FINGERPRINT CLASSIFICATION MODEL BASED ON NEW COMBINATION OF PARTICLE SWARM OPTIMIZATION AND SUPPORT VECTOR MACHINE

http://www.iaeme.com/IJCIET/index.asp 78 editor@iaeme.com
International Journal of Civil Engineering and Technology (IJCIET)
Volume 9, Issue 11, November 2018, pp. 78–87, Article ID: IJCIET_09_11_008
Available online at http://www.iaeme.com/ijciet/issues.asp?JType=IJCIET&VType=9&IType=11
ISSN Print: 0976-6308 and ISSN Online: 0976-6316
© IAEME Publication Scopus Indexed
FINGERPRINT CLASSIFICATION MODEL
BASED ON NEW COMBINATION OF PARTICLE
SWARM OPTIMIZATION AND SUPPORT
VECTOR MACHINE
Radhwan Hussein Abdulzhraa Al_Sagheer
Department of Computer Science, Faculty of Education for Girls,University of Kufa,Iraq
Jammel Mona
Computer center, Faculty of Medicine,University of Kufa, Iraq
Abdulhussein Abdulmohson
Faculty of Education, University of Kufa, Al-Najaf, Iraq
Mohammed Hasan Abdulameer
Department of Computer Science, Faculty of Education for Girls,University of Kufa,Iraq
ABSTRACT
Fingerprint is one of the broadly utilized biometric distinguishing proof to
recognize the persons due steady and worthiness. The classification of fingerprints
gives indexing to the dataset to decrease the seeking and matching manner. There
researches have been developed numerous approaches for fingerprint classification
model, for example, the Neural Network (NN) methods and support vector machine.
Nevertheless, there are a lot of developments to develop the classification procedure.
In this paper, a new classification model based on PSO-SVM technique is used to
achieve fingerprint classification process. The proposed methodology comprises three
phases: preprocessing phases, feature extraction phase and classification phase using
the proposed PSO-SVM model. The CASIA V5 fingerprint dataset and a property
dataset are adopted in this study to test the proposed model. The experimental results
showed that the proposed PSO-SVM classification model was better than the standard
SVM in terms of accuracy, sensitivity and specificity for both datasets.
Keywords- Biometric, Particle swarm optimization, Support Vector Machine,
classification, Fingerprint
Cite this Article: Radhwan Hussein Abdulzhraa Al_Sagheer, Jammel Mona,
Abdulhussein Abdulmohson andMohammed Hasan Abdulameer, Fingerprint
Classification Model Based on New Combination of Particle Swarm Optimization and

Radhwan Hussein Abdulzhraa Al_Sagheer, Jammel Mona, Abdulhussein Abdulmohson
andMohammed Hasan Abdulameer
Support Vector Machine, International Journal of Civil Engineering and Technology,
9(11), 2018, pp. 78–87
http://www.iaeme.com/IJCIET/issues.asp?JType=IJCIET&VType=9&IType=11
1. INTRODUCTION
Biometrics are computerized approaches for distinguishing a person depends on his/her
physiological or behavioral features. In the past years, the biometrics approaches were based
on recognizing the people via their unique body features like their heights, eye colors and
appearance. Currently, there are many advanced techniques are based on biometrics such as
face recognition techniques, fingerprints techniques, hand geometry methods, iris techniques,
etc. The physiological biometrics are considered the first choice of most of the developers in
general due the uncomplicatedness and cost effectiveness of the physiological biometrics [1].
The fingerprint is one of the most popular physiological biometrics due its achievability,
uniqueness, stability, accurateness, and acceptability. The general definition of fingerprint
recognition fingerprint compared to another fingerprint in order to decide if the printings are
from the exact finger or not. It comprises two procedures: verification and identification [2].
The identification is called 1:N match and it happens when the identity is unknown and has to
search through the entire database to find it or not. On the other hand, the verification is called
1:1 match and it is the process of checking when the identity is claimed from someone. The
search is take place to find the claimed person’s identity only. Recently, neural networks,
genetic algorithm and support vector machine are among the well-known machine learning
techniques that are adopted to create a good classifier for fingerprints [3][4]. Particle swarm
optimization is one of the highly common methods in artificial intelligence field [5], and many
versions have been exploited with support vector machine as a classifiers methods for human
faces [6][7][8][9]. In this paper, we will use a combined classifier based on “particle swarm
optimization (PSO)” and “support vector machine (SVM)” in order to introduce a new
classifier able to distinguishing fingerprints powerfully.
2. RELATED WORKS
Several researches to classify fingerprints have been industrialized by the researchers. Based
on the number of core and delta point and their topological relation, The singular points based
fingerprint classification method is developed in order to identify fingerprint type by [10-13].
This technique depend on the mining accuracy of the singular points. Though, it was sensitive
to the fingerprint quality and that could make the false detection rate very high [14]. Another
fingerprint technique is developed based on the structure of fingerprint [15] that utilizes the
variance of geometrical structure. It uses the global structure of fingerprint image rather than
local information and that makes it more powerful than other methods. However, the
geometrical structure of the fingerprint is hard to attain for low quality images. To solve the
challenge of low quality of fingerprint images, A classification technique for fingerprint based
on “j-divergence entropy” and [16-17] SVM. In addition, and based on DWT in combination
with SVM is proposed by [18-19]. Their methods were powerful in dealing with the problem
of low quality fingerprint images. So that in this paper, we will develop classification method
uses particle swarm optimization and support vector machine (PSO-SVM) that can deal with
low quality fingerprint images.

