Available online at www.sciencedirect.

com

ScienceDirect
Available online at www.sciencedirect.com
Procedia Computer Science 00 (2017) 000–000
www.elsevier.com/locate/procedia
ScienceDirect
Procedia Computer Science 115 (2017) 482–492

7th International Conference on Advances in Computing & Communications, ICACC-2017, 22-

24 August 2017, Cochin, India

Multi-Algorithm of Palmprint Recognition System Based on Fusion

of Local Binary Pattern and Two-Dimensional Locality Preserving
Projection
Mouad M.H.Alia , *, Pravin L.Yannawarb A. T. Gaikwadc
a
Research Scholar at Department of CS & IT, Dr. Babasaheb Ambedkar Marathwada University, Aurangabad, (M.S),431001, India
b
Vison and Intelligent System Lab, Department of CS & IT, Dr. Babasaheb Ambedkar Marathwada University, Aurangabad,(MS) India
c
Institute of Management Studies and Information Technology , Aurangabad, (M.S)431001, India

Abstract

This work based on fusion of multi-algorithm used for palmprint identification system. This piece of work is primary
addressing different mechanism like Competitive Valley Hand Detection methods (CHVD) which used for extract
Region of Interest (ROI). While in feature extraction the work was divided into three scenarios based on feature
extraction like Local Binary Pattern (LBP), Two-Dimensional Locality Preserving Projection (2DLPP) and fusion of
LBP+2DLPP.The experimental results show that fusion of LBP and 2DLPP give the best result with high accuracy
reach to 98.55% which improve by 1.22% and 2.85% for LBP and 2DLPP respectively when they are applied
separately.

© 2017 The Authors. Published by Elsevier B.V.

Peer-review under responsibility of the scientific committee of the 7th International Conference on Advances in Computing &
Communications.

Keywords: Biometrics; Palmprint; ROI; LBP; 2DLPP; Fusion; Identification;

* Mouad M.H.Ali. Tel.: +919823235210.

E-mail address: mouad198080@gmail.com

1877-0509 © 2017 The Authors. Published by Elsevier B.V.

Peer-review under responsibility of the scientific committee of the 7th International Conference on Advances in Computing &
Communications.

1877-0509 © 2017 The Authors. Published by Elsevier B.V.

Peer-review under responsibility of the scientific committee of the 7th International Conference on Advances in Computing &
Communications
10.1016/j.procs.2017.09.091
 Mouad M.H. Ali et al. / Procedia Computer Science 115 (2017) 482–492 483
2 Mouad.M.H.Ali et al. / Procedia Computer Science 00 (2017) 000–000

1. Introduction

Today’s the technology is grow day by day, in another side the security system also is increase related to the
technologies. The biometrics recognition system is active research work nowadays, which including many
biometrics characteristics such as (biological and behavior)[1],[2]. Conventional authentication methods such as
passwords, PINs, tokens, and smart card no relevant for application on systems that require security high .The
biometrics system replacing conventional methods by utilizing physical characteristics or behavior characteristics of
human that actually represent a person's identity and advantages that are difficult to duplicate, stolen , and
falsified[3]. There are three main challenges facing the biometric system [4], namely accuracy, scale, and usability.
Various proposed ways to improve the accuracy of biometrics systems such as by combining more than one
biometric characteristic for the introduction or referred to as multimodal biometric system [5]. Here let us first
definition of palmprint which defines as a small area of palm surface which containing more information which is
useful for person authentication system, in additional it has a unique feature (uniqueness means no two people has
same this feature) also it called permanence it will not change in all period of time in the life. For this reason,
palmprint are reliable and confident modality between the same categories of palmprint like fingerprint and face
.etc.In the recent years, there are numbers of technologies were developed related to biometrics authentication
system but the palmprint get less development depend on reliability and cost [6]. The palmprint approach can be
classified into three categories depend on the palmprint image data type such as grayscale[7],[8], 3D[9]and
multispectral. There are many of researchers working in grayscale image compare with the less researcher working
in 3D and multispectral palmprint images. Recently the multispectral data are used in many areas such as face[10]
,iris[11] and palmprint[12]. The palmprint recognition system is most confident and reliable system compare with
another biometrics modalities, also comparing with fingerprint, palm has a lot of feature such as minutiae feature
which is a similar to fingerprint. The second feature is principle line feature which includes three types of lines:
heart, head and life. Also it has texture feature. Furthermore there are geometry, wrinkle and Delta feature [1]. So
this many features are inside the palm area .There are many problem of palm print like Skin distortion, Diversity of
different palm regions and Computational complexity [13]. The increasing in security system the palmprint
recognition system has applied with different feature extraction techniques, also with different results which are
improving the performance of palmprint identification and verification system. There are various techniques of
feature extraction is proposed to improve the performance of biometric systems among which the LDA, PCA, ICA,
LBP, and the LDP [14],[15],[16]. This feature extraction technique are classify into 5 categories like local
feature[17],[18],[19],[20],[21],statistical feature[22],[23],[24],25],[26], appearance feature[27],[28], [29],30],[31],
texture feature [32],[33],[7],[34],[35],[36] and hybrid feature[37],[38]. The work concentrated on texture feature
specially in Local Binary Pattern (LBP) [39], and Two-Dimensional Locality Preserving Projection (2DLPP) for use
in identity recognition proven successful with high accuracy [40],[41]. The LPP techniques is one of appearance
based approach for biometrics system. And the main objective of LPP is to preserve the local structure of the image
space for this D.Hu et al[42] they are working on 2DLPP by extract the feature directly from images matrices, while
the X.He et al[43] they are used LPP for image feature extraction and dimension reduction and apply to face
recognition and called first one implemented this technique and they get good results. In general the 2DLPP
technique use to solve the generalized Eigen values problem [43], also in another hand the 2DLPP required more
coefficients which are one of the disadvantages of 2DLPP technique for image representation and recognition [44].
2DLPP can apply in column direction with help of 2DPCA projection in row direction [45] and some of related work
paper show in[45],[46],[47],[48]. This paper is arranged in four sections, the remaining sections will be explained
the methodology given in section 2. In section 3 the Experiment and result analysis are given. Finally the conclusion
and future work are given in section 4.

