Content uploaded by U. Hariharan
Author content
All content in this area was uploaded by U. Hariharan on Aug 19, 2020
Content may be subject to copyright.
Multimodal Biometric Authentication System using
Deep Learning Method
Sandeep Singh Sengar
Department of Computer Science and
Engineering
SRM University-AP, Amaravati
Andhra Pradesh, INDIA
sandeep.iitdhanbad@gmail.com
U. Hariharan
Department of Information Technology
Galgotias College of Engineering and
Technology
Greater Noida, Uttar Pradesh, India
*udhayakumarhariharan@gmail.com
K. Rajkumar
Department of Information Technology
Galgotias College of Engineering and
Technology
Greater Noida, Uttar Pradesh, India
rajonline7@gmail.com
Abstract—For specific identification process, Identity
Management details an ailment of supplying authorized
owners with secure and easy admittance to information and
solutions. For choosing the individual's identity, the primary
goal is actually executing secured identification feature. PINs,
keys, gain access to cards, passwords, tokens are actually the
private determining elements which are actually utilized within
standard methods which may have a tendency to drawbacks
such as cracking, stealing, copying and posting. Biometrics
grounded identification is needed having a perspective to stay
away from the drawbacks. Due to intra category variants, non-
universality, sound as well as spoof strikes are impacted.
Multimodal biometrics are actually employed to get rid of the
episodes which are actually a grouping of countless modalities.
For an authentication supply, Fingerprint and Palmprint
identification are popular systems these days. For minutiae
thing detection as well as attribute extraction, with this paper,
rich neural community (DNN) were definitely projected. The
confinements of unimodal biometric structure lead to
substantial False Acceptance Rate (FAR) along with False
Rejection Rate (FRR), limited splitting up skill, top bound
within delivery therefore the multimodal biometric product is
designed to satisfy the strict delivery demands. For minutiae
corresponding, values of Euclidean distance are actually used.
The better identification pace is actually attained throughout
the suggested procedure & it's extremely safe only in loud
problem.
Keywords—DNN, Biometrics, FAR, FRR, 2D Gabor Filter,
Deep Neural Network, Euclidean distance.
I. INTRODUCTION
The zone of unique identification is actually mistreating
computer aided devices as a far more protected also
progressively impressive way as well as biometrics is
actually a standout amidst most reliable characteristics which
could be used for specific recognition. For conventional
identification method passwords [1,2] ID cards are already
used to moderate permission to access restricted methods. Be
this since it might, protection could be broken off when an
unapproved prospect or maybe fake endeavours to abuse
them. Inconvenience with making use of conventional
solutions triggered fast increment from the usage of
biometrics. Biometrics are actually an instant identification
which alludes to a searching of decomposing, estimating as
well as evaluating an individual's persona influenced by the
social or physiological characteristics associated together
with the prospect [3]. You will find two types of biometrics
to become Multimodal, Unimodal, and specific. The typical
area biometric capabilities are palm prints, hand geometry,
discourse designs, iris designs, facial features, and
fingerprints. A biometric confirmation structure works within
2 modes, for instance, enlistment function as well as examine
function. Inside enlistment function, the device recognizes an
individual by looking the layouts of significant amount of
clientele within the data source for a match up. Within the
confirmation function, structure approves character of
specific by evaluating the found biometric details together
with the possession of biometric template(s) that are stored
method repository.
Biometric methods that rely on the evidence associated
with an one particular wellspring of information for
validation are actually labelled as Unimodal method [4].
Unimodal biometric methods endure an assortment of
problems, for instance, Commotion within detecting
information, Intra class variations, Uniqueness capability.
The confinements of unimodal biometric structure lead to
substantial False Acceptance Rate (FAR) along with False
Rejection Rate (FRR), limited splitting up skill, top bound
within delivery therefore the multimodal biometric product is
designed to satisfy the strict delivery demands. A biometric
os and that is determined by the nearness of several pieces of
evidence of specific identification is known as multimodal
biometric structure [5]. A biometric os and that is determined
by the nearness of several pieces of evidence of specific
identification is known as multimodal biometric phone
system. It fuses some variety of autonomous biometrics as
well as overcomes a percentage on the restrictions created by
making use of Unimodal biometric. Multimodal biometrics
are progressively important to phony advances, since it's a lot
more difficult to generate numerous biometric characteristics
when compared with fashioning an one particular biometric
trademark hence provides larger accuracy fee as well as from
spoofing steps by which makes it troublesome for an
interloper to simultaneously spoof the a number of biometric
attributes of a real customer [6]. The delivery markers for
specific recognition feature include things like False
Acceptance Rate (FAR) along with False Rejection Rate
(FRR).