Fingerprint Classification Model Based on New Combination of Particle Swarm
Optimization and Support Vector Machine
3. AUTOMATIC FINGERPRINT CLASSIFICATION
Feature extraction and feature matching are the main phases in any fingerprint identification
method. Minutiae extraction method is widely used in for fingerprint feature extraction
compared to the other methods [2]. There are five classes such as, Arch, Tented Arch, Right
Loop, Left Loop and Whorl according to Henry classes as illustrated in Figure 1. In the
matching phase, the gained features matched with the existing features in the database. The
comparison process is one to many comparison and that makes it consumes much time.
Nevertheless, this procedure may perform better if we adopt the classification process which
aids in decreasing the number of comparisons throughout matching phase.
Figure 1 Henry classes of fingerprint
4. THE PROPOSED METHODOLOGY
The proposed methodology comprises of three most important phases. Image preprocessing is
the first phase followed by Phase 2, feature extraction using minutiae extraction method and
Phase 3, Development and implementation of PSO-SVM classifier as illustrated in figure 2
below:

Figure 2 The proposed methodology
By rolling the finger from one side of the nail to the other side over the inked slab, the first
step in preprocessing steps, which is sensing is accomplished. After that, the finger is rolled on
a piece of white paper so that the inked impression of the ridge pattern of the finger appears on
the white paper. Rolled inked fingerprints impressed on paper can be scanned into digital rolled
fingerprints image using flat bed scanners by choosing a "128x128 pixels" gray scale at "100
dpi" resolution with double size. Then, this image is saved as a bitmap extension (.bmp). Figure
3.1 shows some of the fingerprint images that were captured using a rolled inked method.
Figure 3. Sample of fingerprint images that captured using rolled inked method
The second step in preprocessing is to convert the scanned image into a binary image which
is called binaraization . the thresholding process is applied to segment the fingerprint image
and that is done by locating the pixels that has intensity values beyond the threshold in a
foreground and the all residual in a background. The third step is the edge detection which is
done after the binaraization step and it is very important step to prepare the image for feature
extraction pahse. Samples of fingerprint images and their edge results are shown in the figure
4 below:

