EvaBio Platform for the Evaluation Biometric System
Application to the Optimization of the Enrollment Process for Fingerprints Devices
B. Vibert
1
, Z. Yao
1
, S. Vernois
1
, J-M. Le Bars
2
, C. Charrier
2
and C. Rosenberger
1
1
ENSICAEN, 3UMR 6072 GREYC, F-14032 Caen, France
2
UNICAEN, 3UMR 6072 GREYC, F-14032 Caen, France
Keywords:
Evaluation of biometric systems, Evaluation platform, Minutiae template selection, Fingerprint quality metric.
Abstract:
Nowadays, when someone wants to make a payment with a smartcard, the user has to enter a pin code to be
identified. Only biometrics is able to authenticate a user; yet biometric information is sensitive. To ensure
the security and privacy of biometric data, OCC (On-Card-Comparison) has been proposed. This approach
consists in storing biometric data in a secure zone on a smartcard and computing the verification decision in
a Secure Element (SE). The purpose of this paper is to propose an evaluation platform for testing biometric
systems such as the analysis of performance and security on biometric OCC. Based on two examples, we
illustrate its different uses in an operationnal context. The first example focus on the ”Quality module” which
allows to choose the enrollment by considering the fingerprint quality with one proposed metric. The second
one addresses the minutiae reduction of the fingerprint template when the number of minutiae is higher than
expected by the OCC.
1 INTRODUCTION
Nowaday, biometrics is often used in our daily life,
(passport, border control, smartphone ...). This kind
of applications requires in general the use of large on-
line biometric databases which may cause many se-
curity and privacy problems. In order to avoid these
problems, the secure storage of biometric data and
OCC verification are increasingly deployed on a SE
(Secure Element) such as the French passport chip.
The main benefit of this solution is to avoid the trans-
mission of the biometric reference template of the
user. The user has also the control of its own biomet-
ric data stored in the SE. A secure element guarantees
many security issues of the biometric reference (con-
fidentiality, integrity).
The SE is frequently used for several applications
such as border control or face to face bank payment.
Thus, to avoid misused identity for example, it be-
comes very important to define a general methodol-
ogy for evaluating these embedded systems. Some
standards have been proposed as guidelines of bio-
metric systems (ISO, c; ISO, b) but the definition of
a certification process is not yet done. Our lab has
a strong link with industrial companies and many of
them want to compare and evaluate OCC or sensor
in order to choose the best. We need a platform em-
bedding performance and security tools to analyze ex-
isting biometric systems and to help research in this
area. This platform should be modular, able to pro-
cess embedded biometric systems and strongly con-
nected with existing standards. This is why we pro-
posed in this paper, this evaluation platform of bio-
metric OCC for analyzing its performance and secu-
rity.
The paper is organized as follows. Section 2 is de-
voted to the state-of-the-art of evaluation platform of
biometric systems. Section 3 describes the proposed
platform by emphasing on specific modules. In Sec-
tion 4, we illustrate the benefit of the proposed plat-
form through two examples of uses cases. We con-
clude and give some perspectives in Section 5.
2 STATE-OF-THE-ART
In the literature, only few platform exist for assess-
ing the performance and security of biometric sys-
tems. We can cite the NIST platform (Grother et al.,
2011), which is used in many research competitions.
It allows researchers or manufacturers to test their
OCC or minutiae extractors, in term of interoperabil-
329
Vibert B., Yao Z., Vernois S., Bars J., Charrier C. and Rosenberger C..
EvaBio Platform for the Evaluation Biometric System - Application to the Optimization of the Enrollment Process for Fingerprints Devices.
DOI: 10.5220/0005244203290335
In Proceedings of the 1st International Conference on Information Systems Security and Privacy (ICISSP-2015), pages 329-335
ISBN: 978-989-758-081-9
Copyright
c
2015 SCITEPRESS (Science and Technology Publications, Lda.)
ity. NIST reports disseminate information on FMR
(False Match Rate) and FNMR (False Non Match
Rate) for each OCC and extractor. The purpose of this
platform is to compare existing algorithms or systems
by a trusted third party.
