Professional Documents
Culture Documents
Performance Analysis of GLCM-based
Performance Analysis of GLCM-based
What do you think of dblp? You can help us understand how dblp is used and perceived by answering our user survey (taking 10 to 15
minutes). Your help is highly appreciated!
search dblp
[+]
[–] 6. DICTAP 2016: Konya, Turkey
> Home > Conferences and Workshops > DICTAP Dagstuhl
Sixth International Conference on Digital Information and Communication Technology and its
Applications, DICTAP 2016, Konya, Turkey, July 21-23, 2016. IEEE 2016, ISBN 978-1-4673-9609-7
Kerem Kucuk:
Design and implementation of a real-time ERP-OFDM SDR receiver on the USRP2 platform. 24-29
Muhammad Fahim, Asad Masood Khattak, Thar Baker , Francis Chow, Babar Shah:
Micro-context recognition of sedentary behaviour using smartphone. 30-34
Afdallyna Fathiyah Harun, Steve Benford, Claire O'Malley, Nor Laila Md. Noor:
Glyphing it! Children using markers in school. 39-44
Shahid Alam:
SUIS: An online graphical Signature-Based User Identification System. 85-89
https://dblp.org/db/conf/dictap/dictap2016.html 2/4
1/12/2021 dblp: DICTAP 2016
Truncated compact design of monopole antenna with modified ground plane for UWB applications. 115-
118
Syed Yasser Arafat, Muhammad Inzimam Saghir, Mubah Ishtiaq, Umer Bashir:
Comparison of techniques for leaf classification. 136-141
Islam A. T. F. Taj-Eddin, Mahmoud Afifi , Mostafa Korashy, Doha Hamdy, Marwa Nasser, Shimaa Derbaz:
A new compression technique for surveillance videos: Evaluation using new dataset. 159-164
estimation. 183-186
https://dblp.org/db/conf/dictap/dictap2016.html 4/4
Performance Analysis ofGLCM-based
Classification on Wavelet Transform-Compressed
Fingerprint Images
Abstract- Fingerprint detection is one of the primary many distinctive personal biological characteristics such
methods for identifying individuals. Gray Level Co as iris, fingerprint, voice, signature or face recognition that
occurrence Matrix (GLCM) is the oldest and prominent
are utilized for personal identification. Since biological
statistical textual feature extraction method applied in many
features are unique and thus more reliable than traditional
fields for texture analysis. GLCM holds the distribution of
methods during individual identification, such systems are
co-occurring intensity patterns at a given offset over a given
image. However, images occupy excessive space in storage by
more useful and suitable than conventional procedures.
its original sizes. Thus, Discrete Wavelet Transform (DWT) Biometric systems are now commonly used in different
based compression has become popular especially for parts of everyday life such as building access and computer
reducing the size of the fingerprint images. It is important to login. Fingerprint detection is one of the most prevalent
investigate whether GLCM-based classification can be methods for personal identification within all biometric
utilized efficiently on DWT-compressed fingerprint images. systems [3]. Throughout the world, fingerprint recognition
In this paper, we analyze the performance of GLCM-based is accepted by a large part of the population because of its
classification on DWT-compressed fingerprint images. We
fast, secure, and easy way of personal identification.
performed satisfying simulations for different levels ofDWT
Fingerprints are the inherent parts of humans and unique
compressed images. Simulation results identify that
classification performance sharply decreases by the increase
for each individual. Even twins have different fingerprints.
of DWT-compression level. Besides, instead of utilizing all Using the patterns found on the finger tips, fingerprint
Haralick features, it is recognized that 8 of them are the most biometric data is captured. As with every technology,
prominent ones that affect the accuracy performance of the using fmgerprint for biometrics has its own advantages and
classification. disadvantages. Age of the user, dryness and wetness of the
finger, temporary or permanent cuts on the fmger, or dirt
Index Terms- Fingerprint Detection; GLCM; DWT; on the fingerprint scanner all cause the fmgerprints to have
Classification. weak patterns of ridge and valleys that can degrade the
performance of a fingerprint biometric system. The
1. INTRODUCTION
mentioned issues lead to poor recognition rates in
Biometrics refers to the human specific behavioral and fingerprint identification system and users would
physiological characteristics belonging to a person which experience diffi cuIties with the system's authentication
are distinctive of the other individuals and used for the process. Furthermore, as discussed before there can be
personal identification [1-2]. A biometric system includes artificial fmgers which threaten system's security. For
many distinctive personal biological characteristics such creating artificial fingers a copy of the real fmgerprint is
as iris, fingerprint, voice, signature or face recognition that required which can be obtained when the victim is
are utilized for personal identification. Since biological touching objects like glasses and can be replicated in
features are unique and thus more reliable than traditional gelatin. For making the systems secure against artificial
methods during individual identification, such systems are fingers, liveliness detection can be used. Liveliness
more useful and suitable than conventional procedures. detection searches for certain attributes which can only be
Biometrics refers to the human specific behavioral and found on the live fingers such as the pulse in the fingertip.
physiological characteristics belonging to a person which However, individuals have different heart rate which can
are distinctive of the other individuals and used for the be different from time to time. Besides, users need to keep
personal identification [1-2]. A biometric system includes their fingers on the sensor without moving for few seconds
Authorized licensed use limited to: ULAKBIM UASL - Istanbul Arel University. Downloaded on January 11,2021 at 20:35:12 UTC from IEEE Xplore. Restrictions apply.
