Point Detection

Wrist point detection is essential for hand area

computation as the image section below the wrist

cannot be included during hand area computation.

The mentioned image section is not part of the hand,

but the arm. The wrist point detection method

described in (Gulyás Oldal, 2020) proved to be

inaccurate in case of inadequate light conditions on

the imagery data. The algorithm described below

IMPROVE 2021 - International Conference on Image Processing and Vision Engineering

proved to be much more effective in different light

conditions.

The detection starts with extraction of the leftmost

and rightmost edge points in each row, followed by

filtering the points below the thumb. The average

distances between the rightmost and leftmost points

are calculated for every row considering former n

distances as well.

The wrist points are determined by the points with

the biggest rate of change of average distances.

3.4 Finger Valley Detection

The finger valley search starts with elimination of

leftmost and rightmost points from the set of edge

points.

Figure 8: Points for ROI extraction.

The edge points are reduced further by defining a

circle, as follows:

• Centre: 𝐶𝑂𝑀

• Radius: 𝑘 ∙ 𝑑𝑖𝑠𝑡𝐶𝑂𝑀, 𝑚𝑖𝑑𝑑𝑙𝑒_𝑓𝑖𝑛𝑔𝑒𝑟_𝑡𝑖𝑝,

where 𝐶𝑂𝑀 is the center of mass of the hand,

𝑑𝑖𝑠𝑡𝐶𝑂𝑀, 𝑚𝑖𝑑𝑑𝑙𝑒_𝑓𝑖𝑛𝑔𝑒𝑟_𝑡𝑖𝑝 the distance

between the tip of middle finger and center of mass

and 𝑘 a free parameter (𝑘0.55 in the implemented

algorithm). The points inside the circle are kept, the

remaining points are eliminated. The filtered point set

is displayed on Fig. 4.

The following steps of finger valley detection

include blob detection. The blobs extracted are four

objects, where every blob contains points of one

valley (Grycuk, 2014)[4]. The valleys between the

fingers are finally determined by discovering the local

minimum for every object. The detected finger valley

points are displayed on Fig. 5. The process of valley

detection is significantly simplified compared to the

original version (Xu, 2014), as the initial method had

a large margin of error. The original process of valley

detection comprised of multiple steps built on each

other, which made the algorithm more error prone.

The improved method determines the points of

valleys independently from each other.

4 HAND GEOMETRY FEATURES

AND MATCHING

Absolute hand geometry measurements like finger

proportions, lengths or maximal inscribed circle need

to be calculated with respect to other measurements

for the features to be resolution independent

(Rutkowski, 2014).

The features considered during matching are the

following:

• Finger length proportions (with respect to the

middle finger length)

• Area of the largest inscribed circle (with

respect to the area of the hand) (Sanchez-

Reillo, 2000)

• Middle finger length (with respect to the area

of the hand) (Bača, 2012)

The considered finger lengths are the distances

between fingertip and finger valley points. Points

used for finger length calculation are displayed on

Fig. 6. The proportion of finger lengths characterize

well a hands geometrical characteristics. Finger

length proportions are a good descriptor of the hand’s

geometrical attributes (Sanchez-Reillo, 2000).

The finger length proportions are calculated with

respect to the middle finger length, as it proved to

have steadiest position among the fingers.

Another great descriptor of the hand shape is the

area or radius of the largest inscribed circle (Sanchez-

Reillo, 2000).

In the implemented solution, the largest inscribed

circle is determined with a modified hill climbing

algorithm (Kumar, 2003) (Ohashi, 2003). The

detected circle with this method is displayed on Fig.

7. The similarity between two hands geometrical

features is computed with the following formulas:

𝑭 𝑡

.,



𝐿







𝐿







𝐿







𝐿





 ∙ 100

(1)

𝑨𝑡

.,

𝑇′



 ∙ 100

(2)

𝑳𝑡

.,

𝐿′



 ∙ 100

(3)

𝑯

𝟐

𝟖

𝑨

𝟑

𝟖

𝑭

𝟑

𝟖

𝑳

(4)

Where:

• 𝐿′



– 𝑖-th finger length proportion

difference

• 𝑇′



– maximal inscribed circle area

difference

• 𝐿′



- middle finger length difference

Biometric Authentication System based on Hand Geometry and Palmprint Features

• 𝑡

,



𝑥









𝑎function, which normalizes

the input to the



𝑎, 𝑏



range

• 𝐹 – finger proportion similarity

• 𝐴 – maximal inscribed circle area difference

• 𝐿 – finger length difference

• 𝐻 – total hand geometry similarity

The weights of measurements considered during

similarity level calculation are defined with respect to

the steadiness of the feature.