False Acceptance Rate (FAR): When the biometric
telephone system improperly accepts a gain access to try of
an unauthorized pc user is actually recognized as False
Acceptance Rate (FAR). A system's FAR says when the ratio
involving the variety of phony acceptances as well as variety
of identification tries.
ܨ݈ܽݏ݁ܣܴሺΨሻൌ൬ܨܣ
ܰ൰כͳͲͲሺͳሻ
Wherever FA - Lots of incidents of phony acceptances, N
- Total amount of samples.
2020 International Conference on Emerging Smart Computing and Informatics (ESCI)
A
ISSMS Institute of Information Technology, Pune, India. Mar 12-14, 2020
978-1-7281-5263-9/20/$31.00 ©2020 IEEE 309
Authorized licensed use limited to: Cornell University Library. Downloaded on August 19,2020 at 04:33:25 UTC from IEEE Xplore. Restrictions apply.
False Rejection Rate (FRR): When the biometric
telephone system improperly rejects a gain access to try of an
authorized pc user is actually recognized as False Rejection
Rate (FRR).A system's FRR states since the ratio in between
quantity of phony acceptances as well as variety of
identification tries.
ܨܴܴሺΨሻൌ൬ܨܴ
ܰ൰כͳͲͲሺʹሻ
FR - Lots of incidents of phony acceptances. N - Total
amount of samples. Therefore, the equations (one) as well as
(two) are actually utilized for general performance analysis
within biometric methods.
With this paper, fingerprint or palm print documents
biometrics are actually utilized for multimodal biometric
phone system. The characteristics are actually extracted
making use of the distinction algorithm Deep Neural
Networks (DNN). Minutiae based mostly extraction as well
as detection is completed by using Multilayer Perceptron
(MLP) as well as Backpropagation Algorithm.
II. PROPOSED MODEL
The suggested method depends upon Multimodal
biometrics whereby all fingerprint as well as palmprint
biometric is actually used like a wellspring of authentication.
Fig. 1 suggests multimodal biometrics of palmprint as well
as fingerprint. They're viewed as essentially the most
popular, reliable as well as top biometrics. Automated
Fingerprint Identification System (AFIS) makes use of
methods generally determined by specifics concentrates.
Palmprint has edges, central lines, wrinkles, for example, and
features by what specifics concentrates could be extricated
[7]. Because of abundant floor information fingerprint as
well as palmprint are believed to be as incredible techniques
to come down with around house identification. Artificial
Neural Networks
(ANN), one of several Soft Computing methods [8] used for
Multimodal biometric identification.
Fig. 1. Multimodal Biometric
A. Image Acquisition
The info impression received is actually for starters
standardized to reduce variations to come down with dark
dimension esteems on the tips as well as valleys. Histogram
feature is actually carried out to enhance the clearness of this
picture [15]. Binarization is actually carried out to alter 8 bit
dim scope picture directly into twofold picture with zero
share for tips as well as one share for valleys. Advantage
segmentation is completed to determine advantage finishing,
edge department as well as distinguish return on first capital
expense. Within that time, the launch treatment is actually
carried out to split up the parallel picture in w *w squares
[8]. Fig. 2 shows the information stream chart for picture
securing.
Fig. 2. Flow Sequence of Image Acquisition
B. Feature Extraction
Extraction of appropriate feature of the brother ql is
easily the most crucial task. DNN is actually used for
element extraction. It's a distinction algorithm which is made
up of big amount of neurons (units) interconnected as levels.
Information is delivered only one means and possesses
absolutely no cycles or even sectors. The information
concentrates are actually divided by DNN show in Fig. 3
coming from the edges, valleys, wrinkles [14]. Deep-learning
networks and rule lines are actually realized using the much
more normal single concealed level neural networks by the
profundity of theirs; that's, the volume of hub levels whereby
info needs to go within a multistep process of instance
acknowledgment. Prior versions of neural networks, for
instance, the very first perceptron’s had been superficial,
produced from 1 info as well as one particular yield level, as
well as a maximum of 1 shrouded level within the centre.
Several levels (counting info as well as yield) qualifies as
"deep" learning [9]. Therefore deeply is not merely a
fashionable phrase to affect computations to show up as
although they come across Tune and also Sartre directly into
organizations you have not understood around just yet. It's a
meticulously recognized phrase which suggests greater than
a person shrouded level.
