(2010) Fingerprint recognition performance evaluation for mobile ID applications

Fingerprint Recognition Performance Evaluation for Mobile 10 Applications
Shimon Modi, Ph.D.
BSPA Lab
Purdue University 401 N. Grant Street
West Lafayette, IN, 47907
USA
Ashwin Mohan
BSPA Lab
USA
Abstract - According to a report by Frost and Sullivan in
2007, revenues for non-AFIS fingerprint devices in notebook
PC's and wireless devices is anticipated to grow from $148.5
million to $1588.0 million by 2014, a compound annual growth
rate of 40.3% [1]. The AFIS market has a compound annual
growth rate of 15.2% with revenues of $445.0 million in 2007.
With the development of mobile applications in a number of
different market segments, such as healthcare, retail, and law
enforcement, this paper analyzed the performance of
fingerprints of different sizes, from different sensors
(commercialy available optical and capacitance) which were
generated in accordance with NIST Mobile 10 best practices
document using an automated image cropping process. The
minutiae count, image quality scores, false non match rates
(FNMR) and false match rates (FMR) were evaluated for
images of seven different image sizes, ranging from
0.098"XO.126" to 0.578"XO.618". The results provide insight
into constraints of fingerprint image sizes, interoperability and
fingerprint matching of different sizes in a mobile 10
environment.
Index Terms - mobile 10 biometrics, fingerprint recognition
performance, minutiae count.
I. INTRODUCTION
The need for stronger authentication and identity
management has lead to an increased use of biometric
technologies. Traditional authentication technologies like
passwords and tokens are based on something that a person
knows or possesses, respectively. Biometric technologies use
behavioral and biological characteristics of humans for
authentication [2]. Biometric technologies have been used in
law enforcement applications ever since Henry formulated his
system of fingerprint recognition in the 1890's [3]. Traditionally
law enforcement agencies have captured fingerprints when an
individual is brought to the booking station which allows for
greater control over the fingerprint acquisition process. Today,
increased connectivity outside of the traditional police booking
area allows law enforcement agents to use mobile devices for
a diverse range of activities in the field. Various law
enforcement agencies have started exploring the use of
mobile devices to capture fingerprints, face and iris images
from individuals at remote locations and sending the
information to a central server for further processing. Andhra
Pradesh, a state in India, has deployed a fingerprint analysis
and criminal tracing system based on a centralized storage
with decentralized acquisition architecture, with the specific
978-1-4244-7402-81101$26.00 ©2010 IEEE
Benny Senjaya
BSPA Lab
USA
Stephen Elliott
BSPA Lab
USA
intent of capturing information from the field and checking it
against a central repository [4]. Mobile devices give law
enforcement agencies the advantage of capturing biometric
information in various locations without the inconvenience of
bringing the subject back to a central processing location [5].
Using mobile devices to capture biometric data can improve
the effectiveness of agents by providing the capability of
identifying people of interest in the field and generally
improving workflow. But there are several technical and
operational challenges to implementing a mobile device
biometric system. Mobile devices embedded with biometric
technologies currently do not have a standardized sensor
capture area thereby producing different sized images.
Knowledge of the optimal size of the fingerprint image and its
impact on quality and performance of the overall system can
make the system more efficient. The research presented in
this paper evaluated the impact of fingerprint size and
fingerprint sensor interoperability on recognition performance.
The purpose of this research was to examine image quality,
minutiae count and performance in terms of matching error
rates of different sized fingerprint images collected from an
optical and capacitive fingerprint sensor. The objective of this
assessment was to provide performance related information
on feasibility of using mobile fingerprint devices in the field and
thereby improving best practices for such devices.
II. LITERATURE REVIEW
One of the earliest works in evaluating impact of image
size on fingerprint recognition performance was performed
by [6]. Watson and Wilson examined the effect of cropping
fingerprint images on the ability to match the fingerprints.
The conclusion of their evaluation was to not use fingerprint
images of less than 320X320 pixels for matching. They also
concluded that compression degrades performance at a
slower rate than the size of the fingerprint image.
In [7], Schneider et aI., described results of a study that
analyzed correlation of fingerprint image size on matching
performance. They used a dataset of images collected from
259 people with an ultrasonic fingerprint reader. They
cropped their fingerprint images using a center point
calculated by two different methods. The first method used a
fixed pOint on the platen of the sensor as the cropping center
point. The second method calculated the center of mass
using centroiding techniques. Their results showed a
decreased performance with a decrease in image size. This
study demonstrated correlation between image size and
performance, but did not attempt to analyze impact of quality

