View metadata, citation and similar papers at core.ac.

uk brought to you by CORE

provided by Defence Science Journal

Defence Science Journal, Vol. 66, No. 6, November 2016, pp. 600-604, DOI : 10.14429/dsj.66.10800
 2016, DESIDOC

Biometrics in cyber Security

Jaspreet Kour!, M. Hanmandlu , and A.Q. Ansari*
#

!
Galgotias College of Engineering and Technology, Greater Noida - 201 306, India
#
Indian Institute of Technology Delhi, New Delhi - 110 016, India
*
Jamia Millia Islamia, New Delhi - 110 025, India
*
E-mail: [email protected]

ABStrAct
Computers play an important role in our daily lives and its usage has grown manifolds today. With ever increasing
demand of security regulations all over the world and large number of services provided using the internet in day
to day life, the assurance of security associated with such services has become a crucial issue. Biometrics is a key
to the future of data/cyber security. This paper presents a biometric recognition system which can be embedded
in any system involving access control, e-commerce, online banking, computer login etc. to enhance the security.
Fingerprint is an old and mature technology which has been used in this work as biometric trait. In this paper a
fingerprint recognition system based on no minutiae features: Fuzzy features and Invariant moment features has been
developed. Fingerprint images from FVC2002 are used for experimentation. The images are enhanced for improving
the quality and a region of interest (ROI) is cropped around the core point. Two sets of features are extracted from
ROI and support vector machine (SVM) is used for verification. An accuracy of 95 per cent is achieved with the
invariant moment features using RBF kernel in SVM.
Keywords: Cyber security, biometrics, fingerprint recognition, fuzzy features, invariant moment features

1. IntroductIon template stored in the database whereas an identification system

The rapid growth of internet and computer recognises an individual by searching the entire database
interconnectivity for various applications such as online for a match1. Physiological biometrics includes fingerprint,
banking, e-commerce, m-commerce has lead to the concern of face, palm print, knuckle print, iris, ear, hand geometry
security attacks associated with these transactions. Along with etc whereas behavioral biometrics includes signature, gait,
enormous benefits offered by fast growing digital world, there keystroke dynamics, speech, gesture etc. A biometric modality
are significant threats and risks to the government, defense must satisfy properties such as universality, distinctiveness,
and other critical departments in the country. Cyber security permeance and collectability to be used in a biometric system.
is a major issue in todays digital world with cyber crime is Finger print has several desirable properties to be
increasing day by day. The losses and distress caused by cyber considered for an effective recognition system. Fingerprint
attacks are prompting the pioneers in the field of information recognition is one of the oldest and mature biometric technology
security to look for reliable and robust security measures. and has applications in many areas. Ridges and valleys on
The growing use of online banking, e-commerce has led the surface of a fingertip form a pattern, ridges are dark and
to the use of biometric technology to secure these transactions. valleys are bright. Ridges and valleys generally run in parallel,
Biometrics provides a strong defense against cyber security bifurcating and terminating at various points. Minutiae are the
attacks. Biometrics is considered as a safeguard against cyber endpoints and crossing points of ridges. Finger prints possess
crime. Now a day, banks encourage the use of fingerprint to one or more regions where the ridge lines assume distinctive
authenticate transactions. Applications are available in the shapes called singularities. These regions are classified as
markets which are used to unlock computer or phones using loop, delta and whorl as shown in Fig. 1. The core is defined
facial recognition, fingerprint recognition or iris recognition. as the point on the innermost ridge and a point where three
A biometric system recognises a person using a specific different direction ridges meet is defined as the delta. The
physiological and or behavioral characteristic possessed by information provided by singular points is used for alignment,
a person1. A biometric system is called either a verification authentication and classification of fingerprints.
system or an identification system depending on the application The approaches for fingerprint recognition are mainly
where it is used. A verification system verifies an individual by divided into three categories: minutiae based, non minutiae
comparing its test biometric characteristic with the biometric based (image-based) and hybrid system which combines the
minutiae and non-minutiae based approaches. A feature vector
in the case of minutiae based approach involves extracting a
Received : 27 February 2016, Revised : 19 May 2016
Accepted : 03 August 2016, Online published : 31 October 2016 set of minutiae from fingerprints as shown in Fig. 2 in multi-

