Improving the accuracy of fingerprinting system using multibiometric approach

Safa M. AL-Taie Int. Journal of Engineering Research and Application www.ijera.com
ISSN : 2248-9622, Vol. 6, Issue 7, ( Part -4) July 2016, pp.15-19
www.ijera.com 15|P a g e
Improving the accuracy of fingerprinting system using
multibiometric approach
Safa M. AL-Taie
Electrical and computer engineering, Florida Institute of Technology, Melbourne, Fl, USA
ABSTRACT
Biometric technology is a science that used to verify or identify the individual based on physical and/or
behavioral traits. Although biometric systems are considered more secure than other traditional methods such as
password, or key, they also have many limitations such as noisy image, or spoof attack. One of the solutions to
overcome these limitations, is by applying a multibiometric system. Multibiometric system has a significant
effect in improving the performance of both security and accuracy of the system. It also can alleviate the spoof
attacks and reduce the fail to enroll error. A multi-sample is one implementations of the multibiometric systems.
In this study, a new algorithm is suggested to provide a second chance for the genuine user who is rejected, to
compare his/her provided finger with the other samples of the same finger. Multisampling fingerprint is used to
implement this new algorithm. The algorithm is activated when the match score of the user is not equal to a
threshold but close to it, then the system provides another chance to compare the finger with another sample of
the same trait. Using multi-sample biometric system improved the performance of the system by reducing the
False Reject Rate (FRR). Applying the original matching algorithm on the presented database produced 3
genuine users, and 5 imposters for the same fingerprint. While after implementing the suggested condition, the
system performance is enhanced by producing 6 genuine users, and 2 imposters for the same fingerprint. This
work was built and executed depending on a previous Matlab code presented by Zhi Li Wu. Thresholds and
Receiver Operating Characteristic (ROC) curves computed before and after implementing the suggested
multibiometric algorithm. Both ROC curves compared. A final decision and recommendations are provided
depending on the results obtained from this project.
Keywords: Biometrics, Multibiometrics, Multi-sample, FRR, FAR, ROC curves.
I. INTRODUCTION
A biometric system determines one or more
physical or behavioral traits, including fingerprint,
face, iris, voice, signature, gait, hand vein, odor, and
DNA. The set of attributes that is related to a specific
person represents hisher identity, and gives the
system the ability to identify the uniqueness of that
individual.
There is no doubt that a multibiometric
system plays a critical role in most applications and it
receives a lot of attention due to their advantages in
overcoming limitations in unibiometric systems.
Applying multibiometric system can improve the
accuracy of authentication system rather than single
biometric system. Using multi algorithm to process
the same biometric traits can minimize the cost,
because it does not require an additional sensor to
capture those traits. (Gokberk et al., 2005) combined
multiple algorithms for 3D face recognition.
In order to recognize the individual, the
most important task: is to establish the relation
between the individual and his identity. Person
identity refers to the ability of verifying the claim of
the user. (Todorov, 2007) To verify the identity of
any individual, three fundamental methods can be
used: (a) “what he knows”: this method depends on
the fact that the person has a special knowledge of a
secret information (e.g., password, personal ID
number, cryptographic key), (2) “what he
possess extrinsically”: this method relies on the fact
that the person has exclusive possession of the
external token (e.g., passport, driver license, personal
device as mobile phone), (3) “Who he is
intrinsically”(biometric recognition): this method
establish the association between the individual and
his/her identity based on his/her physical inherent or
behavioral traits.
In general, biometric recognition is defined
as the science that establish the identity of the
individual based on his/her physical or behavioral
traits. Biometric recognition can be either fully
automated (i.e. without human intervention) or semi-
automated (i.e. with human intervention).
II. METHODOLOGY
While a biometric system with the single
biometric trait (unibiometric) can improve the
security of an application in a significant way, it can
also be the weakest point in this system. Usually
when a biometric is chosen to be the trait that will
determine users’ originality, because of the unique
properties for this trait. Unibiometric systems can be
exposed to many limitations in the different
RESEARCH ARTICLE OPEN ACCESS