on performance of images of different sizes.
Ortega-Garcia et.al in [8] described a speech and
fingerprint multimodal matching system where a mobile
device is used to collect the biometric data. The data was
evaluated using three different multimodal fusion techniques
and an adaptive fusion strategy, and the results showed an
improvement in all strategies thus demonstrating the
feasibility of using biometric data collected from mobile
devices.
Interoperability of fingerprint sensors is an issue that has an
impact on performance of fingerprint recognition [9]. The
variations and distortions introduced by different acquisition
technologies are not consistent which generates matching
errors. Most commercially available optical sensors are based
on the concept of Frustrated Total Internal Refraction (FTIR)
which is affected by skin distortion, outdoor lighting and
residue on the sensor platen [10]. Most capacitive sensors
measure the difference in capacitance on a charged surface
when a finger comes in contact with it. The difference in
capacitance corresponds to the ridges and valleys of the
fingerprint [11]. The impact of sensor interoperability can be
reduced by improving the quality of fingerprint images and
fusing features of fingerprints collected from different sensors
[12]. This study expanded on the work performed by [6] and
[7] with respect to image sizes by analyzing images collected
from different acquisition technologies and including minutiae
count and image quality information in the analysis framework.
III. DATA SET & METHODOLOGY
Fingerprints from 190 subjects were used for analysis in
this study. Six fingerprint images were collected from the index
finger of the subject's dominant hand using two different
fingerprint sensors. The fingerprints were collected in a single
session and the subjects were seated during the fingerprint
acquisition session. A large area capacitive touch sensor and
optical sensor were used in this study. Both sensors were
FIPS 201 certified for single capture devices. Table 1 gives a
summary of the participant demographics.
Total Participants
Gender
Occupation
Number of samples
TABLE I
DATASET SUMMARY
190
Male
131
Manual Laborer
17
Female
59
Office Worker
173
190'6 - 1140
In order to test the effect of image size on image quality and
performance, seven different image sizes were selected. The
original image collected during the data collection session was
cropped to form the images for these seven different sizes.
The image sizes were chosen from page 17 of the Mobile ID
Device Best Practice Recommendation [3]. A Matlab ™
application developed in-house was used to crop the images
using the core values as the center of the cropping region. If a
fingerprint image had two cores then the core with the higher
y-axis coordinate was chosen. Table 2 lists the image sizes
and image examples from the 2 sensors. Please note the
samples images shown here are not the original size images.
These samples are an illustration of the amount of detail in
each image. It was observed that images at level 7 were large
enough to capture the entire fingerprint image.
TABLE II
SAMPLE IMAGES
Image Size Level
0.098" X 0.126"
0.118" X 0.154" 2
0.154" X 0.194" 3
0.31 4
5
7
The purpose of this research was to examine image quality,
minutiae count and performance in terms of matching error
rates of different sized fingerprint images collected from an
optical and capacitive fingerprint sensor. One of the objectives
of this study was to simulate matching of fingerprints collected
by different sensors and of different sizes, the fingerprint were
cropped to simulate the size of images that would be collected
from a mobile device.
IV. DATA SET & METHODOLOGY
A. Minutiae Count
This step of the analysis was to examine the average
minutiae count for each of the seven image sizes noted above
in Table 2. Minutiae counts were generated using
Neurotechnology's VeriFinger 6.0 extractor. The results
indicated that fingerprint images from image size level 4
(O.31"X 0.29") and above had an average minutiae count
higher than 10.