600
 KOuR, et al.: BIOMETRICS IN CyBER SECuRITy

Integrated wavelet and fourier mellin transform (WFMT)

is used for the feature extraction by Jin6, et al. Multiple WFMT
features constitute a set of reference invariant features. Euclidean
distance is calculated between various WFMT feature vectors
for obtaining the percentage match. Nanni and Lumini7 used
linear binary patterns (LBP) as a feature descriptor. Alignment
of fingerprint images ,which is based on minutiae is done
before dividing the image into several overlapping windows.
Gabor filter with LBP is the feature vector for the fingerprint
recognition system. The fingerprint descriptor as proposed by
Figure 1. Singular regions and core point1. Nanni and Lumini8 is based on HOG that represents an image
by a set of local histograms containing occurrences of gradient
orientation at each local cell of image. Decision level fusion
of orientation, texture and spectral features of fingerprint
image is proposed by Helfroush and Ghassemian9. Whole
fingerprint image is used extracting features instead of using
only points around core. yang and Park10 discussed moment
features having fixed length as the feature vector extracted
from tessellated cells around the core point of the fingerprint
image. Similarity measure based on the eigen value weighted
cosine distance is used to find the matching percentage of query
Figure 2. common minutiae types1.
and template. Fingerprint matching is based on the measures
dimensional plane. But this approach needs time consuming of similarity to the eigen value-weighted cosine distance to
fingerprint image quality improvement using various match the two corresponding feature vectors of the test and
enhancement techniques, applying filters, segmentation and also template. In the image based fingerprint matching method as
involves binarisation and thinning process and it also suffers presented by Kour and Awasthy2, the orientation field is used
from rejection of true or genuine minutiae and introduction of as a feature vector and the Euclidean distance is employed as
unreal and false minutiae. The rich discriminatory information a classifier. In fingerprint recognition system developed by
available in fingerprint is rarely capitalised in this approach Khalil11, et al. 16 co-occurrence matrices are used to compute
having high computational complexity. Image based fingerprint four statistical descriptors namely, correlation, contrast, energy
recognition does not involve detection and matching of and homogeneity and Euclidean distance is used as a classifier.
minutiae points. Features in these methods are extracted using Fusion of spectral and directional features is discussed by
different type of filters like Gabor filter, Sobel filter or by using Helfroush12 and Mohammad pour. Hartley transform is used
transforms like Fourier transform ,wavelet transform etc. Apart for feature extraction by Bharkad13 and Kokare. Hartley
from fingerprint ridge pattern other features in the category transform is faster than the Gabor filter requiring a set of filters
of non minutiae based approach include texture, ridge shape, oriented at different angles and frequencies. Finally, city block
orientation and frequency of local ridge pattern etc. Global distance is used for computing the similarity between two
information is exploited in such methods and they require fingerprint images. Discrete wavelet transform (DWT) based
less preprocessing compared to minutiae based methods2 fingerprint recognition system as described by Kumar14, et al.
.Combination of minutiae and non minutiae based methods are has been used as feature vector followed by Euclidean distance
categorised as Hybrid methods for fingerprint recognition. as a classifier. The features of fingerprints such as Directional
Information, Centre Area and Edge Parameters are obtained by
2. relAted WorK considering three DWT levels at four sub bands. Two features
A bank of Gabor filters is proposed by Jain3, et al. to obtain a used for fingerprint verification are wavelet and Pseudo
compact fixed length finger code which provides both local and Zernike Moments in Pokhriyal15, et al. The hybrid approach
global details from a fingerprint. The two corresponding Finger has the advantage of combining the most discriminating
Codes are matched using a distance measure like Euclidean features for fingerprint recognition. Multi resolution features of
distance. A hybrid fingerprint matching system utilising both aligned fingerprints are computed using 2-D discrete wavelet
minutiae and ridge direction information to represent and transform. Principal component analysis (PCA) is used for
authenticate fingerprints is described by Ross4, et al. A set of dimensionality reduction followed by Euclidean classifier
8 Gabor filters, whose spatial frequencies correspond to the for verification. Pseudo Zernike moments are used to create
average inter-ridge spacing in fingerprints, is used to capture invariant features and Bayesian classifier is used to validate the
the ridge strength at equally spaced orientations. moments in Deepika16, et al.
The feature vector proposed by Tico5, et al. consists of
the standard deviations of the Discrete Wavelet Transform 3. ProPoSed Method
coefficients of the whole image at various scales and orientations. The main steps involved in the proposed fingerprint
K-Nearest Neighbour classifier employing Euclidean distance recognition system are: Preprocessing, extracting region of
is used for matching. interest (ROI), feature extraction and verification. The block