We can also mention the online FVC-Ongoing
platform (Biolab, 2009) dedicated to algorithms for
fingerprint verification (evolution of the FCV com-
petitions). The platform offers multiple databases
grouped into two parts. The first one (Fingerprint
Verification) quantifies both enrollment and verifica-
tion modules, while the second one (ISO Fingerprint
Matching) quantifies only the verification module on
ISO Templates (ISO, a) based on minutiae. Perfor-
mance metrics are: the failure to acquire rate (FTA)
and the failure to enroll rate (FTE), the false non
match rate (FNMR) for a defined false match rate
(FMR) and vice versa, the average enrollment and
verification times, the maximum size required to store
the biometric template on the SE, the distribution of
legitimate and impostors users scores and the ROC
curve with the associated equal error rate (EER). The
main drawback of this platform is that it is necessary
to submit the executable or source code of the OCC
algorithm to the online platform which can cause con-
fidentiality issues.
Another platform is actually in development
within the BEAT (Biometric Evaluation And Testing)
European project (Project, 2013). At the end of the
project, a framework to evaluate the performance of
biometric technologies using several metrics and cri-
teria (performance, vulnerabilities, privacy). The goal
of this project is to have a common platform for the
industrial and researchers to evaluate their products
and to have an independent and certified result with
common criteria. This platform is not yet released
actually and does not focus embedded biometric sys-
tems.
We have seen the main platforms in the state-of-
the-art and we have presented their possibilities and
drawbacks. However, no platform answer to our cri-
teriae (usability, modularity,...). This is why we have
decided to develop our platform we present here.
3 EvaBio PLATFORM
EvaBio platform permits to make the link between in-
dustrial companies and researcher. This strong link
with industrial permits to improve the platform to re-
spond to their needs and also permit to share results
with academics. Figure 1 summarize this idea.
Figure 1: Links between EvaBio platform with industrial
and research.
3.1 General Scheme
The general synopsis of the proposed platform is
given Figure 2. This evolution of our first platform
(Vibert et al., 2013) allows us to have more functional
modules such as Sensor, Computing, Quality metrics,
Security for OCC and Audit. The new modules yield
to developers or researchers to have different kinds of
methods to evaluate OCC or to choose a sensor.
3.2 Modules
The platform is composed of different modules with
specific treatments, and all modules are independent.
This modularity allows us to modify a module with-
out changing the overall operation of the platform.
For example we can quantify the benefit of quality
checking of the fingerprint during the enrollment
process. The platform uses active mechanisms of
communication by event allowing multiple modules
simultaneously access data exchanged between the
client application and the OCC, thus offering ”on the
fly” analysis of results. All the main modules such as
Core, Scenario, Performance, GUI interface
are explained in a previous paper(Vibert et al., 2013).
In this paper, only Sensor, Computing, Evaluation,
Fingerprint quality assessment modules are
described.
The Sensor module is a little platform which
permits to acquire real and fake fingerprint databases
with real finger and specific protocols. This module
is used to evaluate the performance of a sensor and
to provide attacks on it. We also went on mortuary
to test if the sensor is able to acquire dead fingerprint
(Vibert et al., 2014). This sensor platform could be
used in input on Core, to acquire in live one or more
fingerprints to compare on an OCC algorithm.
Computing module yields to have a distributed
computation to improve the efficiency of the evalua-
tion of OCC. For example, from three OCC on three
smartcards, we are able to run three different tests
in parallel. We divide by three the evaluation time
within a campaign.
ICISSP2015-1stInternationalConferenceonInformationSystemsSecurityandPrivacy
330
Figure 2: General scheme of the EvaBio platform.
Fingerprint quality assessment module is
devoted to the quality metric of fingerprint images
or minutiae templates. Fingerprint quality metric
is an auxiliary solution to guarantee the matching
performance by dropping the bad quality samples
(Grother and Tabassi, 2007) in both the enrollment
and matching sessions. This purpose can be simply
achieved because good fingerprint quality could
provide more precise and reliable features. Obvi-
ously, this is also beneficial to the OCC operations,
especially when it is necessary to consider minutiae
selection. There is a much higher probability that
a minutiae extractor can correctly localize minutiae
points within good quality images than that within
bad quality prints (Chen et al., 2005). Therefore,
a reduced minutiae template can preserve correctly
detected minutiae as much as possible rather than
the spurious points, and the performance could be
ensured as well. The quality metric module in the
platform is combined with a validation component
which allows the user to measure the performance
of variant metrics, which enables making a further
decision to choose an appropriate metric.