5 PALMPRINT EXTRACTION

AND MATCHING

5.1 Region of Interest (ROI) Extraction

The palmprint analysis starts with determining an

image section, where the principal palm lines should

be analyzed (Tóth, 2020). A square ROI is chosen in

the proposed method, as two points (𝑃



,𝑃



) are

sufficient for the determination of the whole square.

The remaining two points calculated from 𝑃



,𝑃



points are 𝑃



,𝑃



marked on Fig. 8.

Figure 9: A binary image of extracted palm line before and

after additional noise removal.

Figure 10: Dilated palm line images and the intersection.

𝑷

𝟏

 𝑥





–

𝑑

,𝑥







𝑑



(5)

𝑷

𝟐

 𝑥







𝑑

,𝑥







𝑑



(6)

𝑷

𝟑

𝑥





 𝑏,𝑦





𝑎

(7)

𝑷

𝟒

𝑥





 𝑏,𝑦





𝑎

(8)

On Fig. 8 the 𝑃



, 𝑃



, 𝑃



are the points of finger

valleys, and 𝑑 the vertical distance between the 𝑃



and 𝑃



valleys.

5.2 Palmline Extraction

The palmline extraction method is highly inspired by

the (Bruno, 2014) study. After the extraction of the

ROI image section on the original image, the palm

image is downsized to 128∙128 size, the method

described below produces best results on small

resolution image. The process is followed by gray

scaling, contrast stretch and smoothing. Averaging

filters are applied on the processed image with 4

different averaging filters with a mask size of 5∙5. The

different averaging filters smooth the palm lines in

different directions, which are vertical, horizontal and

two diagonal directions (Lin, 2016). On the resulting

images a bottom hat filter is applied. The four

resulting images are consolidated into one image by

addition of the partial results (Bruno, 2014). The

process is followed by the elimination of low

intensity pixels, thresholding and noise removal for

the purpose of erasing isolated spikes and blocks

(Bruno, 2014). The method produces satisfying

results, however, a considerable amount of noise was

present on the output images (Fig. 9).

An alternative noise removal method had to be

introduced, which is filtering the blobs in the palm

image by height, width and area (Grycuk, 2014). The

discontinuity of palm lines requests an additional step

which is extending every blob larger than a certain

size with a certain number of pixels. Following this

step, the blobs are filtered by height, width and area.

This method eliminates larger noise objects, which

were not connected to the main palm lines during line

extension. An essential last step is the removal of the

added pixels on the filtered image.

5.3 Palmprint Matching

The result of palm line extracting algorithms were

grayscale images containing the principal lines of the

palm. Template matching is the initial step of palm

image comparison. Points of the two images are

linked based on the similarity of the surrounding

pixels (Yuan, 2011).

The output of template matching is linked pairs of

points containing true matches and mismatches

(Hisham, 2015) (Kertész, 2018). The linked points

are further filtered with RANSAC (Random Sample

Consensus) algorithm. The algorithm creates a

homography matrix based on the inlier pairs of points

(Batyrshin, 2019) (Baráth, 2019).

IMPROVE 2021 - International Conference on Image Processing and Vision Engineering

With the use of homography matrix, an affine

transformation is performed on one of the input

images (Yaniv, 2010). With this method the feature

points of one image are projected to the other input.

However, the projected image and the reference

image might not completely overlap, the two images

are dilated and the intersection is computed from the

two. (Fig. 10). The highest intensity image sections

are the most relevant during matching, as these

represent the main principal lines. The pixels which

are taken into consideration are the pixels which

satisfy the following condition:

𝐼𝑚𝑔



𝑥, 𝑦



𝑘 ∙ 𝑀

!

(9)

where 𝑀

!

is the mean intensity value of non-black

pixels and 𝑘 is a free parameter (𝑘 = 1.7 in the

proposed method). The presence of the extracted

pixels on the intersection determines the similarity

level of the two palms.

6 COMBINING HAND

GEOMETRY AND PALMPRINT

The result of the palmprint matching and hand

geometry feature matching are summarized into one

single value considering the quality of the extracted

features (Dubrofsky, 2009). Therefore, different

weights are used for the consolidation of the partial

results.

𝑺  𝟎. 𝟐 𝑯  𝟎. 𝟖 𝑷

(10)

where:

• 𝑆 – total similarity score

• 𝐻 – similarity score of hand geometry features

• 𝑃 – similarity score of palm features

Figure 11: Similarity levels of cases given identical hands.

Figure 12: Similarity levels of cases given different hands.

7 RESULTS

The algorithms were tested on 214 2592∙1944

resolution images of 14 different subjects.

The ages of the subjects vary from 20 to 60, half

of the subjects are females, and the other half are

males. The outcome of keypoint detection was tested

and analyzed manually.

