Performance Comparison Of Biometric System
With And Without Evcs
Study Case For Palmprint Authentication System
Rosmawati Dwi1,2
Engineering and Information Technology,
Swiss German University – Master of IT Programme1
Tangerang Banten, Indonesia 15143
rosmawati.dwi@gmail.com1
Information Systems, STMIK Raharja2
Tangerang Banten, Indonesia 15119
rosmawati.dwi@raharja.info2
Abstract—Biometric nowadays has becomes popular in term
of authentication system. Many institutions and companies start
to use biometric in addition using password or smartcard to
authenticate a person in order to prevent unauthorized person to
access their system. As the development of biometric in
authentication system, the attackers try to search the
vulnerability of it in order to have the authority into the system.
One of the vulnerability of the biometric system that the attacker
addressed is in the biometric database. Many system of biometric
still save the biometric template characteristic in database in the
form of plain image. There is no encryption method used to
secure the biometric database, which makes easier for the
attacker to sabotage the biometric system. This research focuses
on how to secure the biometric database using EVCS (Extended
Visual Cryptography Scheme). The study is applied on 198
training set palmprint images and 10-validation set of palmprint
images. Experiment results showed that EVCS could be used for
securing biometric template database without losing much
information and the quality of the decrypted image reach 40.95
dB. The authentication system with EVCS also successfully
authenticated decrypted image with the percentage of true
positive that is 82.45% compared with the one without EVCS that
reach 76.89% respectively.
Keywords— EVCS, Biometric, Palmprint, Security,
Authentication.
I. INTRODUCTION
Biometrics is unique physical or behavioral
characteristics, which can be used to automatically identify
individuals. Biometric technologies capture, process and
measure these characteristics electronically and compare
them against existing records to create a highly accurate
identity management capability. It has been popularly used
as a technique for authentication system from several other
ways of authentication; using what we have, what we know,
who we are and somebody we know [1]. Palmprint has been
used as an alternative biometric instead of fingerprint due to
the easy pattern to captured and extracted. Various methods
have been researched such as Classifier for palmprint
recognition which using unsupervised technique 2DPCA
(Two Dimensional Principal Component Analysis) where
features are extracted using 2DLDA (Two Dimensional
Linear Discriminant Analysis)[2]. Another research, use
Double-line-single-point (DSLP) for palmprint acquirement
in mobile platform [3] and did a palmprint acquisition system
based online scan sensor [4].
Lukas3,
Cognitive Engineering Research Group (CERG),
Electrical Engineering Dept, Faculty of Engineering3
Universitas Katolik Indonesia Atma Jaya, Jakarta,
Jakarta, Indonesia 12930
lukas@atmajaya.ac.id3
As Biometric technology develops and the use of
biometric technology becomes popular, there exist several
possible attacks that become issues in securing biometric
system. Below is the diagram of possible attack points in
Biometric System [5]:
Fig.1 Possible Attack Points in Biometric System [5]
1. Involves presenting fake biometric
2. Replay attack
3. Feature extractor module is replaced with a Trojan
horse program
4. Genuine feature values are replaced with values
selected by attacker
5. The matcher is replaced with Trojan horse
6. Template database attack
7. Templates are replaced or altered in transition medium
between template and database
8. The matcher result can be overridden by attacker
The paper is focus on point 6; how to counter the template
database attack using Extended Visual Cryptography
Scheme.
Visual cryptography (VC) is a secret sharing technique,
which allows encryption of a secret image among a number
of participants [6]. Different from any traditional
cryptography methods like AES, DES, 3DES, RSA, etc that
needs complex mathematical computation, VC only use
simple image manipulation. Particularly in a k-out-of-n
visual secret sharing scheme (VSS), a secret image is
cryptographically encoded into n shares. Each share
resembles a random binary pattern. Extended Visual
Cryptography Scheme (EVCS) is a visual cryptography that
is combined with steganography. Extended visual
cryptography decodes images with no particular
cryptographic calculation where three images are input and
generate two images (are called sheets), which correspond to
two of the three input images. The third image (is called
target) is reconstructed by printing the two output images
onto transparencies and stacking them together [7]. VC
(Visual Cryptography) and EVCS can be implemented in any
application that used image, such as for Banking Application
using a (2,n) Visual Cryptography technique [8], for data
Fig. 2. EVCS Palmprint Recognition steps
matrix codes like 2D barcodes using VC [9], and for
financial document [10]. From the Fig. 1 above, there are two steps of palmprint
This paper proposes EVCS as a technique to secure recognition, such as:
palmprint authentication system using standard algorithm of 1. Enrollment consists of preprocessing, feature extraction,
visual cryptography and added with the modification of EVCS encryption and embedding with cover images,
EVCS that using color images as a camouflage instead of save result images in database.
gray images. Several studies of Visual cryptography in
biometric systems are the using of visual cryptography 2. Authentication system consists of EVCS decryption,
method in securing fingerprint authentication for ID card comparing image from database with image from input
[11], visual cryptography in securing iris authentication [12], system.
visual cryptography in securing palmprint authentication
A. Preprocessing
based on DCT algorithm [13], and visual cryptography in
securing face authentication [14]. Basically, the four of them Preprocessing step as shown in Fig. 3 is a process to get
applied visual cryptography algorithm in their system that the template of the image or data after the image was
raise suspicion of the image may have hidden information. enhanced for the excellent quality of an image. The steps are
Therefore, the use of EVCS in palmprint biometric system is normalization of image position, finding palmprint Region of
proposed to secure authentication system without any Interest (ROI), and edge detection and dilation.
suspicion of random dots images like in the visual
cryptography. This research also evaluates the performance Preprocessing
of the system using EVCS compared with the system without 1. Normalization
EVCS. a. Image framing
II. EXPERIMENT DATASET b. Find degree of slopes of hand
c. Rotate image
The samples of this research used dataset from a database 2. Finding ROI of Image
of Biometric Ideal Test from Prof. Tieniu Tan's research
a. Find Horizontal and vertical line of hand
group at the center of Biometric Security Research, National
Laboratory of Pattern Recognition (NLPR), Institute of b. Find border ROI
Automation, Chinese Academy of Sciences (CASIA)[15] c. Crop image
that consists of 312 different person’s palmprint. Each 3. Enhance image Quality
person’s palmprint were captured 8 samples from left hand a. Contrast Enhancement
and 8 samples from right hand. Using CMOS camera, the b. Edge detection
samples were taken invariant angle (00 – 450), the size of c. Dilation
samples are 480x640 pixels that contains 8 bit gray scale
images. From 312 persons, 198 are above 17 years old that
consists of both female and male.
Fig. 3. Preprocessing step in Palmprint Recognition using EVCS
The hardware and software that were used for this
experiments are computer with processor Intel Core i5- 1. Normalization of image position
4210u CPU @1.70Hz 2.40 Hz with RAM 16 GB and Matlab
version R2015a as a tool for simulation system and image
processing.
III. EVCS PALMPRINT RECOGNITION STEPS
1. Palmprint Enrollment