Figure 3, presents two examples of an im-
age quality metrics distribution involved by the
Fingerprint quality assessment module : 1)
NFIQ (Tabassi and Wilson, 2005) and 2) GREYC Q
metric. The NFIQ generates five quality levels from 1
to 5 (Figure 3(a)), where the best quality is indicated
by the lowest value and the maximum level denotes
(a)
(b)
Figure 3: Image quality distribution for NFIQ and Q.
the samples of very poor quality. GREYC Q metric
(El-Abed et al., 2011), estimates the quality of finger-
print with five score groups (Figure 3(b)), poor (0-20),
EvaBioPlatformfortheEvaluationBiometricSystem-ApplicationtotheOptimizationoftheEnrollmentProcessfor
FingerprintsDevices
331
bad (20-40), medium (40-60), good (60-80) and very
good (80-100). Such a continuous quality score could
generate a better distribution of sample qualities than
those using only few quality levels. This module is
also a modular unit so that other quality metrics are
also employable in the experiments.
The Evaluation module used metrics commonly
used in the literature and ISO (ISO, c) with more spe-
cific ones:
False Match Rate (FMR): it measures how many
times the biometric data of a user provides pos-
itive verifications with biometric data of another
user.
False Non Match Rate (FNMR): it measures how
many times the biometric data of a user gives a
negative verification of biometric data with the
same user,
Success rate of Attack: it measures the ratio of
successful attacks (number of positive result over
a number of transactions).
Measuring Interoperability: it quantifies the ratio
of successful tests when providing an ISO tem-
plate to the OCC.
ROC Curve: It describes the behavior of the bio-
metric OCC for each value of the decision thresh-
old (from which a test is positive). This implies
that it is possible to obtain the comparison score
from the OCC or to set decision threshold. For
industrial OCCs, this is rarely the case but for re-
search ones, this information is always available.
Verification Time: we measure the time required
to achieve a OCC enrollment or to obtain a veri-
fication result (after sending the ADPU (Applica-
tion Data Protocol Unit defined in (ISO, d)) to the
SE. It is also possible to generate several statistics
on computation times such as histogram verifica-
tion time, average, minimum or maximum time.
4 EXAMPLE OF USES CASES
In this section, we present experimental results on
a commercial OCC with the selection of enrollment
template when we have the quality of the original im-
age.
4.1 Quality Checking During
Enrollment
In this study, a method which permits to choose an en-
rollment template with the best quality and the max-
imum number of minutiae accepted by the OCC has
been proposed. This approach is tested with NFIQ
and Q metrics, and we obtain a better result than be-
fore only with the selection of enrollment template.
Figure 4, illustrates how we choose a template with-
out quality selection 4(a) and when we use a quality
assessment process 4(b).
(a)
(b)
Figure 4: Selection protocol on the reference template en-
rollment with and without quality checking where one col-
umn corresponds to samples from an individual.
Concerning the protocol, the used biometric data
have been collected in earlier experiments with 39 in-
dividuals. We have made three capture sessions with
two fingers : left and right index finger, with five cap-
tures per session per individual. In total, we captured
1170 fingerprint images and ISO Compact Card tem-
plates. On each image, we compute the NFIQ and Q
quality metrics an processed by the evaluation mod-
ule of the EVABIO platform. To choose the reference
templates, we have selected, for each person, the tem-
plate with the best quality value considering the two
metrics and the most number of minutiae under the
maximum size allowed by the OCC.
The performance without template selection is
also computed (Vibert et al., 2013). In Table 1, we
present the results of the comparison on a commer-
cial OCC with or without quality selection.
Table 1: Performance values for each quality metric selec-
tion method.
FMR FNMR
Without selection 0.41% 17.36%
NFIQ selection 0.05% 14.36%
Q selection 0.003% 4.75%
As a conclusion of the performance, there is a
fairly good robustness to imposture for all methods
ICISSP2015-1stInternationalConferenceonInformationSystemsSecurityandPrivacy
332
and the FNMR is relatively good with Q selection but
not with others. We observed quite satisfactory results
without quality selection, in terms of false match rate
compared to those found in (Grother et al., 2011) and
a very good performance with quality selection. The
false non match rate appears too high, however by im-
proving the selection of the enrollment template, we
are able to reduce the FNMR around 10% with NFIQ
selection and an other 10% with Q selection in com-
parison with NFIQ. This experiment shows that the
selection of enrollment template is very important to
achieve good performance.
In the EVABIO platform, a quality metric is used
as filter to choose the reference template for each user
in the database. We can quantify the gain of opera-
tional quality checking during the enrollment process.
It also helps us to improve quality metrics by consid-
ering on different databases similar experiments de-
tailed in Table 1.
4.2 Minutiae Selection
We also have developed a module to reduce the ISO
Compact Card template when we have too many
minutiae. The truncation method defined in the state-
of-the-art (Grother and Salamon, 2007) has been ini-
tially embedded in the module. To determine if the
truncation is the best method in computation time and
performance, two methods have been tested.