601
 DEF. SCI. J., VOL. 66, NO. 6, NOVEMBER 2016

diagram of the proposed system is shown in Fig. 3. The Short using:

Time Fourier Transform (STFT) is applied to fingerprint image 1
∑ ∑ I (i, j )
W W
for enhancement which analyses it in space and frequency I av = i =1 i =1
(1)
WXW
domain. The Fourier domain-based block wise contextual
filter enhances the fingerprint images17. After preprocessing 3. Obtain the maximum intensity, Imax in the kth window.
a landmark point is needed in fingerprint image like core 4. Obtain the membership function, μij for every pixel in kth
point around which to earmark a region of interest (ROI). The window from:
core point is defined as point of maximum curvature on the I (i, j ) − I av
convex ridge. Maximum curvature is detected using complex µij = (2)
I max
filter methods to obtain accurate and reliable core point in a
fingerprint image18. After core point detection, area around the 5. Compute the complement of the above membership
core point is cropped for feature extraction. The choice of ROI function as this has values in the low range from 0 to 0.2
is made to ensure the availability of rich information available using
around the core point for fingerprint matching. The ROI of size (µ ) = 1 − µ
ij 1 ij (3)
64 by 64 pixels extracted from the whole fingerprint image is
shown in Fig. 4. 6. Compute the fuzzy feature from k window from: th

∑∑ ((µ ) ⋅ I (i, j ))
ij 1 (4)
I µk =
∑∑ (µ ) ij 1

The number of features obtained is equal to the number of

square blocks in the cropped image.
It may be noted that with cropped image of size 65x65
and window size (W) of 5x5, the number of features extracted
is 169.

Steps for invariant moment features

1. Divide the cropped image into various non- overlapping
square blocks of size W x W.
2. Find seven invariant moments invariant to translation,
scale change, mirroring and rotation for each window.
Figure 3. Block diagram of proposed method.
This set of moments19 is given by
φ1 = η20 + η02 (5)

φ2 = (η20 − η02 ) + 4η211

2
(6)

φ3 = (η30 − 3η12 ) + (3η21 − η03 )

2 2
(7)

φ4 = (η30 + η12 ) + (η21 + η03 )

2 2
(8)

φ5 = (η30 − 3η12 )(η30 + η12 )

(9)
( 30
 η + η12 )2 − 3 (η21 + η03 )2 

Figure 4. Fingerprint Image and roI extracted image.
+ (3η21 − η03 )(η21 + η03 ) 3 (η30 + η12 ) − (η21 + η03 ) 
2 2
 
Two set of features namely fuzzy features and invariant
φ6 = (η20 − η02 ) (η30 + η12 ) − (η21 + η03 ) 
2 2
moment features are obtained from the region of interest (ROI)
  (10)
of the fingerprint image.
+4η11 (η30 + η12 )(η21 + η03 )
3.1 Feature extraction
Two non-minutiae descriptors are presented here for φ7 = (3η21 − η03 )(η30 + η12 )
extracting features from cropped images which are ROIs. (η30 + η12 )2 − 3 (η21 + η03 )2 
These descriptors overcome the shortcomings of the minutiae  
(11)
based approach for fingerprint recognition. + (3η120 − η30 )(η21 + η03 )
3 (η30 + η12 )2 − (η21 + η03 )2 
Steps for fuzzy features  
1. Divide the cropped image into various non- 3. Reduce dynamic range of the moments using the
overlapping square blocks of size W x W. logarithmic function by
2. Calculate the mean intensity, I av in k th window φi = log (φi ) (12)