2. Palmprint Authentication System Fig. 4. Original sample CASIA palmprint images in variant angle.

There are three steps of normalization. First, original

image of CASIA palmprint with different gradient as
seen in Fig. 4. To make samples consistent, normalize the
image by making a frame of hand as shown in Fig. 5(a).
 Then, from the bottom-left corner (for left hand image),
increment the pixel until find the first white pixel. Those
points marked as point A and B respectively as shown in
Fig. 5(b). From those two points, draw the line (yellow
line) and find the angle between yellow line and
horizontal line using equation (1) where v1 is vector of
yellow line, v2 is vector of horizontal line and θ is the
angle of the hand. Last, rotate the image as far as θ
degree approaching the horizontal line like in the Fig.
5(c).
θ = cos -1 ((v1.v2)/(|(v1||v2|)) (1)
(a) (b) (c)
Fig. 5. (a) A frame of hand, (b) frame of hand with line of slope, (c)
frame of hand after rotated
2. Finding palmprint’s Region of Interest (ROI)
Fig. 6. Summarize graph of pixel vertically (up) and horizontally
(down)
(a) (b) (c)
Fig. 7. Horizontal and vertical line (a) ,the ROI border (b), original
cropped image (c)
Focus area of image that is extracted is the palm area
without finger. There are three steps of finding palmprint’s
ROI; find horizontal center line and vertical border line, find
border box line, crop image.
Before cropping the palm area, first summarize pixel of
image vertically and horizontally that can be seen in Fig. 6.
In the vertical graph, it can be seen that the threshold of the
value of vertical borderline position is where the value is 40
and position is greater than 380 for left hand. While for the
horizontal centerline in the range of pixel 230 until 250
(about half of the horizontal size image). In that range, the
horizontal centerline is found where the summarization of
horizontal pixel value is not zero as shown in Fig. 7(a).
Then, to create ROI border of 300 x 300 pixels, there are
four lines that should be drawn. First, subtracting horizontal
centerline with 150 to draw horizontal above line. Second,
adding horizontal centerline with 150 to draw horizontal
below line. Third, subtracting vertical borderline with 330
for vertical left line and with 30 for vertical right line. The
results of borderline can be seen in Fig. 7(b). After that, crop
the original normalized image using that border. The result
image can be seen in Fig. 7(c).
3. Enhance Image Quality
There are three steps of enhancing the quality of image
before next process, which is feature extraction, such as
Enhance the contrast, Edge detection, and Dilation – erosion
process.
Contrast enhancement in this image is needed because
the principle lines of the palmprint still not firm and clear.
By using histogram equalization, the contrast of image can
be improved. Fig. 8(a) shows the graph of histogram before
normalized where the values of pixels were concentrated
between 125 and 190. To normalize the histogram, the
formula (2) is used to increase the contrast of image.
cdf(v)- cdfmin
h(v) = round ( (MxN) - 1
x (L –1)) (2)
Attributes M and N are row and column of the image;
meanwhile L is the depth of image in each pixel (for grey
scale L=256). Cdf min in that equation is the minimum non-
zero value of cumulative distribution function. After being
adjusted, the value of pixel spread equally in histogram as
shown in Fig. 8(b). The result of contrast enhancement
image can be seen in Fig. 9(b) where the principle lines of
palmprint is bolder and clearer than the original image (Fig.
9(a)).
(a) (b)
Fig. 8. Histogram of original image (a), image after contras
enhancement (b)
(a) (b)
Fig. 9. Cropping of original image (a), Contrast enhanced image (b)