Level
1
2
3
4
5
6
7
TABLE III
AVERAGE MINUTIAE COUNT
Optical Capacitive
Dataset Dataset
1.98 2.04
2.91 2.92
4.28 4.32
11.04 10.83
15.11 15.06
16.54 16.37
32.31 32.12
The number of minutiae is fundamental to the process of
matching fingerprints. The 12-point guideline used in forensic
science states that assuming an expert can correctly extract
all minutiae points from a latent fingerprint, a 12-point match
with a full fingerprint can be considered as a sufficient
evidence of fingerprint matching [13]. Figures 1 and 2 show
the minutiae histograms across the seven different image
sizes.
0.6 s---------------------,
0.5
0.4
..
'1i
8 0.3
0.2
0.1
OJ
0.6
0.5
!: 0.4
·iii
c: 2: 0.3
0.2
0. 1
1
1,---Level l
II
II
II
1
II
II
J
I Level 2
lev el 7
40 50
Data
Fig. 1. Optical Dataset Minutiae Histogram
o 10 20 30
Data
Lev el l
40 50
Fig. 2. Capacitive Dataset Minutiae Histogram
60
60
B. NFIQ Image Quality
Image quality scores using the NFIQ image quality
algorithm were calculated [14]. NFIQ generates quality scores
in the range of 1-5 where 1 indicates best possible score and
5 indicates the worse possible score. NFIQ scores are
predictive of performance that should be expected for the
fingerprint image. Due to the extremely low minutiae count for
levels 1, 2 and 3 (Table 3) image quality scores were
generated only for fingerprints at levels 4, 5, 6, and 7. The raw
images were converted to WSQ format at 15:1 ratio using a
certified package [15]. Table 4 gives a summary of descriptive
statistics at each level.
TABLE IV
NFIQ SCORE DESCRIPTIVE STATISTICS
Level Optical Dataset Capacitive
Dataset
4 Median : 3.0 Median : 3.0
Mean: 3.0 Mean : 3.0
5 Median :3.0 Median :3.0
Mean:2.8 Mean:2.7
6 Median: 3.0 Median: 3.0
Mean 2.8 Mean: 2.7
7 Median : 2.0 Median : 2.0
Mean:1.9 Mean:1.7
Quality assessment is important to ensure high quality
fingerprint images are stored for matching purposes. Previous
research has shown that the effect of low quality images is
predictable, but the impact of high quality images is harder to
determine. Images at levels 4, 5, and 6 had the same median
score and images at level 7 had the best NFIQ median score.
Both the optical and capacitive data sets exhibited similar
behavior for image quality scores and average minutiae count.
C. Performance of Native Datasets
To compare performance of biometric systems a modified
Receiver Operating Characteristic (ROC) curve called
Detection Error Tradeoff (DET) curve can be used [16]. A DET
curve plots the false match rate (FMR) on the x-axis and false
non match rate (FNMR) on the y-axis as function of decision
threshold.
01
O(r. 01" l·
l."md 11 U U L< '" L.' FAIl
Fig. 3. Optical Dataset DET Curves
lOIn:

.g
_ ...... ......... fh $ . . .... �'
Ol¶
01111:
L¾ll U
-
´
FAR
... =- ""'
1µ lln
Fig. 4. Capacitive Dataset DET Curves
Figure 3 and 4 show superimposed DET curves for the
image size levels 1 to 7 for the optical and capacitive data sets
respectively. Tables 5 and 6 list the FNMR and FMR
respectively at an operational threshold of 0.1 % FMR using
the VeriFinger matcher.
TABLE V
VERI FINGER FNMR (IN %)
ImaQe Size Level Optical Dataset Capacitive Dataset
1 98.52 98.37
2 81.76 81.00
3 46.24 41.25
4 1.27 1.70
5 0.00 0.58
6 0.12 0.29
7 0.00 0.00
TABLE VII
VERI FINGER FMR (IN %)
ImaQe Size Level Optical Dataset Capacitive Dataset
4 0.28 0.26
5 0.30 0.23
6 0.16 0.14
7 0.00 0.01
Both DET curves showed an improvement in performance
as the size of the images was increased. The best
performance for the fingerprint images was found to be at
level 7 for both the optical and capacitive datasets. The results
from Tables 3 and 4 showed that images at level 7 had the
best quality and the highest average of minutiae count which
resulted in lower number of false non match and false match
errors. This was expected since level 7 covered the entire
fingerprint region. The median NFIQ scores for images at
levels 4, 5 and 6 (Table 4) showed a similar median score but
the DET curves showed a marked improvement in
performance for both the optical and capacitive data sets.
Although the median image quality score at level 4 and 5 were
the same, the average number of minutiae count increased
which lead to a decrease in the FNMR.
D. Zoo Plot Analysis
The relation between genuine score distribution and
imposter score distribution can be analyzed using the
biometric zoo plot as described by Dunstone and Yeager [17).
They described four different categories based on the
relationship between genuine and imposter scores:
Chameleons, Phantoms, Doves and Worms. Chameleons
generate high genuine and imposter scores. Phantoms
generate low genuine and imposter scores. Doves generate
high genuine scores and low imposter scores. Worms
generate 1?"¥J!enuine score~ and high imposter scores.
Performlx , a commercially available software package
provides the ability to visually analyze large biometric
datasets, was used to generate so called "zoo plots". Fig. 5
and 6 are the zoo plots for level 7 datasets for optical and
capacitive sensors.
Zoo p lot 14r---.----r---,--9r_:;--_r--_,----r_<
12
: . : QI
( . . • •
I: c
` l5u 10 • :' _ Fm"'"!=n. o8 •... :.=W ''-7. d ..' ..'
;.. '!I .t· ... .. . ......... , ..... , ......•... •..•.• .• •...• •. ³ •...•.... ; .. ....... , .. .h6 •
: .:) , ...• , ...
• :
a .
.. : I : • .. •.• :.. : .• •... e . • : .... •..... •
: ' ': . ¹ , º.......... i : .......... i
: QI
4 ^
:. : : .: • t· • • • • • • •• • • • • • • • • • . • • • • • • • • • • • • • • • • • . • • • • • • • • • • . . ·
.
´ • • • • • • • • • • . .
·
°0---;1;2-:ono --M4O:n 0--''*60;;-0 -i;80;:n0---:;1t,00;;!; 0--;-:;--,1 2!n.·0;;-0 -:;-1+,4·:-;;:00--;-:--,1 6=:·0::-0 -:-:J-1800
Average genuine match score
Fig. 3. Optical Dataset Level 7 Zoo Plot
Zoo Zp lot
14r-----.------.---,-------r¿---.------.
12 .............. , ---------------:- - .- .. . . . •. . . . µ p.-q---
] '
. . . » . .
¶ 1,,- '
· 10 . ..... ' [ .. - - .- . ... •. . .* #.:+-= j.k .. b . . . . . . . . . .' -. : •. ;Z ..•. ..... •.¼ . • ------ • ½ ...... . .'
• •
-
' : . ' : ' ' r
:
... , . , .. :. .. ' . .. ... 8.. 8 .i,. . .. .. ..... .. . . . . • . . . . ·
·:, ··
··•• ·
··
··
.
l .w :