Subsequently, the feature extraction and feature

matching was tested on 450 combinations given

identical hands, and 450 combinations given different

hands.

The results of the algorithm on identical subjects

is shown on Fig. 11, while the results considering

different subjects is shown on Fig. 12.

Considering the distribution of similarity levels

given identical hands, it can be said, that in 75% of

cases the algorithm confirmed an identity of a person

with high confidence. The similarity level in these

cases fell in the 80-100% range. In more than 96% of

differing hand combinations, the method resulted

with a similarity level between 0% to 70%, while in

most cases the similarity level falls in the 30-60%

range. The results suggest, that the algorithm in

majority of different hand cases produces a value very

different compared to identical hands.

This behavior allows us to view the effectiveness

of the proposed method by analyzing FAR (False

Acceptance Rate) and FRR (False Rejection Rate) on

different acceptance thresholds (Ross, 2004). In Fig.

13. the FAR and FRR percentages are displayed

considering a few acceptance thresholds. By defining

68% as the acceptance threshold, FAR (False

Acceptance Rate) is 4.6%, and FRR (False Reject

Rate) is 14,2%.

Biometric Authentication System based on Hand Geometry and Palmprint Features

Figure 13: FAR and FRR considering different acceptance

thresholds.

8 CONCLUSION

The method described in the paper provides a simple

solution for a low-cost contact-free authentication

system based on palmprint and hand geometry

features. The implemented method produced

promising outcome and enthuses further

development. A future work for the method is an

improvement of the palm feature matching algorithm,

where the method will use distance transform for

determining the similarity of the palm features

following the affine transformation of one of the palm

images. This improvement would provide a solution,

where the varying intensities of palm lines would

affect the effectiveness of the method to a lesser

extent. A further future work for the method consist

of improving the hand geometry feature matching. In

the future, the method will use a machine learning

algorithm for determining the optimal weights

considered during matching with a machine learning

algorithm. The current method in comparison with

the previously published method (Gulyás Oldal,

2020) is significantly simpler, and therefore less error

prone. The keypoint detection takes place on a

downsized image, which made a significant

difference in the run time of the algorithms.

Furthermore, valley detection consists of independent

simple steps, which proved to have a smaller margin

of error and is much faster. During hand geometry

feature matching, largest inscribed circle area is

considered in replacement of the palm area, as the

determination of the latter consisted of determining

two additional root points of the fingers.

The two points could not be determined with high

confidence, only a rough estimation was made. This

resulted in a deviating palm area in different image

samplings, and made the hand geometry feature

matching less accurate. The method was tested on a

significantly larger data set, which gives a better

overview of the method’s accuracy.

REFERENCES

Laura Gulyás Oldal, András Kovács, "Hand geometry and

palmprint-based authentication using image

processing," 2020 IEEE 18th International Symposium

on Intelligent Systems and Informatics (SISY),

Subotica, Serbia, 2020, pp. 125-130

Alessandro Bruno, Paolino Carminetti, Gentile Vito, Marco

La Cascia, Emanuele Mancino, “Palmprint principal

lines extraction” BIOMS 2014 - 2014 IEEE Workshop

on Biometric Measurements and Systems for Security

and Medical Applications, Proceedings, 2014

Xiangyang Xu, Shengzhou Xu, Lianghai Jin, Enmin Song,

“Characteristic analysis of Otsu threshold and its

applications” Pattern Recognition Letters, Vol. 32.,

2011, pp. 957-960I.

Rafal Grycuk, Marcin Gabryel, Marcin Korytkowski, Rafal

Scherer, and Sviatoslav Voloshynovskiy, “From Single

Image to List of Objects Based on Edge and Blob

Detection”, Lecture Notes in Computer Science, 2014,

pp. 605-615

Ying-Tung Hsiao, Cheng-Long Chuang, Joe-Air Jiang and

Cheng-Chih Chien, "A contour based image

segmentation algorithm using morphological edge

detection," 2005 IEEE International Conference on

Systems, Man and Cybernetics, Waikoloa, HI, 2005, pp.

2962-2967 Vol. 3

Vladimir Maksimović , Branimir Jakšić , Mile Petrović ,

Petar Spalević, Mirko Milošević, “Analysis of Edge

Detection on Compressed Images with Different

Complexities”, Acta Polytechnika Hungarica, Vol. 17,

No. 4, 2020

Ronald L. Graham, “An Efficient Algorithm for

Determining the Convex Hull of a Finite Planar Set”,

Information Processing Letters, Vol 1., 1972, pp. 132-

133

Rutkowski, L., Scherer, R., Korytkowski, M., Pedrycz,

W., Tadeusiewicz, R., Zurada, J.M. Artificial

Intelligence and Soft Computing, ICAISC 2014,

Lecture Notes in Computer Science 2014, vol 8468.