to detect the fingertip pulse. However, there are artificial defmes a wavelet family If/.,,((jJ(s, r) E R+* xR) from
fingers called wafer thin artificial fmgers that can fool such . ,
'\I s s
uniqueness and continuity, the use of fmgerprints is
considered to be one of the most reliable methods of
By extension, the families of functions on sub manifolds
personal verification. Due to the reasons such as
maintaining the public order as is proper, law enforcement of Rn invariant under a group of locally isomorphic to
requirements and increasing interest of civilian application the group affine transformation can also be described as
developers, automated fingerprint verification systems wavelet families. The Daubechies wavelet transfonnation
have become globally prevalent and been extensively is shown in Fig. 1. Tn this figure the red curve shows the
examined by the researchers and communities [4]. As a wavelet coefficient and the blue color shows the scaling
result, fmgerprint recognition is today being increasingly function. This curve is used for 4-tap Daubechies wavelet
used in diverse number of applications such as access which is used for implementation in this paper.
control or online identification [5]. Structure-based
Daubechies 4 tap wavelet
classification of fmgerprint images grooves local images
1.5
[6, 8] are used to classify. [6] In the fifth class to separate
classification accuracy report rate (with 18% rejection) is 1
90% and the source [7] to classify the four classes is
91.5%. A change in the classification of syntactic grammar 0. 5
fingerprint and hence their classification [9, 11] are used.
Good accuracy report [10] was in the classification of five o
classes (with 20% rejection) of 95.6%.
-0.5
In this paper, we perform a comprehensive performance
analysis of GLCM-based classification over DWT -1
compressed images. During DWT, Daubechies wavelets
are applied. First-level of wavelet transformation is -1.5 scaling function -
utilized and subsequently the GLCMs are generated from wavelet function --
Authorized licensed use limited to: ULAKBIM UASL - Istanbul Arel University. Downloaded on January 11,2021 at 20:35:12 UTC from IEEE Xplore. Restrictions apply.
image. If the intensity values of the image vary in the
interval [0, L-I} then the size of the GLCM becomes (L-
1) x(L-I). An example of configuration of the four GLCM
matrixes (GLCMoo, GLCM45o, GLCM90o, GLCM135o) of an (2)
image/for d=O is demonstrated in Fig. 2. k=l
The pOSItIOn of a point in an Euclidean n-space is
an Euclidean vector. Euclidean norm, Euclidean length,
or magnitude of a vector measures the length of the vector:
D. Euclidean norm
For i = (h iJ, ..., in) andj = Ol,h, ...,jn) are two points
in Euclidean n-space, the Euclidean distance d from i toj,
or fromj to i is given by the Pythagorean formula:
Authorized licensed use limited to: ULAKBIM UASL - Istanbul Arel University. Downloaded on January 11,2021 at 20:35:12 UTC from IEEE Xplore. Restrictions apply.
100
---a--- 1 image train
-- 2 image train
90 ---a--- 3 image train
------*-- 4 image train
'�"'j Li D ;. � 5 image train
�.
80
-----+--- 6 image trBin
-- 7 image train
70
C
'"
" 80
Q; Fig. 8. Flowchart of the GLCM feature extraction with DWT.
c..
50
In Fig. 9, the results for one-level DWT with 8 Haralick
40
features is illustrated. As shown in this figure, the highest
accuracy is obtained with 7 training image data.
30
20
10 15 20 25 30
person
5 10 15 20 25 30
person
In Fig. 7 the classification performance for the GLCM Tn this study, we make a comprehensive performance
feature extraction without DWT is shown. As shown in analysis of three different methods for fingerprint
this figure the highest accuracy is acquired for 7 training identification. In the first method, feature extraction and
image data. classification is applied to the raw DWT utilized images.
Second method considers only GLCM based feature
100 ����----�------.-----�------,-----, extraction without DWT. Finally, GLCM-based feature
---e--- 1 image train
-- 2 image train
extraction and classification is applied to the DWT
90
---a--- 3 image train compressed images. The third method gives promising
-----+--- 4 image train
5 image train
classification performance regarding other methods. The
80
-----+--- 6 image train best identification accuracy around 93% is achieved with
-- 7 image train
7 training images for each individual. Results of the
C 70 simulations identifY that Increase In the DWT
'"
f:!
t1J compressIOn level significantly decreases the
c..
80
performance.
50
40
30
5 10 15 20 25 30
person
Authorized licensed use limited to: ULAKBIM UASL - Istanbul Arel University. Downloaded on January 11,2021 at 20:35:12 UTC from IEEE Xplore. Restrictions apply.
TABLE I. Classification Performances
N urn ber of training images for each oerson
Scenario Method
1 2 3 4 5 6 7
wavelet feature extraction
1 44.5751 63.5165 69.0334 62.4762 80.4706 61.0157 75.1084
(without GLCM)
GLCM feature extraction
2 67.5972 68.1317 71.5104 77.7631 69.9185 70.5839 81.0558
without DWT
3 one-level DWT with GLCM 56.6369 50.8868 44.8066 54.9318 75.7699 48.5057 93.2915
Authorized licensed use limited to: ULAKBIM UASL - Istanbul Arel University. Downloaded on January 11,2021 at 20:35:12 UTC from IEEE Xplore. Restrictions apply.