.':1.::. { .. ; .
-
-
·-
.;.!I
6
···:: ..'-. ...-. . .-. ..'-- -'1
Qj • • • • • • • • • • • •• • :. . . . . . . . . . : • • • • • .. ············
I
······ .... . ....
¸ . . . .
4 . . . . . . . •. . . : .. ... . .. : . . . .• % ....... -.. . -. . .. :... . . . ... . ... .
loo 400 600 800 1000
Average genuine match score
1200
Fig. 4. Capacitive Dataset Level 7 Zoo Plot
1400
Both figures 5 and 6 show a weak positive relationship
between average genuine match score and average imposter
score for the datasets. The graphs show a higher probability
of finding worms (the bottom left quadrant) indicating that
subjects were generating low genuine and imposter scores.
Low genuine scores will result in a lower confidence in the
matching operation, as well as a higher probability of false non
matches, which in a law enforcement application is not
desirable. Analysis of the zoo plots did not reveal any specific
trends for any of the zoo categories. Similar zoo plots were
generated for level 5 and level 6 for optical and capacitive
data sets and while the overall relationship remained stable,
the individual data points were more widely dispersed. Those
subjects that generated the highest number of false matches
were identified from the zoo plot. For optical dataset and
capacitive dataset level 7, two subjects were found to have