The next step of quality enhancement is edge detection.
The experiment used canny edge detector to get more detail
of principle lines than other kind of edge detector. Fig. 10(a) -44 -45 -33 308 612 0 11043 147 522
shows the results of canny edge detection from Fig. 9(b). To
gain better quality of three principle lines of palmprint, the -43 -45 -34 194 266 46 10326 266 807
next step is to dilate the image using equation (3) below. The 44 45 42 220 369 1223 3080 757 109
attribute of the equation such as S for structure element, f is
44 45 42 529 406 1008 2362 775 173
the image, Df is the domain of original image, then Db is the
domain of result image after dilation. From the dilation 43 45 42 827 457 1083 2140 780 329
process using some Structure Element, like rectangular and 43 45 43 799 486 1005 3772 528 110
circle, the result image become as shown in Fig. 10(b) and
were used as template images for feature extraction.

TABLE II. EIGHT RECORDS OF THE NINE LEFT AND RIGHT PALMPRINT
f⨁S = max {f (s-x) + b(x) | (s-x) ∈ Df, and x ∈ Db } (3) IMAGE'S FEATURES OF SAMPLE ID NUMBER 20 WITH EVCS

Angle AngleAngleRegionRegionRegion Region RegionRegion

A B C Ia Ib IIa IIb IIIa IIIb
-44 -45 -33 369 486 0 9547 180 437
-42 -43 -33 375 397 0 6412 169 566
-43 -44 -33 187 292 0 10602 143 709
-43 -45 -33 156 227 46 9217 224 667
44 45 42 179 324 1086 2582 684 96
44 45 42 451 374 876 2134 695 145
43 45 43 610 386 977 3157 465 99
43 45 42 710 425 985 2113 680 285
(a) (b)
Fig. 10. (a) Edge detection result image, (b) Image template after
dilation C. EVCS encryption
B. Feature extraction
EVCS Encryption
Feature extraction in this experiment consist of nine
element of features of the palmprint images that were
extracted, such as three angles between principle lines (angle 1. Generate random number as much as image size
A, B, and C in Fig. 11) and six region (Ia,Ib, IIa, IIb, IIIa, 2. Half toning image for each pixel
IIIb). Those six regions were separated by the mainline (line 3. Substituting pixel with one combination shares
from a to b), midline (line from o that is perpendicular with 4. Embedding shares with cover image
the mainline), and three principle lines.
In table I and table II below, there are samples of Fig. 12. Preprocessing step in Palmprint Recognition using EVCS
palmprint’s features without EVCS and samples of
palmprint’s features with EVCS respectively. 4 first records EVCS encryption in this research was used to overcome
are from left palmprint and the rest are from right palmprints. the issue of template database attack. Two share EVCS were
As we can see, between feature extraction of palmprint used in this experiment due to the faster and efficient process
without EVCS and with EVCS have similar value for the of EVCS. The steps of the EVCS like in Fig. 12 are as
respective samples. follows:
a. Generating random integer as much as 300 x 300 pixels,
which is the size of template image.
b. Halftoning process of pixel in the template image. Fig. 13
shows the six possible scheme of pixel halftone for two
shares. Since we choose of using two shares, it means
that each result shares will have 600 x 600 subpixels that
consist of half value of pixel from template image.
c. Substituting each original pixel of each shares with one
of the six schemes as shown in Fig. 13 randomly. Fig. 14
shows the result image of visual cryptography process
using 2 x 2 shares.
d. Embedding each result shares with random cover images
where Fig. 15 shows two examples of those random
Fig. 11. Features of palmprint[16] images. In Fig. 16, there are results embedded share
images.
TABLE I. EIGHT RECORDS OF THE NINE LEFT AND RIGHT PALMPRINT
IMAGE'S FEATURES OF SAMPLE ID NUMBER 20 WITHOUT EVCS