2. Methodology of the System

The system organized by different stages for identification proposed. These stages like acquisition stage, pre-
processing stage, feature extraction stage, fusion stage, matching stages and finally the decision stage which applied
over CASIA palmprint dataset. The figure 1 shows the block diagram of methodology process and the details of each
stage are discussed in the next section.
484 Mouad M.H.etAli
Mouad.M.H.Ali al. et al. / Procedia
/ Procedia Computer
Computer Science
Science 115 (2017)
00 (2017) 482–492
000–000 3

Fig.1. Block diagram of methodology of Palmprint Recognition

2.1. Image Acquisition stage

It is the first step of any biometric system, which is the answer of many questions like how can we get the
palmprint image? , which devices used to capture the palm image? and what are the characterizations of these
devices? Here the data collected from CASIA dataset which contain 7200 images for both right and left hands.

2.2. Pre-processing stage

The pre-processing steps are very important steps for building any biometric system (identification or
verification) which used to remove the noise and clarity of the palm image. The pre-processing of palmprint system
is the way to extract the very important parts in palm surface which include more information and called as Region
of Interest (ROI), which defines as a rectangle area on the palm. In the general, the main steps of palmprint pre-
processing start with converting the palm image into gray scale, enhancement, Binarization, and boundary detection;
detect the reference points, extreme point, valley point, scaling finally cropping the ROI. This all the steps needed to
do the pre-processing process on palmprint images. There are many methods to extract the ROI, such cropping
directly from palmprint image without applying any algorithm [47].The other method are called competitive Hand
valley Detection method (CHVD) and Euclidean distance method. The difference between them is the way to get the
reference point. In this study the CHVD method is used to extract ROI which derived from [49] to improve the
palmprint matching performance.

2.3. Feature Extraction Stage

The feature extraction is applied on the output of pre-processing phase which is a fixed size of the image to
extract the feature of palms. In this study there are two types of techniques are used for feature extraction namely
LBP and 2DLPP. The advantage of 2DLPP is the ability of taking a characteristic topology or information on the
local structure of an image with precision. The disadvantage of this algorithm is it has not resistant to variations in
lighting and orientation. The lighting variation overcome by image composition and normalized the ROI. The LBP
is a pioneer of local structural model quantization and the concatenate histogram statistics to generate feature vector
and well describe multiple interest region of an image.
 Mouad M.H. Ali et al. / Procedia Computer Science 115 (2017) 482–492 485
4 Mouad.M.H.Ali et al. / Procedia Computer Science 00 (2017) 000–000

2.3.1. Feature Extraction using LBP

LBP operator applied to extract the texture feature as well as the shape of grayscale image. It refers to a binary
code for an image-pixel and it provides us by information regarding the local neighbourhood of that pixel. The
principal LBP operator was given by Ojala et al[51]. The idea behind the LBP operator is to search a centre pixel
value of an image and take this value as threshold and check as below cases:
if neigborhood pixel  T hreshold
1    ;     
L B P _ B inary _ C ode  
0 ; O therw ise
If neighbor pixel has higher value or equal to centre pixel value this leads to give that pixel the value 1, otherwise it
takes zero. The example of LBP and how to calculate a binary code is represented in Fig.2.