Raul Sanchez-Reillo, Carmen Sanchez-Avila, Ana-

Gonzalez-Marcos, “Biometric Identification through

Hand Geometry Measuerements”, IEEE Transactions

on Pattern Analysis and Machine Intelligence, Vol. 22,

2000, pp. 1168-1171Y.

Miroslav Bača, Petra Grd, Tomislav Fotak, New Trends

and Developments in Biometricsm, Chapter 4: Basic

Principles and Trends in Hand Geometry and Hand

Shape Biometrics, 2012

Ajay Kumar, David C. M. Wong , Helen C. Shen, and Anil

K. Jain, “Personal Verification Using Palmprint and

Hand Geometry Biometric”, Audio- and Video-Based

Biometric Person Authentication, 2003, pp. 668-678

Takumi Ohashi, Zaher Aghbari, Akifumi Makinouchi,

“Hill-Climbing Algorithm for Efficient Color-Based

IMPROVE 2021 - International Conference on Image Processing and Vision Engineering

Image Segmentation”, Signal Processing, Pattern

Recognition, and Applications, 2003

Bence Tóth Tamás, Sándor Szénási, “Tree Growth

Simulation based on Ray-Traced Lights Modelling”,

Acta Polytechnika Hungarica, Vol. 17, No. 4, 2020

Lin, Sen & Xu, Tianyang & Yin, Xinyong. Region of

interest extraction for palmprint and palm vein

recognition. 9th International Congress on Image and

Signal Processing, BioMedical Engineering and

Informatics (CISP-BMEI), 2016, 538-542.

Weiqi Yuan,Sen Lin, Haibin Tong, Shudong Liu, “A

Detection Method of Palmprint Principal Lines Based

on Local Minimum Gray Value and Line Following” ,

International Conference on Hand-Based Biometrics,

2011

M. B. Hisham, Shahrul Nizam Yaakob, Raof R. A. A, A.B

A. Nazren, N.M.Wafi, “Template Matching Using Sum

of Squared Difference and Normalized Cross

Correlation”, 2015 IEEE Student Conference on

Research and Development (SCOReD), 2015

Gabor Kertész, Sándor Szénási, Zoltán Vámossy, “Multi-

Directional Image Projections with Fixed Resolution

for Object Matching”, Acta Polytechnika Hungarica,

Vol. 15, No. 2, 2018

Ildar Z. Batyrshin ,“Constructing Correlation Coefficients

from Similarity and Dissimilarity Functions”, Acta

Polytechnika Hungarica, Vol. 16, No. 10, 2019

Baráth Dániel, Matas Jiri, “Gráf-vágás RANSAC”, KÉPAF

2019: Képfeldolgozók és Alakfelismerők Társaságának

12. országos konferenciája, 2019

Ziv Yaniv, Random Sample Consensus (RANSAC)

Algorithm, A Generic Implementation, Imaging

Science and Information Systems (ISIS) Center, Dept.

of Radiology, Georgetown University Medical Center,

Washington, DC, USA, 2010

Elan Dubrofsky, Homography Estimation, A Master’s

Essay Submitted in Partial Fulfillment of the

Requirements for the Degree of Masters of Science in

the Faculty of Graduate Studies, The University of

British Columbia, 2009

A. Ross and A. K. Jain, "Multimodal biometrics: An

overview," 2004 12th European Signal Processing

Conference, Vienna, 2004, pp. 1221-1224.

M. Golfarelli, D. Maio and D. Malton, "On the error-reject

trade-off in biometric verification systems," in IEEE

Transactions on Pattern Analysis and Machine

Intelligence, 1997, vol. 19, no. 7, pp. 786-796

Huda, Syamsul; Yusuf, Enjang Burhanudin: Biometrics in

the Quran Perspective: Definition, History, and

Type. Didaktika Religia, December 2019, pp. 276-298

G.S. Badrinath, Phalguni Gupta: Palmprint based

recognition system using phase-difference information,

Future generation Computer Systems, Vol. 28 No. 1,

January 2012, pp. 287-305

Wei Li, Bob Zhang, Lei Zhang and Jingqi Yan: Principal

Line-Based Alignment Refinement for Palmprint

Recognition, IEEE Transactions on Systems, Man, and

Cybernetics, Part C (Applications and Reviews), Vol.

42, No. 6, November 2012, pp. 1491-1499

De-Shuang Huang, Wei Jia, David Zhang: Palmprint

verification based on principal line, Pattern

recognition, Vol 41. No. 4, April 2008, pp. 1316-132

Biometric Authentication System based on Hand Geometry and Palmprint Features