Angle AngleAngleRegionRegionRegion Region RegionRegion

A B C Ia Ib IIa IIb IIIa IIIb
-43 -44 -33 277 240 0 11985 250 822
-43 -45 -33 422 439 0 7508 195 675
 this experiment MAX = 256. From those formulas, we can
see the MSE and PSNR value of 8 records (4 records are
taken from left hand and the others are from right hand) from
sample 20 in Table III below. The formula also applied for
the rest 197 samples and we got average MSE value 0.017
and average PSNR value reach 40.95 dB.

MSE ∑m-1 n-1

i=0 ∑j=0 I(i,j)-K(i,j) (4)
mn

MAXI 2
PSNR=10. log10
MSE

=20. log10 (MAXI ) -10. log10 (MSE) (5)

The lowest value of MSE indicated better quality of

Fig. 13. Six schemes of pixel halftone for two shares of palmprint image
encryption
image. From this experiment, we got 0.017 for MSE score,
which means decrypted EVCS image is similar with the
original template image and has little error from the
decryption process. In the opposite of MSE, higher value of
PSNR indicated better quality of image. So, from the
experiment, it is measured that the average PSNR reach
40.95 dB, which indicated the images have good quality.

TABLE III. EIGHT RECORDS OF MSE AND PSNR OF LEFT AND RIGHT
PALMPRINT IMAGE'S FEATURES OF SAMPLE ID NUMBER 20

Left/Right Hand MSE PSNR

Left 1.783800 105.0378

Fig. 14. Share 1 (left) and share 2 (right) for image size 600 x 600 pixels
Left 1.936311 104.2174
Left 1.839511 104.7303
Left 1.711489 105.4516
Right 2.005767 103.8650
Right 1.929589 104.2522
Right 2.127811 103.2743
Right 2.197678 102.9513

Fig. 15. Original cover images 2. Authentication System Performance

To measure the performance of the system, we use
Euclidean Distance. Euclidean distance equation as in
Equation (6), is used to measure distance between any two
points in space corresponds to the length of straight line
between them. From 198 dataset, we took each sample 4
records as training sample. Then the rest become testing set.
After we measure all of the sample records, we define the
threshold. From that threshold, each sample was evaluated
whether the distance is below or above the threshold. The
distance below the threshold indicated more similarities and
the records were marked as “TRUE”.

Fig. 16. Embedded share image result 2

E(x,y)= ∑ni=0 xi -yi (6)
IV. PERFORMANCE EVALUATION P
S%= *100% (7)
In this experiment, there are two points that were N