produced 36 false matches but they were not generated
against the same subjects. The same analysis was performed
on level 5 and 6 for both datasets to examine trends in
subjects generating false matches and false non-matches.
The results are given in Table 7.
Table 1. Veri Finger FMR (in %)
Level Subject Number of
ID(s) False Matches
Optical L7 56,70 36
Capacitive L7 58,178 36
Optical L6 79,178 160
Optical L5 4,58 252
For images at level 7, 6 and 5 there were no common
subjects between optical and capacitive datasets. Identifying
subjects that generate errors in the optical dataset would not
be useful in identifying subjects that generate errors in the
capacitive dataset and vice-versa. It was also observed that
there were no common subjects for the optical dataset across
different levels. The same observation holds true for the
capacitive dataset across different levels.
E. Performance of Interoperable Datasets
In a mobile 10 infrastructure, the ability to match fingerprints
from different sensors is important. Previous research has
shown that matching fingerprints from different sensors has an
effect on error rates [18]. Interoperability analysis was
performed to evaluate the FNMR of matching fingerprints from
the optical and capacitive datasets. The analysis concentrated
on FNMR as a first step towards a comprehensive analysis of
performance analysis. VeriFinger 6.0 was used to generate
FNMR. Every subject provided 6 images on each sensor. The
interoperable dataset was generated by combining 3 images
from the optical dataset with 3 images from the capacitive
dataset. FNMR was calculated for the interoperable datasets
at every image size level by comparing the optical fingerprint
images to capacitive fingerprint images. Table 8 shows the
results of the interoperable FNMR at fixed 0.1 % FMR.
Table 2. FNMR (in %) at FMR of 0.1%
Image Size Level Interoperable
Dataset
1 99.79
2 91.42
3 61.34
4 4.56
5 1.03
6 0.27
7 0.27
Interoperability FNMR reduced significantly between level 3
and level 4, which can be attributed to higher number of
minutiae points for level 4 images. This improvement indicates
the need for larger size images when comparing images from
different sensors. The interoperability FNMR for images for
level 5 was different from level 6 and this seems to have been
caused in part by the higher number of minutiae points and
not due to image quality scores. This indicates a need for a an
assessment framework of input fingerprint images which takes
into account factors other than just image quality scores.
F. Matching Performance for Different Sized Images
The size of fingerprint images collected from the field will not
remain constant. The size of the fingerprint image is a function
of the type of mobile device used for capturing the fingerprint
images. The analysis in this section focused on understanding
the impact of comparing fingerprint images of different sizes.
Images at level 7, (0.578XO.618) - the largest size were
considered to be the enrollment or reference template.
Fingerprints from levels 3, 4, 5 and 6 were compared to the
reference template and the FNMR was calculated, which are
shown in Table 9.
Table 3. FNMR (in %) at FMR of 0.1 %
Enrollment Level Comparison Optical Capacitiv
Level Dataset e Dataset
FNMR% FNMR%
3 70.28 68.53
7 4 2.09 2.16
5 0.12 0.26
6 0.24 0.26
The lowest FNMR was generated when images from level 5
and 6 were compared to the enrollment images. Both optical
and capacitive fingerprint images showed a similar trend in
decrease of FNMR among the different levels of images. The
ability to match fingerprints of different sizes is crucial, and
even more so in a mobile device environment. When results
from Table 5 are compared with results from Table 9 it can be
seen that matching two fingerprints from image size level 4
yields better results than matching fingerprints of level 7 to
level 4. It was also observed that matching fingerprints from
different image size levels had better results than matching
fingerprints from different sensors. Operational decisions
about mobile fingerprint devices will depend on the
performance of fingerprints captured, and these results will aid
in forming policies for comparing fingerprints of different sizes
and fingerprints collected from different sensors. For example,
comparison of fingerprints from level 7 and 3 should not use a
single fingerprint to make a decision, whereas confidence in
comparison of fingerprints from level 7 and 4 would be
increased by using 2 or more fingers.
V. CONCLUSIONS
The results from this research show that single fingerprint
images of sizes at or below level 3 are unsuitable for matching
purposes. For mobile fingerprint devices the number of
minutiae extracted from the image is as crucial as capturing
high quality fingerprint images. Although fingerprints at level 7
showed the best results, fingerprint images at levels 5 and 6
also showed performance that would be acceptable in law
enforcement applications. Interoperability FNMR reduced as
the size of the fingerprint image was increased, which
indicates that the number of minutiae was important to