Within deep learning networks, every level of nodes
preps on an unmistakable plan of spotlights determined by
history layer's yield [15]. The additional you progress
straight into the neural web, the greater perplexing the
spotlights the nodes of yours are able to perceive, as they
total as well as recombine spotlights as a result of history
coating.
C. Minutiae Detection
Multilayer Perceptron (MLP) strategy is used to
distinguish the specifics, regardless of whether it's on
advantage department or even edge finishing. It contains
several levels, for instance, feedback, covered up as well as
yield levels [12]. The networking is actually divided using
Back propagation algorithm that is cooking or even
mastering algorithm in which a community process is
actually done by isolating the picture directly into 3*3
310
Authorized licensed use limited to: Cornell University Library. Downloaded on August 19,2020 at 04:33:25 UTC from IEEE Xplore. Restrictions apply.
windowpane square measurement, if the windowpane is
actually placed on specifics the self-esteem is actually one or
even different zero. If perhaps within pixel worth is actually
one and possesses simply 1 one self-esteem neighbor, then
simply particulars factor is actually on advantage completing
[11]. On the away possibility which center pixel worth is
actually one and possesses simply 3 one self-esteem
neighbor, then specifics factor is actually on advantage
department. With lengthy final, Euclidean splitting up is
actually discovered in between specifics concentrates and set
away within a repository.
Fig. 3. A Sample DNN
D. Minutiae Matching
Specifics coordinating are actually completed by
evaluating the specifics esteems place at bay within data
source using the specified info picture. Specifics scores are
actually produced having a stove someplace inside all the
different zero as well as one. If perhaps the coordinating
score is much more noteworthy compared to predefined cap
worth, the prospect is actually accredited mostly is a faker
[5].
III. SYSTEM IMPLEMENTATION
The structure consists of 2 phase forms: Identification
and also Enlistment. Fig. 4 suggests System Design for
Particulars Coordinating Based Individual Identification
System. Amid enlistment procedure, the prospect enlists
equally palmprint as well as fingerprint as character. The
functions, for instance, tips, valleys, main collections, lines
and wrinkles are actually extricated of specifics concentrates
are actually acquired. With, specifics recognition, the
information concentrates are actually divided as end as well
as bifurcation (edge endings) concentrates [10,11]. At this
point together with the information concentrates, the
Euclidean splitting up esteems are actually computed and set
away inside repository. Amid identification process,
fingerprint as well as palmprint are actually procured for
identification. Similar process occurs till particulars
recognition. With specifics coordinating, the coordinating
score esteems are actually computed by evaluating the
specifics contained in the specified information and also the
camera put away inside repository. The option is performed
influenced by the coordinating score. On the away possibility
which coordinating score tends to be more noteworthy
compared to the advantage worth, the prospect is actually
accredited as well as recognized. Another thing, prospect is
actually unapproved as well as rejected.
Fig. 4. System Architecture
IV. RESULTS AND DISCUSSIONS
Making use of Matlab, the pictures of all fingerprint as
well as palmprint are actually selected and set away within
an MS - Access repository. At this point amid the
identification process, the pictures are actually packed for
picture enhancement in which standardization, histogram
getting used to as well as advantage launch are actually
performed [10]. With the much better photos, the specifics
concentration are actually divided. Within the first coat level
the bifurcation as well as advantage endings are actually
realized by executing community process. Euclidean
splitting up computation is actually performed [13]. The
information esteems coordinating are actually completed by
computing the scores as well as validation is actually
examined. The delivery of Gabor channel switches for both
real component as well as fanciful component as much as
FAR and also FRR influenced by the amount of assessments.
The variations are showing up Table I as well as Table II.
The delivery on the suggested means with different
limitations, Tar, and they limit the phony accept pace as well
as bogus refuse pace is actually been seen [3]. Therefore, the
suggested strategy grounded techniques to enhance
distinction exactness involving frauds and also genuine
clientele the prediction of FAR and FRR shown in Table III.
TABLE I. RELIABILITY MEASURES FOR GABOR REAL PART
No. of Samples
FRR FAR ACCURACY
25 1.10 3.0 91.00
125 0.70 2.3 96.23
600 0.02 1.3 97.60
TABLE II. RELIABILITY MEASURES FOR GABOR IMAGINARY
PART
No. of Samples
FRR FAR ACCURACY
25 1.10 2.0 91.00
125 0.70 1.4 96.23
600 0.02 01.4 97.60
311
Authorized licensed use limited to: Cornell University Library. Downloaded on August 19,2020 at 04:33:25 UTC from IEEE Xplore. Restrictions apply.