Fig.2. Example of original LBP operator

After that the LBP was extend to utilize neighborhoods of various sizes. In such case a circle is produced
having radius R from the center pixel. P sampling points on the circle edge are determined and compared with center
pixel value to find the values of all sampling points in neighborhoods for any radius and any number of pixels. Fig.3
demonstrates three neighbor sets for various values of P and R. Additionally it known as the multi-scale LBP
feature.

Fig.3. Circular (8, 1), (8, 2), and (16, 2) neighborhoods

In the case of our work the LBP feature extraction is to calculate the LBP for every pixel in the image. This
occurs by divided the palmprint image into blocks or region. Here we divided image into different block size like
8x8, 16x16 and 32x32. To create the feature vector of the image, each block has histogram and by combining all
histogram for each block to create the LBP feature vector that represent the palmprin image Fig.4 shows the process
of our LBP technique.

Fig.4. LBP for Palm image and divided into regions

2.3.2. Feature Extraction using 2DLPP

The hallmark of the (Characteristic) palmprint is obtained by projecting the image using a transformation matrix
obtained from the 2DLPP algorithm. Representations characteristics 2DLPP results referred to Laplacian palm. The
transformation matrix obtained from 2DLPP process that is applied to a set of training images X = {x1, x2, ..., xN}
which N is the number of samples in the dataset. The algorithm of 2DLPP [40],[29] is described in flowchart as
486 Mouad M.H. Ali et al. / Procedia Computer Science 115 (2017) 482–492
Mouad.M.H.Ali et al. / Procedia Computer Science 00 (2017) 000–000 5

show in figure 5.

Fig.5. Flowchart of 2DLPP Algorithm

Establishment of nearest-neighbor graph: Graph nearest-neighbour(S) is represented as a matrix in which each

element Sij-ij indicates the closeness between images i and j in a dataset. The vertex i and j are connected if the image
I is close to the neighboring point j using k- nearest neighbor (KNN).
Giving weight to the graph: If vertices i and j are not connected to each other, then the value Sij = 0. If i and j are
connected, the weight can be calculated in two ways: (i) simple-minded which give Sij = 1 and (ii) heat- kernel
which gives weight using the Eq. (1).
2

xi - x j
t (1)
S ij
 e
Where xi and xj are two images of observation and t is a constant that has been determined. Output of this phase is
the similarity matrix is symmetric (S) dimension N x N.
Resolution on the problem of generalized: Eigenvalue (Eigen map). At this stage, eigenvector and eigenvalue are
calculated from the following Eqs. (2, 3 and 4):

XLX T w                              
 XDX T w (2)

Dij  ij Sij (3)

(4)
LD–S
Where X is a collection of images in a dataset matrix, D is the vector sum of the column or row of S. S is symmetric.
L is the Laplacian matrix reduction result matrix with the vector S. The equation (2) will produce a solution matrix
w = {w1, w2, ..., wd} which is an eigenvector corresponding to Eigen values in ascending sequences.
Projection: Projection done by performing matrix multiplication between the matrix image matrix equations W
which results from equation (2) to get feature vector.
xi 
 Yi X
iW , i 1, 2, , N (5)
Vector line dimensionless from (Yi) is the feature vector that represents the image of Xi. The Fig.6 shows
Laplacianpalm of some palmprint samples.

Fig.6. Representation Laplacianpalm from some samples

 Mouad M.H. Ali et al. / Procedia Computer Science 115 (2017) 482–492 487
6 Mouad.M.H.Ali et al. / Procedia Computer Science 00 (2017) 000–000

2.3.3. Fusion of LBP and 2DLPP

In this proposed system, the fusion is used for LBP and 2DLPP, where the entire feature from different algorithm
applied to palmprint dataset and combining them to improve the accuracy of the system. Here we used simple
feature fusion called as concatenated. The process of concatenated is to take the feature which generated by LBP
with the feature generated by 2DLPP in to one feature vector and apply feature reduction technique using PCA then
feature selection technique using LDA and store the new features as template in database for matching purpose.