4.2.1 No Selection
The first method keeps all minutiae from the initial
template. The performance associated to the initial
template is used as reference for the experimental re-
sults. We expect that reduced templates could lead to
lower performances than the initial template.
4.2.2 Selection by Truncation
This method is based on a simple truncation i.e., we
only keep minutiae from the initial template the first
N
max
minutiae. The efficiency of this simple approach
depends on the method used to generate the finger-
print template. For many commercial biometrics sys-
tems, a fingerprint template is generated with a spe-
cific method. It can be generated considering minu-
tiae with the ascending locations Y as for example.
In the case where multiple captures have been made,
high quality minutiae (always present in the different
captures as for example) can be placed at the begin-
ning of the template. Selecting the N
max
first minutiae
could be in this case a very efficient and simple.
4.2.3 Barycentre Selection
This method based on a pruning mechanism is sim-
ple and fast (few milliseconds). It has been proposed
by the NIST for minutiae selection in (Grother and
Salamon, 2007). It has been shown that minutiae lo-
cated near the core of a fingerprint minutiae are the
most useful ones for the matching process (Weiwe
and Wang, 2002). Given a fingerprint template, the
core location is usually unknown. However, the cen-
troid of minutiae can be a good estimate (when no
other information is available). This minutiae selec-
tion approach tends to only keep minutiae near the
centroid for this reason. We have four steps for the
computation process:
1. Compute the centroid of the minutiae from the fin-
gerprint template (containing N
j
minutiae);
Centroid = (X
c
, Y
c
) =
1
N
j
(
N
j
i=1
X
i
,
N
j
i=1
Y
i
) (1)
2. Compute the distance of each minutiae to the cen-
troid;
r
i
=
q
(X
i
X
c
)
2
+ (Y
i
Y
c
)
2
, i = 1 : N
j
(2)
3. Sort in ascending order minutiae according to the
distance r
i
, i = 1 : N
j
;
4. Select the first N
max
minutiae.
4.2.4 Performance Evaluation
Concerning the protocol, we have used the
FVC2002DB2 (Maio et al., 2002) database to
illustrate results. This database is composed of 8
fingerprints per person and 100 individuals, with
a total of 800 fingerprints. All minutiae templates
used in the experiments have been extracted using
the NBIS tool, MINDTCT (Watson et al., 2007)
from the NIST. In order to realize the matching of
fingerprint templates, we used a very well known
minutiae matching algorithm proposed in 1997 by
Jain et al (Jain et al., 1997). This method consists of
an alignment stage (translation and rotation estima-
tion between the two templates to compare) and a
matching stage after transformation.
To evaluate the performance of minutiae selec-
tion algorithms, we use the AUC (Area Under the
Curve) metric since it is often considered as global
performance criterion. We use this value to quantify
the efficiency of a minutiae selection method. We
compute the AUC value for each selection method
with N
max
varying from 30 to 50 by step of 2. The
Confidence Interval (CI) is also used to weight the
EvaBioPlatformfortheEvaluationBiometricSystem-ApplicationtotheOptimizationoftheEnrollmentProcessfor
FingerprintsDevices
333
Table 2: AUC values for each minutiae selection method for different values of N
max
on FVC2002DB2.
N
max
30 34 38 42 46 50
No selection (%) 11.2 ± 0.15 11.2 ± 0.15 11.2 ± 0.15 11.2 ± 0.15 11.2 ± 0.15 11.2± 0.15
Truncation (%) 10.2 ± 0.28 9.97 ± 0.2 9.29 ± 0.14 8.93 ± 0.11 9.41 ± 0.07 9.48 ± 0.05
Barycentre (%) 8.73 ± 0.31 9.01 ± 0.2 9.00 ± 0.15 9.26 ± 0.10 9.17 ± 0.07 9.47 ± 0.04
Figure 5: Example of minutiae selection on a fingerprint sample: stars represent selected minutiae by the different methods.
For the barycenter selection approach, the green point represents the estimated CORE point (barycenter of minutiae).
Figure 6: Evolution of the AUC value face to minutiae selection on FVC2002DB2.
results of AUC value.
We show in Figure 5 the results of minutiae
selection using the two tested methods. Selected
minutiae are represented by a red star and others
with a blue circle. We can see with the barycentre
approach, select minutiae are near the estimated
CORE. With the truncation method, we loose the
right part of the template.
Table 2 details the AUC value for each minutiae
selection method for the FVC2002 DB2 database.
On this database, the two selection methods permit to
obtain a better performance (Cf. Figure 3) compared
to the initial template (no selection).