602
 KOuR, et al.: BIOMETRICS IN CyBER SECuRITy

where i = 1,2,…7. 5. concluSIon

It may be noted that for ROI of size 64 x 64 and window This paper presents a fingerprint recognition system
size (W) of 16 x 16, the number of features extracted is 112. which can be used for computer login, access control, internet
banking or other such applications where security is a concern.
3.2 Verification Biometrics is being effectively used as a safeguard against
Support vector Machine (SVM) is a widely used tool cyber crime. In this work a fingerprint recognition system is
in pattern reconition field. It makes use of the Structural developed using non-minutiae features: fuzzy features and
Risk Minimisation (SRM) principle. It provides very good invariant moment features extracted from ROIs as these result
performance in non linear patterns by transforming the input in fixed length feature vectors and are computationally simple.
vectors into a high dimension feature space with the help of a SVM is used for verification. A comparative study shows that
kernel. It locates optimal hyper plane for linearly separable data fuzzy features give an accuracy of 87.5 per cent whereas the
and extends to nonlinearly separable data by mapping them to invariant moment features give an accuracy of 95 per cent both
high dimension feature space. SVM has many applications with RBF kernel. Biometrics is a key to the future of cyber
like biometric based authentication, information management, security.
text categorisation, bioinformatics etc. SVMs are based on
the principle of Structural Risk Minimisation rooted in the reFerenceS
statistical learning theory. Details about SVM are available in 1. Maio, D. & Maltoni, D. Handbook of fingerprint
Vapnik20 and also in Byun21 and Lee. recognition. Springer Verlag, 2009.
LIBSVM22 has been used for verification. Four built in 2. Kour, J. & Awasthy, N. Non minutiae based fingerprint
kernels that we have at our disposal are; linear, polynomial, matching. In International Association of Computer
radial basis function, and sigmoid. Science and Information Technology –IEEE Spring
Parameter tuning is required in Support Vector Machines Conference, 2009, 199 – 203.
like other learning algorithms so as to find the optimal doi : 10.1109/IACSIT-SC.2009.127
parameters which result in improved accuracy on the test 3. Jain, A.K.; Prabhakar, S.; Hong, L. & Pankanti, S.
data. Three kernels are used and for each kernel, 5-fold cross Filterbank-based fingerprint matching. IEEE Trans. Image
validation is performed on the training set. A logarithmic ‘grid Proces., 2000, 9(5), 846-859. doi : 10.1109/83.841531
search’ is made to get the parameters c and γ. The values of 4. Ross, A.; Jain, A.K. & Reisman, J. A hybrid fingerprint
c and γ giving the best average cross-validation accuracy are matcher. Pattern Recog., 2003, 36(7), 1661-1673.
employed at the testing phase. doi : 10.1016/S0031-3203(02)00349-7
5. Tico, M.; Kuosmanen, P. & Saarinen, J. Wavelet domain
4. exPerIMentAl reSultS features for fingerprint recognition. Electron. Lett., 2001,
4.1 database 37(1), 21–22. doi : 10.1049/el:20010031
The benchmark database FVC2002DB1B23 has been 6. Jin, A.T.B.; Ling, D.N.C. & Song, O.T. An efficient
used in our experiments. It consists of ten users with eight fingerprint verification system using integrated
impressions each. The database is divided into training and wavelet and fourier-mellin invariant transform,
testing set. The training set consists of seven fingerprint images Image Vis. Comput., 2004, 22(6) 503–513.
and testing set consists of one fingerprint image. The images doi : 10.1016/j.imavis.2003.12.002
are taken with an optical sensor having a size of 388 x 374 at a 7. Nanni, L. & Lumini, A. Local binary patterns for a hybrid
resolution of 500dpi. fingerprint matcher. Pattern Recognition, 2008, 41(11),
3461-3466. doi : /10.1016/j.patcog.2008.05.013
4.2 experiments 8. Nanni, L. & Lumini, A. Descriptors for image-based
Different fingerprints are selected for training and testing fingerprint matchers. Expert Sys. Appl., 2009, 36(10),
and this process is repeated four times. The mean of the four 12414-12422. doi : 10.1016/j.eswa.2009.04.041
experimental trials is considered as the final performance. Each 9. Helfroush, S. & Ghassemian, H. Nonminutiae based
set of features is set aside for verification using SVM. decision level fusion for fingerprint verification. EURASIP
Different window sizes (W) have been experimented on J. Adv. Signal Proces., 2007. doi : 10.1155/2007/60590
ROIs. The best results are obtained on fuzzy features for a 10. yang, J.C & Park, D.S. A fingerprint verification
window size of 5x5 and on invariant moments for a window algorithm using tessellated invariant moment features.
size of 16x16. The features are normalised before inputting Neurocomputing , 2008, 71, 1939– 1946.
to SVM doi : 101016/j.neucom.2007.12.034
table 1. Accuracy of proposed system 11. Khalil, M. S.; Mohammed, D.; Khan, M. K. & Al-Nuzaili,
Q. Fingerprint verification using statistical descriptors.
Accuracy(%) Digital Signal Proces., 2009, 20(4),1264-1273.
FVc2002 doi : 10.1016/j.dsp.2009.12.002
linear rBF Polynomial
12. Helfroush, M.S. & Mohammadpour, M. Fingerprint
Fuzzy features 75 87.5 77.5 verification system: a non-minutiae based approach.
In International Conference on Computer, Control and
Invariant moment features 92.5 95 90
Communication , 2009, pp. 1-4