2.4. Matching stage

The matching is done by comparing input image (Query) with the template store in database which taken at time
of enrollment and compute the degree of similarity or dissimilarity from two templates. To achieve the score the
Euclidean distance measure is used to compute the similarity or the difference between the templates, The Euclidean
distance which gives in Eq. (6) is used to calculate the distance of two feature vector fv1 and fv2:
 d ( f v1, f v 2 ) 
 nN1 fv n1  fv n 2  2
(6)
Where, N is determining the number of feature in fv1 and fv2. In the case of identification, the result is "known"or
"unidentified" depend on threshold value. The threshold () used in our system is obtained from an array [50] by Eq.
(7).

max(TA ) - min (TA)

 = (7)

Threshold Array (TA) Euclidean distance is obtained from the distance between all the feature vector samples in
a dataset. If a dataset contains N individuals with each M samples then there are N pieces of (TA), each TA has M
pieces threshold.  is a constant predetermined which is used to divide threshold value into N parts. Then, N values
threshold will be tested to obtain the optimal values of FAR and FRR by Eq. (8)

1 min  TA  

2 min TA  2 (8)

N min TA  N

i (i = 1, 2,…,N) is selected when the value approaches the FRR or FAR value is very small50 depending on the
specifications required.

2.5. Decision stage

In the final step of propose of study is the decision of either “Accepted “or “Rejected “with help of threshold
value (T). ( 𝑖𝑓 𝑆 ≤ 𝑇 ) That means “Accept” but ( 𝑖𝑓 𝑆 > 𝑇 ) that means rejected. Where, S is the distance score.

3. Experimental Results

3.1. Database
The experiments are evaluated on Chinese Academy of Sciences' Institute of Automation (CASIA)Multi-
Spectral Database v1.0 palmprint database which has 8 bits gray level (JPEG file), image size (768x576). This
database contains 7200 images for 100 subjects for both left and right hands. For evaluate the algorithm the 100
subjects are selected for left hand with ID 001-100 and each subjects has 6 samples each labelled 01-06 and the
total dataset containing 1200 images. The experimental applied on the laptop Dell, Intel core i3, CPU 2.20 GHz with
RAM 8.00GB on 64-bit operating system (windows 7). The figure7 shows some samples taken from CASIA dataset
and their ROI part.
488 Mouad M.H. Ali et al. / Procedia Computer Science 115 (2017) 482–492
Mouad.M.H.Ali et al. / Procedia Computer Science 00 (2017) 000–000 7

Fig.7. Samples taken from CASIA dataset with extract ROI by CHVD

In this proposed system the pre-processing approach is used as similar to literature47 which used the different
steps like Binarization, boundary extraction, extract the ROI by CHVD method and finally the ROI with dimension
size (155x155) is cropped with help of reference point. The figure 8 shows the pre-processing steps.

Fig.8. Pre-processing steps

3.2. Evaluation of parameter

To evaluate any biometric system related to specific application there are different parameter namely False
Accepted Rate (FAR),False Rejected Rate (FRR) and Equal Error Rate(EER) These parameters should be the
lowest values to achieve the better performance of the system. The FAR which is the ratio of imposter score
exceeding the threshold values divided by all the imposter score generated by the system and calculated by Eq.(9).
Impostor Score exceeding thershold
FAR  100 (9)
All Impostor Score
While the FRR is the Ratio of genuine score falling below the threshold value divided by all the genuine score
generated by the system and calculated by Eq. (10).

G enu ine S cores falling below thershold (10)

FRR  1 00
A ll G enuine S cores
The EER can be calculated according to the Eq. (11)
F AR  F R R (11)
EER 
2
In addition to the above parameter there is Genuine Accept Rate (GAR) which shows the relation between
FAR and FRR with help of threshold value and the Receiver Operating Characteristic (ROC) curves shows the FAR
values which are changed related to GAR values and show the performance of the system. The GAR value is
calculated by Eq. (12).
G A R 1  F RR (12)

3.3. Experimental Result using LBP

For evaluate the palmprint system with LBP method, the score(S) and threshold (T) values are generated in
 Mouad M.H. Ali et al. / Procedia Computer Science 115 (2017) 482–492 489
8 Mouad.M.H.Ali et al. / Procedia Computer Science 00 (2017) 000–000

the matching stage. According to the threshold values, the score matrix can be divided into genuine score matrix and
Impostor score matrix (that means from same person or from different persons). In case of genuine score ( S ≤ T )
while in case of impostor score ( S > T ) and by the help of these two matrixes, the FAR and FRR can be calculated.
Fig.9. shows the relation between FAR vs. FRR based on threshold values (T). The system was achieved the best
result in EER equal to =0.043 and FAR,FRR equal to 0.040, 0.044 respectively. And the maximum GAR reach to
95.7%. Table (1) shows the performance of LBP with Histogram length of 2800 and feature length after apply PCA
method for dimensionality reduction and select the optimal feature by using LDA method is 753 feature points. The
(Fig.9a) depict the relation between FAR and FRR with the help of threshold value while (Fig.9b) depict the ROC
curve of FAR vs. GAR.