Figure 4 The images after edge detection process
The result of the previous stage is a binary image with 126x126 pixels, because thick ridges
with some individual pixels can be noticed in this image. No features can be got from this
image because of the thick edges; therefore a thinning algorithm must be applied to obtain a
skeletonized fingerprint image to make the feature extraction easier. Figure 5 below
demonstrates the resultant thinned image.
Figure 5.The resulted thinned images
The second essential phase in our proposed model is Feature extraction. Minutiae extraction
process is much easier when the grey scaled image undergo binarization and thinning process.
There are 150 types of minutiae on the fingerprint but, only ridge ending and ridge bifurcation
were widely used for differentiate each fingerprint.
5. CLASSIFICATION USING PSO-SVMS MODEL
One of the famous supervised learning classification technique is support vector machine
that uses the statistical learning mechanism. The goal of SVM is to find a suitable hyperplane,
which can classify two classes through manipulating the datasets after training [7]. Assume the
training database is represented by , where x is nput vector and the label of the class
is ∈ 1, 1 . The formula to calculated the required hyperplane is . 0, where
means the point positioned on the required hyperplane, describes the locating of the requited
hyperplane and h signifies the bias of the space from the starting point to hyperplane. The top
dividing hyperplane is created through minimizing ‖ ‖ with the constraint .
1, 1,2, … … . , For that reason, getting of the best hyperplane is vital to response to the
optimization problem that stated via:
min ‖ ‖ (1)
. 1, 1,2, … … . ,
In order to exchange the optimization problem, the “positive slack variables” which comprise
ξ#, ξ#
∗
are introduced, and then the method is extended to work with “nonlinear decision
surfaces”. The extended optimization problem is supposed as:
%
, &
‖ ‖ ' ∑ & (2)
. 1 & , & 0, 1,2, … … . ,
In the above-mentioned formula, C is point to “penalty parameter”, that is responsible on
guiding the tradeoff between margin expansion and “error minimization”. As a result, the result
function of the classification is becomes:
f x sin ∑ α#
-
# y#k0x#, x12 h (3)

In the preceding formula, α# is indicate to “Lagrange multipliers” and k0x#, x12 = ϕx#, ϕx1
is the used kernel function. The role of kernel function is to map the data in to “higher
dimensional space” through nonlinear mapping functions ϕx. The RBF kernel function is
commonly utilized in the previous researches for images classification purposes. For that
reason, we utilized RBF in the proposed PSO-SVM model. The RBF is represented by
exp(−7x# − x17/2σ ), σ is indicates to a real number that is positive.
6. PARAMETERS SELECTION OF SVM WITH PSO
PSO introduced by Kennedy & Eberhart in 1995, and it is motivated by means of the social
activities between entities like the birds cooperative or the fish groups. PSO particles search
for the finest location with respect to the equivalent finest solution in the virtual search space
for an optimization problem. Throughout each generation of particles, the generated particles
travels to the direction of its own best location and its best global location. The traveling
procedure of the particle is designated as:
: ;
<=
= <
: ;
<
+ > . ?@ A. (B ;
C
− ;
<) + > . ?@ A. 0BDEFG<
<
− ;
<
2 (4)
;
<=
= ;
<
+ H: ;
<
(5)
In the aforementioned formula, t means the tth
iteration and C1 and C2 are velocity learning
factors. Also, the rand is represents number between the interval 0 and 1 which is positive under
normal distribution. ɑ is the constraint factor that is able to control of the velocity weight.
represents the inertial weight factor and ; indicates to the a particles i velocity while
: ; indicates to the velocity of particle i. the personal best location of particle i is represented
by B ; and the finest one of all personal best locations of all particles in the swarm is
represented by B DEFG<. The “RBF kernel function” is used to build the SVM, that has two user-
determined parameters C and σ. Hence, the particle is composed of those parameters > and .
The process of parameters optimization of SVM with PSO is illustrated in the Figure 6 below:
Figure 6. SVM training parameter using PSO
7. EXPERIMENTAL RESULTS
To develop an efficient and consistent model, we need to use well-known dataset besides a new
collected property dataset in order to train and test the model. For this purpose, we utilized
“CASIA Fingerprint Image Database Version_5.0” [18] and property fingerprint dataset in
order to assess the proposed model. The CASIA Fingerprint Database comprises 20,000
fingerprint images have been collected from 500 persons. The candidates were graduate