reducing FNMR. An interesting result was the similarity of the
interoperability FNMR at level 6 and level 7, even though the
number of minutiae count and NFIQ score was significantly
different at the two levels. The comparison of fingerprint
images across different levels also showed acceptable FNMR
at levels 5 and 6.
The results from this research have shown the
different operational issues that could arise due to use of
fingerprint recognition in a mobile 10 environment. There are
several other avenues of research which require attention to
get a better understanding of mobile fingerprint recognition.
This study used images collected using peripheral fingerprint
scanners. Collecting fingerprint images using mobile devices
would also highlight usability issues, data collection errors and
differences between providing fingerprint sequentially or
simultaneously to the fingerprint devices. This study used
single fingerprint comparisons for calculating FNMR. Future
studies should examine the improvement in performance by
using 2, 3, and 4 finger matching. Another important aspect in
mobile 10 is the data transfer rates between the mobile device
and the central processing station. The size of the data
packets would determine the processing time, and the impact
of different WSQ compression levels at each size level also
needs to be analyzed. The INC ITS 378 template standard was
not used to store the fingerprint details in this study. The
impact of standardized templates also requires evaluation in
order to ensure interoperability between different systems. As
the deployments of mobile fingerprint devices increases
further research into these operational issues will become
crucial to its success.
[1]
VI. REFERENCES
Non-AFIS Fingerprint in Notebook PC's and
Wireless Market Devices, 2008.
[2] ISO/IEC JTC1 SC37 SD2 - Harmonized Biometric
Vocabulary, 2006.
[3]
[4]
[5]
[6]
[7]
[8]
R. Bolle, S. Cole, and N. Ratha, History of
Fingerprint Pattern Recognition, New York:
Springer, 2004, pp. 1-25.
CMC, Fingerprint Analysis and Criminal Tracing
System.
Mobile ID Device Best Practice Recommendation,
NIST, 2008.
C. Watson and C. Wilson, Effect of Image Size
and Compression on One-to-One Fingerprint
Matching, 2005.
J. Schneider, C.E. Richardson, F.W. Kiefer, and V.
Govindaraju, On the Correlation of Image Size to
System Accuracy in Automatic Fingerprint
Identification Systems. in Audio-and Video-Based
Biometrie Person Authentication, Audio-and VideoBased
Biometrie Person Authentication, Guildford,
U.K.: LNCS, 2003.
J. Ortega-Garcia, J. Fierrez-Aguilar, J. Bigun, and
J. Gonzalez-Rodriguez, Multi modal Biometric
Authentication using Quality Signals in Mobile
Communications, Mantova, Italy: 2003.
[9]
[10]
[11]
[12]
[13]
[14]
[15]
[16]
[17]
[18]
A. Jain, D. Maltoni, and A. Ross, Biometric Sensor
Interoperability, Berlin: Springer-Verlag, 2004.
L. O'Gorman and X. Xia, Innovations in fingerprint
capture devices, Pattem Recognition, vol. 36,
2001, pp. 361-369.
R. Bolle, N. Ratha, and D. Setlak, Advances in
Fingerprint Sensors Using RF Imaging Techniques,
New York: Springer-Verlag, 2004, pp. 27-53.
S. Elliott, S. Modi, and H. Kim, Performance
Analysis for Multi Sensor Fingerprint Recognition
System, New Delhi, India: Springer Verlag, 2007.
S. Prabhakar, D. Maltoni, D. Maio, and A. Jain,
Handbook of Fingerprint Recognition 2nd Edition,
Springer, 2003.
E. Tabassi and C. Wilson, A novel approach to
fingerprint image quality, IEEE Intemational
Conference on Image Processing, 2005. ICIP 2005,
Genoa, Italy: 2005, pp. 37-40.
Aware Inc WSQ1 000, 2008.
G. Doddington, T. Kamm, A. Martin, M. Ordowski,
and M. Przybocki, The DET curve in assessment of
detection task performance, Greece: 1997, pp.
1895-1898.
N. Yager and T. Dunstone, Design, Evaluation, And
Data Mining, New York, New York: Springer-Verlag,
2008.
S. Modi, Analysis of Fingerprint Sensor
Interoperability on System Performance, 2008, p.
176.
VIII. VITA
Shimon Modi graduated from Purdue University in 2008
with Ph. D. in Technology focusing on evaluation of biometric
system interoperability. He has been the Director of Research
of BSPA Lab, Purdue University and he is currently a visiting
scientist with Center for Development of Advanced Computing
(C-DAC) Mumbai, India. He has written several papers on
statistical evaluation of biometric systems with respect to
interoperability, sample quality and demographics.
Ashwin Mohan is a graduate student currently pursuing the
M. S. degree in Information Security from the Centre for
Education and Research in Information Assurance (CERIAS)
at Purdue University. He received his Bachelors degree from
the Dhirubhai Ambani Institute of Information and
Communication Technology (DA-IICT), India. His research
interests include fingerprint recognition, development of
biometric standards and evaluation of small scale device
forensics.
Benny Senjaya is a graduate student currently pursuing the
M.S. degree in Technology focusing on human interaction
within biometric technology at Purdue University. He received

(2010) Fingerprint recognition performance evaluation for mobile ID applications

More Related Content

What's hot (20)

Similar to (2010) Fingerprint recognition performance evaluation for mobile ID applications (20)

More from International Center for Biometric Research (20)

Recently uploaded (20)

(2010) Fingerprint recognition performance evaluation for mobile ID applications