603
 DEF. SCI. J., VOL. 66, NO. 6, NOVEMBER 2016

doi : 10.1109/IC4.2009.4909271 contrIButorS

13. Bharkad, S. &. Kokare, M. Hartley transform based
fingerprint matching. J. Info. Proces. Sys., 2012, 8(1). dr Jaspreet Kour received her BTech, MTech from REC
doi : 10.3745/JIPS.2012.8.1.085 (NIT) Bhopal and uPTu Lucknow respectively and PhD
14. Kumar, S.; Raja, K.B.; Chhotaray, R.K. & Pattanaik, S. from Jamia Millia Islamia, New Delhi. Currently working as
an Associate Professor in the Department of Electronics and
DWT based fingerprint recognition using nonminutiae
Instrumentaion Engineering at Galgotias College of Engineering
features. Int. J. Comput. Sci., 2011, 8(2), 257-265. and Technology, Greater Noida, u.P. Her research areas are :
ISSN (Online): 1694-0814. Pattern recognition, image processing and biometrics.
15. Pokhriyal, A. & Lehri, S. A new method of fingerprint In the current study, Kour has designed the study and performed
authentication using 2-D wavelets. J. Theoretical Applied the experiments and drafted the mauscript.
Info. Technol., 2010, 13(2), 131 – 138.
16. Deepika, C.L.; Kandaswamy, A.; Vimal, C. & Sathish, dr Madasu hanmandalu received BE (Electricals Engineering)
B. Invariant feature extraction from fingerprint biometric from Osmania university, Hyderabad, in 1973; MTech in Power
using pseudo zernike moments. In International Joint systems from R.E.C. Warangal, Jawaharlal Nehru Technological
university, in 1976, and PhD from Indian Institute of Technology,
Journal Conference on Engineering and Technology,
Delhi, in 1981. Currently working as a Professor in the
2010 . Department of Electrical Engineering at Indian Institute of
17. Chikkerur, S.; Cartwright, A.N. & Govindraju, V. Technology, New Delhi. His current research interests mainly
Fingerprint enhancement using STFT analysis. Pattern include : Soft computing, biometrics, and medical imaging.
Recognition, 2007, 40(1), 198-211. In the current study, he has provided direction and guidance
doi :10.1016/j.patcog.2006.05.036 in the study and revised the manuscript.
18. Nillson, K. &. Bigun, J. Localisation of corresponding
points in fingerprints by complex filtering. Pattern Recog. dr Abdul Quaiyum Ansari received his BTech, MTech, and
Lett. , 2003, 24(13), 2135-2144. PhD from AMu, Aligarh, IIT Delhi and JMI, New Delhi,
doi : 10.1016/S0167-8655(03)00083-7 respectively. is a Professor in the Department of Electrical
Engineering at Jamia Millia Islamia, New Delhi. His research
19. Gonzalez, R.C. & Woods, R.E. Digital image processing,
interests include : Computer networks, networks-on-chip, fuzzy
second ed., Prentice-Hall, 2002 logic, and image processing.
20. Vapnik, V.N. Statistical learning theory. New york: In the current study, he has revised the manuscript for important
Wiley-Interscience, 1998. technical content and guided throughout the study.
21. Byun, H. & Lee, S. SVM survey applications of support
vector machines for pattern recognition: A Survey.
LNCS 2388, Springer-Verlag Berlin Heidelberg, 2002,
pp. 213-236.
22. Chih-Chung, Chang & Chih-Jen, Lin. LIBSVM: a library
for support vector machines, 2001. http://www.csie.ntu.
edu.tw/~cjlin/libsvm (Accessed on 5 Feburary 2011).
23. FVC2002 Web site: http://bias.csr.unibo.it/2002
(Accessed on 3 Feburary 2012).

604