Table 1. Performance of the system based on LBP method on optimal threshold values

Database size Feature length T FAR (%) FRR (%) EER (%) GAR (%)
218.3551 0.001 0.456 0.229 54.4
235.0596 0.010 0.231 0.121 76.9
253.3551 0.015 0.220 0.118 78
1200 753
264.2073 0.019 0.071 0.045 92.9
281.6211 0.043 0.043 0.043 95.7
285.4711 0.040 0.044 0.042 95.6

a FAR
b 1

FRR 0.9
0.25
0.8

0.7
0.2
0.6
1-FRR (GAR)
FAR/FRR

0.15 0.5

0.4
0.1
0.3

0.2
0.05
0.1

0 0
200 220 240 260 280 300 320 340 360 380 0 0.005 0.01 0.015 0.02 0.025
Threshold FAR

Fig.9.The performance of LBP (a) Relation of FAR vs. FRR based on threshold value (b) ROC curve of FAR vs. GAR

3.4. Experimental Result using 2DLPP

Evaluation of the performance of 2DLPP system is similar to the process of LBP for generated the score
matrix, genuine score and Impostor score for calculated the FAR and FRR of the system which already discuss in
previous section. Figure 10 shows the result of the system with EER equal to 0.026 and threshold =295.2073 with
FAR and FRR equals to 0.030, 0.023 respectively. And the maximum GAR = 97.33%. The table (2) show the result
achieved by 2DLPP with feature length 942 after reducing the dimension and select optimal feature .The (Fig.10a)
depict the relation between FAR and FRR with help of threshold value while (Fig.10b ) depict the ROC curve of
FAR vs. GAR.
Table 2. Performance of the system based on 2DLPP method on optimal threshold values

Database size Feature length T FAR (%) FRR (%) EER (%) GAR (%)
218.3551 0.000 0.281 0.1405 71.9
291.0596 0.040 0.040 0.040 96
1200 942 295.2073 0.030 0.023 0.0265 97.33
286.7623 0.030 0.037 0.0335 96.3
281.6211 0.038 0.040 0.039 96
301.1312 0.040 0.031 0.0355 96.9
 Mouad.M.H.Ali et al. / Procedia Computer Science 00 (2017) 000–000 9
490 Mouad M.H. Ali et al. / Procedia Computer Science 115 (2017) 482–492

a FAR b 1

FRR 0.9
0.25
0.8

0.7
0.2
0.6

1-FRR (GAR)
FAR/FRR

0.15 0.5

0.4
0.1
0.3

0.2
0.05
0.1

0 0
200 220 240 260 280 300 320 340 360 380 0 0.01 0.02 0.03 0.04 0.05 0.06
Threshold FAR

Fig.10.The performance of 2DLPP (a) Relation of FAR vs. FRR based on threshold value (b) ROC curve of FAR vs. GAR

3.5. Experimental Result using Fusion of LBP +2DLPP

In this experiment, the feature vectors generated by LBP and 2DLPP are concatenated by simple feature
fusion and generate the fused feature vector of LBP +2DLPP and no need to normalize the feature of LBP and
2DLPP because both are from same domain. The process is carried out for evaluation the system as mention in the
previous section for generated the score matrix, then divided it to genuine score and impostor score with the help of
threshold value then FAR and FRR are calculated. Table (3) shows the performance of fusion of LBP+2DLPP
system. The system can achieve the best results related to the lower values of FAR, FRR and EER as 0.0221, 0.0145
and 0.0183 respectively. And the maximum GAR equal to 98.55%. The (Fig.11a) depict the relation between FAR
and FRR with help of threshold value while (Fig.11b) depict the ROC curve of FAR vs. GAR
Table 3. Performance of the system based on Fusion of LBP+2DLPP method on optimal threshold values

Database size Feature length T FAR (%) FRR (%) EER (%) GAR (%)
235.0596 0.000 0.201 0.1005 79.9
253.3551 0.000 0.110 0.055 89
264.2073 0.021 0.052 0.0365 94.8
1200 1695
301.1312 0.040 0.031 0.0355 96.9
318.7761 0.020 0.020 0.020 98
321.4321 0.0221 0.0145 0.0183 98.55

a FAR b
1

FRR 0.9
0.25
0.8

0.7
0.2
0.6
1-FRR (GAR)
FAR/FRR

0.15 0.5

0.4
0.1
0.3

0.2
0.05
0.1

0 0
200 220 240 260 280 300 320 340 360 380 0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1
Threshold FAR
Fig.11. performance of Fusion (LBP + 2DLPP) (a) Relation of FAR vs. FRR based on threshold (b) ROC curve of FAR vs. GAR

Finally, table (4) show the final result of the system while table (5) comparing between propose system and
existing system on different dataset size and feature length.