students, workers, waiters, etc. every candidate added 40 fingerprint images of his eight fingers
and 5 images for every finger. The resolution of the fingerprint images is 328*356 and the
entire dataset images are 8 bit gray-level. The CASIA dataset is divided into five sub datasets
named as (CAS1, CAS2,CAS3,CAS4 and CAS5) and each sub dataset contains 1000 images
for 100 subject. Each sub dataset is divided 40% for training and 60% for testing. The figure 7
below shows sample from CASIA fingerprint dataset
Figure 7. Sample image from CASIA fingerprint dataset
In addition, we used a property dataset that contains 500 fingerprint images for 50 subjects
as shown in figure 8. The property dataset is divided to five sub dataset termed as (prop 1,
prop2, prop3, prop4 and prop5) and each sub dataset contains 100 fingerprint images. Every
property sub dataset is divided to 40% for training and 60% for testing.
Figure 8. Sample from property dataset
In order to assess the performance of the proposed model, the common 10 folds cross
validation test is applied over each dataset. The statistical performance measures are used to
evaluate the discrimination power of the proposed algorithm, which are described below:
Accuracy= (TP+TN)/ (TP+TN+FP+FN) , Sensitivity= TP/ (TP+FN) AND Specificity=
TN/ (FP+TN) . To perform n-fold cross validation, ten folds of training and testing datasets are
generated by folding operation. The cross validation results for 1: N recognition over CASIA
five datasets is tabulated in Table 1 and figure 9 and for property five sub datasets in Table 2
and figure 10.
Table 1 Average Cross validation results of PSO-SVM and the standard SVM for CASIA five sub
datasets
SVM PSO-SVM
Accuracy Specificity Sensitivity Accuracy Specificity Sensitivity
CAS
1
0.8 0.7 0.8 0.9 0.9 1
CAS
2
0.9 0.9 0.9 1 1 1
CAS
3
0.8 1 0.9 0.9 1 0.9
CAS
4
0.8 0.5 0.9 1 1 1
CAS
5
0.7 0.4 0.9 0.8 0.7 0.9

Figure 9 Average Cross validation results with CASIA dataset
Table 2 Average Cross validation results of PSO-SVM and the standard SVM for the property five
sub datasets
SVM PSO-SVM
Accuracy Specificity Sensitivity Accuracy Specificity Sensitivity
Prop
1
0.9 0.9 0.9 1 0.9 1
Prop
2
0.8 0.8 0.9 0.9 1 1
Prop
3
0.8 0.8 0.8 1 1 0.9
Prop
4
0.9 0.9 0.9 1 0.9 0.8
Prop
5
0.7 0.8 0.8 0.9 1 1
Figure 10 Average Cross validation results with the property dataset
From table 1 and table 2, the average accuracy performance of the proposed PSO-SVM
model is outperform the standard SVM in most cases of CASIA sub datasets and the property
sub datasets. Moreover, the results of sensitivity and specificity measures of the proposed PSO-
SVM model are better in most cases as well. In addition, the optimal training parameters of C
and σ of SVM for CASIA sub datasets are shown in table 3 and while the optimal parameters
of SVM for the property sub datasets are depicted in table 4. The five SVMs for every datasets
are used as a standard case to show the stability of the proposed PSO-SVMs model.

Table 3 The optimal parameters of 5 SVMs via PSO on CASIA sub datasets
C σ
PSO-SVM1 103 1.01
PSO-SVM2 88.57 0.94
PSO-SVM3 75.32 1.24
PSO-SVM4 68.12 1.19
PSO-SVM5 76.88 0.86
Table 4 The optimal parameters of 5 SVMs via PSO on the property sub datasets
C σ
PSO-SVM1 95.44 1.03
PSO-SVM2 114.6 0.32
PSO-SVM3 89.54 1.87
PSO-SVM4 55.82 1.43
PSO-SVM5 76.88 1.87
8. CONCLUSION
In this paper, a new classification model based on PSO-SVM technique has been introduced
to accomplish fingerprint classification process and to decrease the searching process
and matching. The proposed model included three major phases: preprocessing phase,
feature extraction phase and classification phase using the proposed PSO-SVM model.
In the experiments, CASIA V5 fingerprint dataset and collected property fingerprint
dataset are utilized in this paper in order to examine the proposed model. The
experimental results showed that the proposed PSO-SVM classification model was
outperform the standard SVM in terms of accuracy, sensitivity and specificity for both
datasets.
References
[1] Majeed, Adnan, and Tarogil Lahore Pakistan. "Biometric Identification Criminal, Terrorist,
and individual Identification System It ‘sa New Wine & Taste.", international Journal of
Advanced Research in Computer Engineering & Technology (IJARCET), 4 (6), June, pp:
2977- 2986, 2015
[2] Kumar, Krishna, Basant Kumar, Dharmendra Kumar, and Rachna Shah. "Fingerprint
Recognition using Minutiae Extraction." In International Conference on ICT-Initiatives,
Policies & Governance, Dehradun, India, pp. 28-29. 2011.
[3] Alessandro Baffa, "Fingerprints Recognition using Neural Network",Milan, Italy, 2008.
[4] Yousself Elmir, Zakaria Elberrichi and Reda Adjoudj. (2012). Support Vector Machine
based Fingerprint Identification, Algeria.
[5] Kennedy, R. "J. and Eberhart, Particle swarm optimization." In Proceedings of IEEE
International Conference on Neural Networks IV, pages, vol. 1000. 1995.
[6] Abdullah, Siti Norul Huda Sheikh, Mohammed Hasan Abdulameer, Nazri Ahmad Zamani,
Fasly Rahim, Khairul Akram Zainol Ariffin, Zulaiha Othman, and Mohd Zakree Ahmad
Nazri. "2.5 D Facial Analysis via Bio-Inspired Active Appearance Model and Support