evaluated; the quality of image after EVCS encryption and
the performance of the authentication system with EVCS
compared with the system without EVCS encryption.
1. Quality of Image
For the quality of image after being encrypted using
EVCS, we use MSE and PSNR equation as seen in equations
(4) and (5) respectively. Where m and n are the length and
wide of image respectively, I is the original image, K is
result image, and MAX is the maximum value of pixel. In
Then, to measure the performance of the system, we use
equation (7) where P is number of positive (TRUE) value, N
is total number of record that being evaluated (198 samples),
and S is successful rate (%). In this experiment, it is found
that the score of system without EVCS: P= 609, and N = 792
(198 x 4) and S = 76.89%, while the score of system with
EVCS: P = 653, N = 792 (198 x 4) and S = 82.45%. From
the results, we can say that the performances of both system
with and without EVCS are good enough, even the system
with EVCS can gain higher performance due to the use of
quality enhancement in the process of EVCS encryption and
decryption, which means the system with EVCS can become
solution of securing image database from alternation without
losing the performance of the system.
V. CONCLUSION
EVCS (Extended Visual Cryptography Scheme)
successfully secured 198 CASIA palmprint images database,
without losing much information and the quality of the
decrypted image reach 40.95 dB and successfully
authenticate decrypted image with successful rate reach
82.45% compared with the system without EVCS that only
reach 76.89%.
The EVCS can be used in the palmprint recognition
without losing much information of image. The steps upon
198 CASIA palmprint image database consist of 3 stages
such as preprocessing, feature extraction and EVCS
encryption. The experiment showed that MSE and PSNR
scores are 0.017 and 40.95 dB respectively, which means the
decrypted EVCS image has little error and similar to the
original image. Further, PSNR score of 40.95 dB indicates
that decrypted EVCS image has a good quality compared
with original template image.
REFERENCES
[1] Brainard, John, Ari Juels, Ronald L. Rivest, Michael Szydlo, and
Moti Yung. "Fourth-factor authentication: somebody you know." In
Proceedings of the 13th ACM conference on Computer and
communications security, pp. 168-178. ACM, 2006.
[2] I.Rida, R.Herault, G.L.Marcialis and G. Gasso, “Palmprint
recognition with an efficitn data driven ensemble classifier”, Pattern
Recognition Letters, In Press, April 2018.
[3] L.Leng, F. Gao, Q.Chen and C. Kim, “Palmprint recognition system
on mobile devices with double-line-single-point assisstance”, Pers
Ubiquit Comput, vol. 22, issue 1, pp. 93-104, 2018.
[4] D.Zhang, G. Lu and L. Zhang, “Line Scan Palmprint Recognition
System”, Advanced Biometrics, Springer,Cham, pp. 235-257, 2018.
[5] Ratha, Nalini K., Jonathan H. Connell, and Ruud M. Bolle. "An
analysis of minutiae matching strength." In International Conference
on Audio-and Video-Based Biometric Person Authentication, pp.
223-228. Springer, Berlin, Heidelberg, 2001.
[6] Liu, Feng, and Wei Qi Yan. Visual Cryptography for Image
Processing and Security. Vol. 2. New York: Springer, 2014.
[7] M. Nakajima and Y. Yamaguchi, “Extended Visual Cryptography for
Natural Iamges”, Journal of WSCG, vol. 10, no.1-2, pp. 303-310,
2002.
[8] J. K. Pal, J. K. Mandal and K. Dasgupta, “A (2, N) Visual
Cryptographic Technique for Banking Applications”, Int. Journal of
Network Security and Its Applications (IJNSA), Vol.2, No.4, October
2010.
[9] Sharma, M. Agnihotra, and M. Chinna Rao. "Visual cryptography
authentication for data matrix code." International Journal of
Computer Science and Telecommunications 2.8 (2011): 58-62.
[10] Hawkes, L., Alec Yasinsac, and C. Cline. "An application of visual
cryptography to financial documents." Florida State University,
Florida (2000): 1-7.
[11] Rao, YV Subba, Yulia Sukonkina, Chakravarthy Bhagwati, and
Umesh Kumar Singh. "Fingerprint based authentication application
using visual cryptography methods (improved id card)." In TENCON
2008-2008 IEEE Region 10 Conference, pp. 1-5. IEEE, 2008.
[12] Revenkar, P. S., Anisa Anjum, and W. Z. Gandhare. "Secure iris
authentication using visual cryptography." arXiv preprint
arXiv:1004.1748, 2010.
[13] Divya, C., and E. Surya. "Visual cryptography using palm print based
on dct algorithm." International Journal of Emerging Technology and
Advanced Engineering 2, no. 12, 2250-2459, 2012.
[14] Akotkar, Atul Sureshpant, and Chaitali Choudhary. "Secure of Face
Authentication using Visual Cryptography." International Journal of
Innovative Science and Modern Engineering (IJISME) ISSN, 2319-
6386, 2014.
[15] T. Tan and Z.Sun, “Note on CASIA Palmprint Iamge Database”,
Biometrics Ideal Test, access on 18 May 2018 from:
http://biometrics.idealtest.org/dbDetailForUser.do?id=5
[16] D.D.Zhang, “Palmprint Authentication”, Kluwer Academic Publisher,
Boston, USA, 2004.