10 Mouad M.H.
Mouad.M.H.Ali Ali/ Procedia
et al. et al. / Procedia Computer
Computer ScienceScience 115000–000
00 (2017) (2017) 482–492 491

Table 4. Performance of the system Comparison of the top recognition accuracy (%)

Database size Feature Extraction Feature length FAR (%) FRR (%) EER (%) GAR (%)
1200 LBP 753 0.043 0.043 0.043 95.7
1200 2DLPP 942 0.030 0.023 0.0265 97.33
1200 LBP+2DLPP 1695 0.0221 0.0145 0.0183 98.55

Table 5. Comparison of the recognition accuracy (%) and feature length on different Database size
Methods Database size Feature length Time (s) GAR (%)
LBP + Original Image [52] 5021 59 -------- 82.32
PCA + LPP [53] 1730 85 0.4844 93.178
2DLPP [53] 1730 640 0.7760 93.814
I2DLPP [53] 1730 50 0.2109 95.772
Proposed LBP 1200 753 0.5794 95.7
Proposed 2DLPP 1200 942 0.6172 97.7
Proposed LBP+2DLPP 1200 1695 0.8727 98.55

4. Conclusion

In this work, the biometric system for person recognition is designed based on fusion of multi-algorithm used for
feature extraction techniques like LBP and 2DLPP. The system used CHVD algorithm to extract the ROI of
palmprint image. While LBP and 2DLPP uses to utilize the texture feature of palm print image over CASIA Dataset.
The work is divided into three scenarios like LBP, 2DLPP and fusion of LBP+2DLPP. In the case of first scenario
LBP is resulted in 95.7% Genuine Accept Rate (GAR) with the same value for FAR, FRR and EER equal to 0.043.
The second scenario 2DLPP is resulted in 97.33% Genuine Accept Rate (GAR) with 0.030 FAR, and 0.023 FRR
with EER equal to 0.0265. Finally the third scenario is fusion of LBP with 2DLPP and is resulted in 98.55%
Genuine Accept Rate (GAR) with FAR = 0.022, FRR = 0.0145 and EER equal to 0.0183. It is conclude that the
fusion of LBP+2DLPP were applied on CASIA Dataset is achieved best result and the GAR improve by 1.22%
comparing with LBP and 2.85% comparing with 2DLPP separately. The future work may extend to apply palmprint
multi-feature with different level of fusion to improve the performance of the system.
References
[1]. Jain, A. K., P. J. Flynn, and A. Ross. Handbook of Biometrics. Springer. 2007
[2]. Ali, Mouad MH, and A. T. Gaikwad.: Multimodal Biometrics Enhancement Recognition System based on Fusion of Fingerprint and
PalmPrint: A Review. Global Journal of Computer Science and Technology 2016;16( 2):13-26
[3]. Jain, A.K., Flynn, P., Ross, A.A, Handbook of Biometrics. USA: Springer Science and Business Media, LLC, 2008
[4]. A. K. Jain, S. Pankanti, S. Prabhakar, Lin Hong and A. Ross, Biometrics: a grand challenge, Proceedings of the 17th International Conference
on Pattern Recognition, ICPR . 2004;2:935-942
[5]. Ashish Mishra, Multimodal Biometrics it is: Need for Future Systems, International Journal of Computer Applications. 2010;.4:28–33
[6]. Kumar A, Zhang D, Improving biometric authentication performance from the user quality. IEEE Trans Instrum Meas 2010;59(3):730–735
[7]. Kong A, Zhang D, Kamel M, Palmprint identification using feature-level fusion. Pattern Recogn 2006;39(3): 478–487
[8]. Wu X, Zhang D, Wang K , Palm line extraction and matching for personal authentication, IEEE Trans Syst Man Cybern Syst Hum
2006;36(5): 978–987
[9]. Li W, Zhang D, Zhang L, Three dimensional palmprint recognition, IEEE international conferenceon systems, man, and cybernetics, 2009;
4847–4852
[10]. Singh R, Vatsa M, Noore A, Hierarchical fusion of multispectral face images for improved recognition performance, Inf Fusion
2008;9(2):200–210
[11]. Park JH, Kang MG, Multispectral iris authentication system against counterfeit attack using gradient-based image fusion, Opt Eng .
2007;46(11):117003–1117003-14
[12]. Han D, Guo Z, Zhang D, Multispectral palmprint recognition using wavelet-based image fusion, In: Proceeding of the 9th international
conference on signal processing, Oct 2008; 2074–2077
[13]. Jifeng Dai, Jianjiang Feng.: Robust and Efficient Ridge-Based Palmprint Matching. 2012;34(8):1618-1632
[14]. Connie, T., Teoh, A., Goh, M., Ngo, D.,Palmprint Recognition with PCA and ICA, Conference of Image and Vision Computing New
Zealand (IVCNZ'03), 2003; :227-232
[15]. L. Mirmohamadsadeghi and A. Drygajlo, Palm vein recognition with Local Binary Patterns and Local Derivative Patterns, 2011
International Joint Conference on Biometrics (IJCB), Washington, DC, 2011: 1-6
[16]. Xiang-Qian Wu, Kuan-Quan Wang and David Zhang, Palmprint recognition using Fisher's linear discriminant, Proceedings of the
International Conference on Machine Learning and Cybernetics (IEEE Cat. No.03EX693), 2003;5: 3150-3154
[17]. W Shu, D Zhang.: Palmprint verification: an implementation of biometric technology in Pattern Recognition, Proceedings. Fourteenth
International Conference, 1998: 219–221
492 Mouad M.H. Ali et al. / Procedia Computer Science 115 (2017) 482–492
Mouad.M.H.Ali et al. / Procedia Computer Science 00 (2017) 000–000 11