Vector Machine for Forensic Application.", International Journal of Advanced Computer
Science and Applications, 8(7),2017,pp:370-375
[7] M. Hasan, S.N.H.S. Abdullah and Z.A.Othman, “Face recognition based on opposition
particle swarm optimization and support vector machine,” IEEE International Conference
In Signal and Image Processing Applications (ICSIPA), October , Malaka, Malaysia , 2013,
pp. 417-424
[8] M. H. Abdulameer, S.N.H.S. Abdullah and Z.A.Othman, “Support vector machine based
on adaptive acceleration particle swarm optimization,” The Scientific World Journal, 2014.
[9] M. H. Abdulameer, D.Mohammed, , S. A. Mohammed, M., Al-Azawi, Y. M. H. Al-Mayali,
and I. A. Alameri, “Face recognition technique based on adaptive opposition particle swarm
optimization (AOPSO) and support vector machine (SVM),” ARPN Journal of Engineering
and Applied Sciences, VOL. 13(6), (2018), pp.2259-2266
[10] Kalle Karu, Anil K. Jain; Fingerprint classification [J].PatternRecognition,3, 389-404
(1996).
[11] Guijun Nie, Chen Wu, Xijun Ye et al; Fingerprint Classification Based on Both
Continuously Distributed Directional Image and Modified Version of Poincar Index [J].
Acta Electronica Sinica, 5 , 947-952 (2006).
[12] Lin Wang, Mo DaiApplication of a new type of singular points in fingerprint classification
[J]. Pattern Recognition Letters, 13 , 1640-1650 (2007).
[13] Manhua Liu. Fingerprint classification based on Adaboost learning from singularity
features [J]. Pattern Recognition, 43 , 1062-1070 (2010)
[14] Yu Zhang, Yilong Yin, Gongqing Tuo; Fingerprint Image Quality Classification Based on
SVM [J]. Pattern Recognition and Artificial Intelligence, 1, 129-135 (2009).
[15] Cappelli R, Lumini A, Maio D; Fingerprint classification by directional image partitioning
[J]. IEEE Transactions on Pattern Analysis and Machine Intelligence, 5 , 402-421 (1999).
[16] Kadhim Naief Kadhim and Ahmed H. ( Experimental Study Of Magnetization Effect On
[7] Ground Water Properties).Jordan Journal of Civil Engineering, Volume 12, No. 2, 2018
[17] Yang, G., Yin, Y. and Qi, X., 2014. Fingerprint Classification Method based on J-
divergence Entropy and SVM. Appl. Math, 8(1L), pp.245-251
[18] Akbar, S., Ahmad, A. and Hayat, M., 2014. Identification of fingerprint using discrete
wavelet transform in conjunction with support vector machine. IJCSI, 11, pp.1694-0814.
[19] Chinese Academy of Sciences, “CASIA face database,” 2007

FINGERPRINT CLASSIFICATION MODEL BASED ON NEW COMBINATION OF PARTICLE SWARM OPTIMIZATION AND SUPPORT VECTOR MACHINE

More Related Content

What's hot (17)

Similar to FINGERPRINT CLASSIFICATION MODEL BASED ON NEW COMBINATION OF PARTICLE SWARM OPTIMIZATION AND SUPPORT VECTOR MACHINE (20)

More from IAEME Publication (20)

Recently uploaded (20)

FINGERPRINT CLASSIFICATION MODEL BASED ON NEW COMBINATION OF PARTICLE SWARM OPTIMIZATION AND SUPPORT VECTOR MACHINE