biometric system levels figure (1). On the other hand
implementing multibiometrics can insure both
security and accuracy in the authentication system.
Combining multiple techniques can improve the
performance of the authentication system [1].
A. Biometric Used In The Project
In this project Fingerprint Minutiae were
chosen to be the Biometric traits. It has been chosen
because it presents the most Biometric trait adopted
by commercial applications among other traits,
especially with facilities that require speed more than
accuracy [2]. Fingerprints can exhibit sufficient
uniqueness, also fingerprint can be considered
Immutable biometrics [3].
B. Data Base implemented in the project
The first phase of this project was choosing
the DataBase. This is an important step to make sure
that all fingerprints included in this project are
obtained under the same conditions, which is an
important factor in the experimental work. In this
project the FVC2004 DataBase was chosen, it was
obtained from thermal sweeping sensor "FingerChip
FCD4B14CB" by Atmel. All fingerprints images are
of 300x480 (144 K pixels) with 512 dpi resolution
[4]. This fingerprint DataBase contains (80) samples
as each finger of the two hands has (8) samples.
Application used to implement the project
The fingerprint reader and minutiae
extractor implemented by Zhi Li Wu (2003) was
chosen to implement this project. Zhi Li Wu software
was coded in the Matlab platform, including all
functions required to read fingerprint scans to the
real minutiae extractor phase, and getting matching
score results [5]. This software was originally
designed to provide only match scores after
comparing two minutiae properties files. In our
project we added extra stages in the matching
process. These additional stages followed specific
conditions, where these conditions reflected the
proposed algorithm to enhance the performance of
the system in the False Reject Rate (FRR).
C. The Receiver Operating Characteristic
(ROC) curve
The Receiver Operating Characteristic
(ROC) curve, is a graphical presentation of the
system performance. In biometric science the ROC
curve will represent the false rates of the system,
including both False Reject Rate (FRR), and False
Accept Rate (FAR) [6]. The main question in this
experiment was: Can the False Reject Rate (FRR) be
enhanced if the system algorithm was modified at the
matching phase? First to answer this question we
draw the ROC curve for the fingerprint recognition
system before modifying the match process. The
second step was to modify the algorithm by
considering a second chance for scores of match
process which are close to the threshold. So instead
of providing final decision (reject) for scores which
are close to the threshold, a suggestion for the user
asking him/her to provide his/her fingerprint again to
be compared it with another sample which is stored
in the template in the database. This second chance
provided the user an opportunity to overcome some
physical problems that can affect the recognition
process. These problems led to provide less
information to the sensor so the sensor refused the
genuine user, and caused in a False negative (FN). In
this project we proposed to use a multi-sampling
biometric system as a second chance in case of
having close but not the requested matching score
algorithm as shown in figure (2). According to [7],
the following terms are defined as:
Th mean: refers to Th of the system.
Thj : refers to Th value at each sample of the finger.
Smin: refers to the minimum matching score.
Smax: refers to the maximum mathing score.
T: refers to the total number of the samples for the
same finger.
L0: refers to the match score of imposter class.
L1: refers to the match score of genuine class.
L= (L0+L1) the total number of the scores.
Si: refers to the score at each sample.
Figure (1) the original algorithm

Figure (2) the proposed algorithm
Before implementing the proposed algorithm
 The first set of fingerprints with 8 image
samples were all read, enhanced, features were
extracted, and stored in a*.m file for further
matching.
 Calculating the threshold of the system by
implementing the mean deviation equation for
all matching scores obtained during the
matching phase, using equation (1).
 Classifying both genuine L1 match scores
(L1=3) and L0 imposter match score (L0=5),
with a total number of match score L=L0+L1=8.
 Generating a set of thresholds (thj) j=1-T as
shown in table 1, where Smax> (thj)>Smin,
where Smax is the maximum matching score,
and Smin is the minimum matching score. All
thresholds were equally placed by implementing
both equation (2) and equation (3).
 Both FAR and FRR were calculated at each
threshold, using equation (4) and equation (5)
respectively, as shown in table 2 and table 3.
 Connecting the set points of FAR and FRR, to
obtain the ROC curve, as shown in figure (3).
 Determining the best operating point for such
system, with the lowest FRR and FAR. Which
produced FAR and FRR (0.2, 0.33).
Table
I
Table II
Table III
After implementing the suggested algorithm
The second set of fingerprints with 8 image samples,
and all images were read, enhanced, features were
extracted and stored in a*.m file for the further
matching process.
Calculating the threshold of the system by
implementing the mean deviation equation for all
matching scores which are obtained during the
matching phase, using equation (1), and adding the
new condition of matching scores that are near to
threshold and not equal to it.
Figure (3)
Thj Equation Th
Th1 38.3+(1-1)*1.9 38.3
Th2 38.3+(2-1)*1.9 40.2
Th3 38.3+(3-1)*1.9 42.1
Th4 38.3+(4-1)*1.9 44
Th5 38.3+(5-1)*1.9 45.9
Th6 38.3+(6-1)*1.9 47.8
Th7 38.3+(7-1)*1.9 49.9
Th8 38.3+(8-1)*1.9 51.6
Thj FAR
1 Th(30) 1
2 Th(37.85) 0.5
3 Th(45.7) 0
4 Th(53.55) 0
5 Th(61.4) 0
6 Th(69.25) 0
7 Th(77.1) 0
8 Th(84.95) 0
Thj FAR
1 Th(38.3) 1
2 Th(40.2) 0.2
3 Th(42.1) 0
4 Th(44) 0
5 Th(45.9) 0
6 Th(47.8) 0
7 Th(49.9) 0
8 Th(51.6) 0