[18]. D qingyun, Y yinglin, Z dapeng,: A line feature extraction method based on morphological median pyramid. J. South China Univ. Technol.
2001;29(5):14–18
[19]. Han, Chin-Chuan, Hsu-Liang Cheng, Chih-Lung Lin, and Kuo-Chin Fan. : Personal authentication using palm-print features." Pattern
recognition. 2003;36 (2):371-381
[20]. Noble, J. Alison. : Finding corners. Image and Vision Computing 1988;6(2): 121-128
[21]. L. Liu and D. Zhang.: Palm-line detection. IEEE International Conference on Image Processing 2005; III269-III272
[22]. J. y. Gan and D. p. Zhou.:A Novel Method for Palmprint Recognition Based on Wavelet Transform. 8th international Conference on Signal
Processing, Beijing, 2006;3.2006
[23]. Xue yuli.: Palmprint feature extraction based on low frequency distribution. Microcomputer & its Applications. 2011;20(1): 40-43
[24]. Raghavendra, Ramachandra, Bernadette Dorizzi, Ashok Rao, and G. Hemantha Kumar. : Designing efficient fusion schemes for
multimodal biometric systems using face and palmprint. Pattern Recognition 2011;44(5):1076-1088
[25]. Zhang, Yu, Dequn Zhao, Guangmin Sun, Qiang Guo, and Bo Fu.:Palmprint recognition based on sub-block energy feature extracted by real
2d-gabor transform. In Artificial Intelligence and Computational Intelligence (AICI), 2010;: 124-128.
[26]. Fahmy, Maged MM.: Palmprint recognition based on Mel frequency Cepstral coefficients feature extraction. Ain Shams Engineering
Journal. 2010;1(1):39-47
[27]. Lu, Guangming, David Zhang, and Kuanquan Wang.: Palmprint recognition using eigenpalms features. Pattern Recognition
Letters 2003;24(9):1463-1467
[28]. Sang, Haifeng, Weiqi Yuan, and Zhijia Zhang.: Research of palmprint recognition based on 2DPCA. In International Symposium on Neural
Networks, 2009: 831-838.
[29]. Zuo, Wangmeng, Kuanquan Wang, and David Zhang.: Bi-directional PCA with assembled matrix distance metric.IEEE International
Conference on Image Processing, ICIP 2005,2:958–961
[30]. Pan, Xin, and Qiu-Qi Ruan.: Palmprint recognition using Gabor feature-based (2D) 2 PCA. Neurocomputing 2008;71(13):3032-3036.
[31]. Lu, Guang-Ming, Kuan-Quan Wang, and David Zhang.: Wavelet based independent component analysis for palmprint identification.
In Machine Learning and Cybernetics, 2004. Proceedings of 2004 International Conference on, 2004;6:3547-3550
[32]. Xianji Wang, Haifeng Gong, Hao Zhang, Bin Li and Zhenquan Zhuang.: Palmprint Identification using Boosting Local Binary Pattern. 18th
International Conference on Pattern Recognition (ICPR'06), Hong Kong, 2006: 503-506
[33]. Kumar, Ajay, and Helen C. Shen. : Palmprint identification using PalmCodes. Third International Conference on Image and Graphics
(ICIG'04), 2004: 258-261
[34]. Zhenan Sun, Tieniu Tan, Yunhong Wang and S. Z. Li.: Ordinal palmprint represention for personal identification,IEEE Computer Society
Conference on Computer Vision and Pattern Recognition, 2005; 1:279-284
[35]. Wei, Jing, Wei Jia, Hong Wang, and Dan-Feng Zhu.: Improved competitive code for palmprint recognition using simplified Gabor filter.
In International Conference on Intelligent Computing, 2009: 371-377