TABLE III. TEST RESULTS ON DESIGNATED THRESHOLD
STANDARDS
Threshold Tar
False Accept Rate(%)
False Reject Rate(%)
0.325 0.00 1.46
0.335 0.00 0.81
0.345 0.27 0.55
0.355 0.82 0.55
0.365 2.40 0.29
0.375 5.26 0.11
V. CONCLUSION
Biometrics assumes a crucial task within around house
identification. The goal is actually actualizing a secured
unique identification device continues to be dependably
achieved by techniques for making use of Multimodal
Biometrics. Fingerprint as well as Palmprint that are wealthy
within floor information achieves top exactness. DNN
grounded distinction becomes excellent identification fee.
Together the collections, the recommended technique betters
noteworthy delivery as well as distinction accuracy within
factors that are fresh . The results provide far better accuracy
for your offered approach.
REFERENCES
[1] Shankar, K., Elhoseny, M., Kumar, R.S., Lakshmanaprabu, S.K. and
Yuan, X., 2018. Secret image sharing scheme with encrypted shadow
images using optimal homomorphic encryption technique. Journal of
Ambient Intelligence and Humanized Computing, pp.1-13.
[2] Shankar, K., Elhoseny, M., Chelvi, E.D., Lakshmanaprabu, S.K. and
Wu, W., 2018. An Efficient Optimal Key Based Chaos Function for
Medical Image Security. IEEE Access, 6, pp.77145-77154.
[3] A.K. Jain, A. Ross & S. Prabhakar, “An Introduction to Biometric
Recognition”, In: IEEE Transactions on Circuits and Systems for
Video Technology, Vol. 14, N. 1, pp 4-20, January 2004.
[4] Sengar, S.S. and Mukhopadhyay, S. (2017). Moving object detection
based on frame difference and W4. Signal, Image and Video
Processing, 11(7), pp.1357-1364.
[5] Mir. A.H, Rubab .S, Jhat .Z.A,” Biometric Verification: A Literature
Survey “, International Journal of Computing & ICT Research,
Vol.05, No.02, December 2011.
[6] Fawaz Alsaade,” Neuro-Fuzzy Logic Decision in a Multimodal
Biometric Fusion System”, Scientific Journal of King Faisal
University (Basic and Applied Sciences), Vol.11, No.02, 2010.
[7] MuthuKumar .A, Kasturi .C and Kannan .S, ”Multimodal Biometric
Authentication using Particle Swarm Optimization Algorithm with
Fingerprint and Iris”, ICTACT Journal on Image and Video
Processing, Vol.02, No.03, Feburary 2012.
[8] Sengar, S.S. and Mukhopadhyay, S. (2016). Moving object area
detection using normalized self adaptive optical flow. Optik, 127(16),
pp.6258-6267.
[9] K.Shanmugapriyal, M.Karthika, Dr.S.Valarmathy, M.Arunkumar
“Performance Evaluation of Contourlet Transform based Palmprint
Recognition using Nearest Neighbour Classifier” International
Journal of Emerging Technology and Advanced Engineering
Website: www.ijetae.com (ISSN 2250-2459, ISO 9001:2008 Certified
Journal, Volume 3, Issue 1, January 2013.
[10] Sengar, S.S., Mukhopadhyay, S. (2017). Foreground Detection via
Background Subtraction and Improved Three-Frame
Differencing, Arabian Journal for Science and Engineering, vol. 42,
pp. 3621.
[11] SitalakshmiVenkataraman,”A Grid-Based Neural Network
Framework for Multimodal Biometrics”, World Academy of Science ,
Engineering and Technology,2010.
[12] Murugan, B.S., Elhoseny, M., Shankar, K. and Uthayakumar, J.,
2019. Region-based scalable smart system for anomaly detection in
pedestrian walkways. Computers & Electrical Engineering, 75,
pp.146-160.
[13] Sengar, S. (2017). Detection of moving objects based on enhancement
of optical flow. Science Direct, Vol. 145, pp.131-141.
[14] Uthayakumar, J., Metawa, N., Shankar, K. and Lakshmanaprabu,
S.K., 2018. Financial crisis prediction model using ant colony
optimization. International Journal of Information Management.
[15] Sengar, S. S., and Mukhopadhyay, S. (2017). Motion detection using
block based bi-directional optical flow method. Journal of Visual
Communication and Image Representation, 49, 89-103.
312
Authorized licensed use limited to: Cornell University Library. Downloaded on August 19,2020 at 04:33:25 UTC from IEEE Xplore. Restrictions apply.