Table IV
 Classifying both genuine L1 match scores L1=6,
and L0 imposter match scores L0=2, with a total
number of match scores L=L0+L1=8.
 After implementing the proposed algorithm,
both FAR and FRR were calculated at each
threshold using equation (4) and (5) respectively,
as shown in table 4 and table 5.
 Connecting the set points of FAR and FRR to
obtain the ROC curve after implementing the
suggested algorithm, as shown in figure (4).
 Determining the best operating point for the
system, with the lowest FAR and FRR. In this
case, the best point of the system represented by
FAR and FRR (0.5, 0.16).
Table V
Figure (4)
III. Results And Discussion
While it is important for some environments to
have a high security level, it also can produce
positive and negative results. Positive results, will
ensure that only genuine users allowed to pass the
biometric authentication test. On the other hand, the
negative results will increase the potential of refusing
the genuine users. The suggestion of modifying the
original algorithm and adding a new condition,
enhanced the performance of the system by reducing
the FRR errors. The new proposed condition, tends
to provide a second chance to the genuine user who
was falsely rejected, by comparing his/her fingerprint
with other samples of the same fingerprint. This
condition is implemented when the match score of
any individual is not equal to a threshold value but
close to it. Using multibiometric system is better than
using single biometric, because the individual can
have more samples or more traits to use, in case of
any error occurred. This provided another chance to
the user instead of rejecting him/her at first time. The
results show that applying the presented database to
the original algorithm (before modification) provided
3 genuine users, and 5 imposters for the same
fingerprint. While after modifying the algorithm and
implementing the suggested condition, then the
system obtained 6 genuine users, and 2 imposters for
the same fingerprint. This is a good evidence that the
performance of the biometric system is improved and
enhanced.
Both systems are examined with the ROC curve,
by plotting FAR and FRR before and after the
suggested condition. The observed figure (5) shows
that the value of FRR is improved after
implementing the condition. The area under the
curve for ROC2 is closer to the (x-axes) compared to
ROC1. So the performance of system 2 with (ROC2)
is better than system1 with (ROC1).
This additional stage decreased the FRR
value from 0.33 to 0.16, and the authentication
system is improved by reducing the number of False
Negative (FN) cases. These results can be applied to
the authentication systems that do not need a high
security level, where the time factor is more
important than the security level for the system. A
system having this performance is recommended for
facilities that will have a large input number of
people to authenticate, where time becomes more
precious than results. Administrator for such systems
would prefer to have such performance and such
error handling, to overcome more complicated
problems in the future. On the other hand, users that
will enroll in such authentication systems will have
additional chances to compare the same provided
fingerprint with other samples (in case of verifying
the proposed condition) before they are considered as
imposters.
Enrollment time will be a critical phase in
such system, usually in a normal circumstances,
users have to provide the biometric once or twice to
enroll in the system. In this project, an additional
time required during the enrollment phase.
Depending on results, the more samples that users
Thj FRR
1 Th(30) 0.16
2 Th(37.85) 0.16
3 Th(45.7) 0.5
4 Th(53.55) 0.83
5 Th(61.4) 0.83
6 Th(69.25) 0.83
7 Th(77.1) 0.83
8 Th(84.95) 0.83
Thj FRR
1 Th(38.3) 0.33
2 Th(40.2) 0.33
3 Th(42.1) 0.66
4 Th(44) 0.66
5 Th(45.9) 0.66
6 Th(47.8) 0.66
7 Th(49.9) 0.66
8 Th(51.6) 1

provided in the enrollment phase, provide the best
results to the recognition system, especially in the
case of the user is not rejected at the first time.
Figure (5)
References
[1]. Norman Poh, Samy Bengio and Jerzy Korczak,
“A multi-sample multi-source model for
biometric authentication” IEEE International
Workshop on Neural Networks for Signal
Processing (NNSP), 2002.
[2]. Arun Ross, Norman Poh, “Multibiometric
systems: overview, case studies, and open
issues” 273- 292. In Handbook of remote
biometrics, 2009.
[3]. Anil K. Jain, Arun A. Ross, Karthik
Nandakumar, “Introduction to biometrics” 52-
53. Springer Science+Business Media, LLC,
2011.
[4]. Fingerprint verification competition, (2004).
Available:
http://bias.csr.unibo.it/fvc2004/default.asp
[5]. Zhi li wu, “Fingerprint recognition”, Available:
http://www.comp.hkbu.edu.hk/~vincent/resPape
r.htm
[6]. Kar-Ann Toh∗, Jaihie Kim, Sangyoun Lee,
“Maximizing area under ROC curve for
biometric scores fusion” Pattern Recognition,
2008.
[7]. Giorgio Giacinto, Fabio oli, Roberto Tronci,
“Score Selection Techniques for fingerprint
Multi- modal Biometric Authentication” (2005).

Improving the accuracy of fingerprinting system using multibiometric approach

More Related Content

What's hot (18)

Similar to Improving the accuracy of fingerprinting system using multibiometric approach (20)

Recently uploaded (20)

Improving the accuracy of fingerprinting system using multibiometric approach