[36]. He, Xiaofei.: Locality Preserving Projections. In the document dissertation for a degree of Doctor of Philosophy Chicago, Illinois.( 2005)
[37]. Khan, Zohaib, Ajmal Mian, and Yiqun Hu.: Contour code: Robust and efficient multispectral palmprint encoding for human recognition.
In IEEE International Conference on (ICCV), 2011: 1935-1942
[38]. Kumar, Ajay, and David Zhang. "Personal authentication using multiple palmprint representation." Pattern Recognition 2005;38(10):1695-
1704
[39]. Li, Wenxin, Jane You, and David Zhang.: Texture-based palmprint retrieval using a layered search scheme for personal identification. IEEE
Transactions on Multimedia 2005;7(5):891-898
[40]. S. Chen, H. Zhao, M. Kong, B. Luo, 2DLPP: a two-dimensional extension of locality preserving projections, Neurocomputing 2007;70 (4–
6): 912–921
[41]. J. G. Wang, W. Y. Yau, and A. Suwandy, Fusion of palmprint and palm vein images for person recognition based on laplacian palm
feature, Pattern Recognition, 2008;41(5):1514–1527
[42]. D. Hu, G. Feng, Z. Zhou, Two-dimensional locality preserving projections (2DLPP) with its application to palmprint recognition, Pattern
Recognition 2007;40: 339–342
[43]. X. He, S. Yan, Y. Hu, P. Niyogi, H. Zhang, Face recognition using laplacianfaces ,
IEEE Transactions on Pattern Ana1ysis and Machine Intel1igence 2005;27(3):328–340
[44]. Xin Pan, Qiu-Qi Ruan, Yan-Xia Wang, An Improved 2DLPP Method On Gabor Features For Palmprint Recognition ,proceeding in ICIP,
2007: 413-416
[45]. Kong, Adams, David Zhang, and Mohamed Kamel.: A survey of palmprint recognition. Pattern recognition 2009;42(7): 1408-1418.
[46]. Zhang, David, Wangmeng Zuo, and Feng Yue.: A comparative study of palmprint recognition algorithms. ACM Computing Surveys
(CSUR). 2012;44(1):1200-1237
[47]. M. M. H. Ali, P. Yannawar and A. T. Gaikwad.: Study of edge detection methods based on palmprint lines. 2016 International Conference
on Electrical, Electronics, and Optimization Techniques (ICEEOT), India, 2016;1344-1350
[48]. M. M. H. Ali, P. Yannawar and A. T. Gaikwad.: A Review: Palmprint Recognition Process and Techniques. Paper presented at: 2nd
International Conference on Cognitive Knowledge Engineering (ICKE-2016), (December 21-23, 2016)
[49]. Wirayuda, T.A.B., Adhi, H.A., Kuswanto, D.H. and Dayawati, R.N., Real-time hand-tracking on video image based on palm geometry,
International Conference of Information and Communication Technology, 2013; 241-246
[50]. Malik, J., Girdhar, D., Reference Threshold Calculation for Biometrics Authentication, International Journal Graphics and Signal
Processing. 2014; 2: 46 – 253
[51]. Ojala, Timo, Matti Pietikäinen, and David Harwood. : A comparative study of texture measures with classification based on featured
distributions. Pattern recognition . 1996; 29(1):51-59.
[52]. Zhao, Yang, Wei Jia, RongXiang Hu, and Jie Gui. Palmprint Identification Using LBP and Different Representations. In International
Conference on Hand-Based Biometrics (ICHB), 2011: 1-5.
[53]. Pan, Xin, and Qiu-Qi Ruan. "Palmprint recognition with improved two-dimensional locality preserving projections." Image and Vision
Computing 2008; 26(9): 1261-1268.