To conclude with this illustration, we observe than
barycenter is better than truncature method which is
even the standard method. This use case illustrate
ICISSP2015-1stInternationalConferenceonInformationSystemsSecurityandPrivacy
334
the interest of the EVABIO platform. The minutiae
selection method can be seen as a preprocessing to
the enrollment process in an operational application.
The platform allows to test in a very convenient way
other selection methods.
5 CONCLUSIONS
In this paper, we have presented the benefits of the
EVABIO biometric evaluation platform, and we have
illustrated with two examples the capability of the
platform. The first one quantifies how the use of bio-
metric quality metrics on enrollment template selec-
tion is influenced performance. The second is a brief
comparative study of fingerprint minutiae selection
algorithms. To conclude, we demonstrate the facil-
ity to obtain results with the proposed platform.
In perspective, we plan to develop new modules
to evaluate the OCC and sensor on smartphone and to
design new attacks on OCC and sensors. We will also
improve the scenario module to propose new tests.
REFERENCES
ISO/IEC 19795-2. information technology - biometric data
interchange format - part 2: Finger Minutiae data,
2004.
ISO/IEC 19795-7. information technology - biometric per-
formance testing and reporting - part 7: testing of on-
card biometric comparison algorithms, 2011.
ISO/IEC 2382-37. Information technology - vocabulary -
part 37: Biometrics, 2012.
ISO/IEC 7816-1 to 15: Identification cards - Integrated cir-
cuit(s) cards with contacts(Parts 1 to 15). ISO/IEC,
http://www.iso.org.
Biolab (2009). FVConGoing. https://biolab.csr.unibo.it/
FVCOnGoing.
Chen, Y., Dass, S. C., and Jain, A. K. (2005). Fingerprint
quality indices for predicting authentication perfor-
mance. In Audio-and Video-Based Biometric Person
Authentication, pages 160–170. Springer.
El-Abed, M., Hemery, B., Charrier, C., Rosenberger, C.,
et al. (2011). Evaluation de la qualit
´
e de donn
´
ees
biom
´
etriques. Revue des Nouvelles Technologies de
l’information (RNTI), pages 1–22.
Grother, P. and Salamon, W. (2007). Interoperability of
the ISO/IEC 19794-2 compact card and 10 ISO/IEC
7816-11 match-on-card specifications 11.
Grother, P., Salamon, W., Watson, C., Indovina, M., and
Flanagan, P. (2011). Minex ii ”performance of fin-
gerprint match-on-card algorithms” phase iv : report
NIST interagency report 7477 (revision ii).
Grother, P. and Tabassi, E. (2007). Performance of biomet-
ric quality measures. Pattern Analysis and Machine
Intelligence, IEEE Transactions on, 29(4):531–543.
Jain, A. K., Hong, L., Pankanti, S., and Bolle, R. (1997).
An identity-authentication system using fingerprints.
Proceedings of the IEEE, 9:1365–1388.
Maio, D., Maltoni, D., Cappelli, R., Wayman, J. L., and
Jain, A. K. (2002). FVC2002: Second fingerprint ver-
ification competition. In Pattern Recognition, 2002.
Proceedings. 16th International Conference on, vol-
ume 3, pages 811–814. IEEE.
Project, B. (2013). Beat project. https://www.beat-eu.org/.
Tabassi, E. and Wilson, C. L. (2005). A novel approach to
fingerprint image quality. In Image Processing, 2005.
ICIP 2005. IEEE International Conference on, vol-
ume 2, pages II–37. IEEE.
Vibert, B., Lebouteiller, J., Keita, F., Rosenberger, C., et al.
(2014). Biometric sensor and match-on-card evalua-
tion platform. In International Biometric Performance
Testing Conference (IBPC).
Vibert, B., Rosenberger, C., and Ninassi, A. (2013). Secu-
rity and performance evaluation platform of biometric
match on card. In Computer and Information Technol-
ogy (WCCIT), 2013 World Congress on, pages 1–6.
IEEE.
Watson, C. I., Garris, M. D., Tabassi, E., Wilson, C. L., Mc-
cabe, R. M., Janet, S., and Ko, K. (2007). Users guide
to nist biometric image software (nbis). Technical re-
port, NIST.
Weiwe, Z. and Wang, Y. (2002). Core-based structure
matching algorithm of fingerprint verification. Inter-
national Conference on Pattern Recognition.
EvaBioPlatformfortheEvaluationBiometricSystem-ApplicationtotheOptimizationoftheEnrollmentProcessfor
FingerprintsDevices
335