An Efficient Reconfigurable Architecture for Fingerprint Recognition

TLDR: An efficient Finite State Machine FSM based reconfigurable architecture for fingerprint recognition using fusion scores with correlation matching technique for FVC2004 DB3 Database is proposed and performance parameters such as TSR Total Success Rate, FAR False Acceptance Rate, and FRR False Rejection Rate are computed.

Abstract: The fingerprint identification is an efficient biometric technique to authenticate human beings in real-time Big Data Analytics. In this paper, we propose an efficient Finite State Machine FSM based reconfigurable architecture for fingerprint recognition. The fingerprint image is resized, and Compound Linear Binary Pattern CLBP is applied on fingerprint, followed by histogram to obtain histogram CLBP features. Discrete Wavelet Transform DWT Level 2 features are obtained by the same methodology. The novel matching score of CLBP is computed using histogram CLBP features of test image and fingerprint images in the database. Similarly, the DWT matching score is computed using DWT features of test image and fingerprint images in the database. Further, the matching scores of CLBP and DWT are fused with arithmetic equation using improvement factor. The performance parameters such as TSR Total Success Rate, FAR False Acceptance Rate, and FRR False Rejection Rate are computed using fusion scores with correlation matching technique for FVC2004 DB3 Database. The proposed fusion based VLSI architecture is synthesized on Virtex xc5vlx30T-3 FPGA board using Finite State Machine resulting in optimized parameters.

Full text

Research A rticle
An Efficient Reconfigurable Architecture for
Fingerprint Recognition
Satish S. Bhairannawar,
1
K. B. Raja,
2
andK.R.Venugopal
3
1
Department of Electronics and Communication, SDMCET, Dharwad 580002, India
2
Department of Electronics and Comm un ica tion, UVCE, Bangalore 560001, India
3
UVCE, Bangalore University, Bangalore, India
Correspondence should be addressed to Satish S. Bhairannawar; satishbhairannawar@gmail.com
Received  December ; Revised  April ; Accepted  May 
Academic Editor: Lazhar Khriji
Copyright ©  Satish S. Bhairannawar et al. is is an open access article distributed under the Creative Commons Attribution
License, which permits unrestricted use, distribution, and reproduction in any medium, provided the original work is properly
cited.
e ngerprint identication is an ecient biometric technique to authenticate human beings in real-time Big Data Analytics. In
this paper, we propose an ecient Finite State Machine (FSM) based recongurable architecture for ngerprint recognition. e
ngerprint image is resized, and Compound Linear Binary Pattern (CLBP) is applied on ngerprint, followed by histogram to
obtain histogram CLBP features. Discrete Wavelet Transform (DWT) Level  features are obtained by the same methodology. e
novel matching score of CLBP is computed using histogram CLBP features of test image and ngerprint images in the database.
Similarly, the DWT matching score is computed using DWT features of test image and ngerprint images in the database. Further,
the matching scores of CLBP and DWT are fused with arithmetic equation using improvement factor. e performance parameters
such as TSR (Total Success Rate), FAR (False Acceptance Rate), and FRR (False Rejection Rate) are computed using fusion scores
with correlation matching technique for FVC DB Database. e proposed fusion based VLSI architecture is synthesized on
Virtex xcvlxT- FPGA board using Finite State Machine resulting in optimized parameters.
1. Introduction
e reliable personnel authentication [, ] based on bio-
metrics has signicant importance in the present digital
worldandcanbeachievedbyhumanandcomputerinterface
activities. e evolution of biometrics in recent years from
single mode to multiple mode closed systems has made
it possible to consider for Big Data processing [, ]. e
development of new algorithms and parallel processing archi-
tectureshasanimpactforBigDataprocessingresponsetime.
e interoperable feature and many sources and avenues for
collection of biometric samples have made the biometric
evolution of Big Data possible. e biometric physiologi-
cal or b ehavioural samples are captured using sensors or
devices, which are further processed in the next level of
vetting through Oce for Personal Management (OPM)
which can be either verication or identication. Fingerprint
based identication is one of the most important biometric
technologies, which have drawn a substantial amount of
attention recently since the process of acquiring ngerprint
samples are easy and simple. A ngerprint is seen as a set of
interleavedridgesandvalleysonthesurfaceofthenger.e
most ngerprint matching approach relies on the fact that the
uniqueness of a ngerprint can be determined by minutiae,
which are represented by either bifurcation or termination
ofridges.equalityandenhancedminutiae[–],which
inuence recognition rates are discussed in literature.
e features of a ngerprint can be derived using the
following: (i) Spatial domain:thefeaturesofanimageare
carried out directly on pixel value. Examples are L ocal Binary
Pattern [], Complete Linear Binary Pattern [], and Singular
Value Decomposition []. (ii) Transform domain:inthisany
transform is applied to an original image to get a t ransformed
imageonwhichfurtherprocessingisdone.Examplesare
Fast Fourier Transform [], Discrete Cosine Transform [],
Discrete Wavelet Transform [], and Dual Tree Complex
Hindawi Publishing Corporation
VLSI Design
Volume 2016, Article ID 9532762, 22 pages
http://dx.doi.org/10.1155/2016/9532762

 VLSI Design
Wavelet Transform []. (iii) Fusion: in this technique [, ]
it combines the advantages of both spatial and transform
domain.
e automated ngerprint recognition system is used
for both identication and verication against standard
database law enforcement agencies to identify the suspect
for committing cr i me or for attendance verication process
to verify t he claimed identity. e performance speed of
ngerprint system is a critical factor to be addressed while
dealing with large databases. e real-time processing of a
ngerprintrecognitionsystemisitsabilitytoprocessthelarge
data and produce the results within certain time constraints
in the order of milliseconds and sometimes microseconds
depending on the application and the user requirements.
In this category, Field Programmable Gate Array (FPGA)
outperforms other processors. e FPGAs are specially built
hardware optimized for speed and are suitable for real-time
biometric data processing. Multicores and HPC clusters have
reasonable real-time processing capabilities, but not ecient
as FPGA with many processing cores and high bandwidth
memory.
In real time, speed of the algorithm becomes crucial
which in turn denes the throughput. e ecient FPGA
architectures [–] for ngerprint processing and existing
algorithms to identify a ngerprint based on minutiae [],
ridge, multiresolution features, and Hough transform were
discussed.
Vatsaetal.[]proposedRedundantDiscreteWavelet
Transform based on local image quality assessment algorithm
followed by extraction algorithm using Level  features.
ese features are combined with Level  and Level  in
the ngerprint identication scheme. Finally, the matching
performance was improved by using quality based likelihood
ratios. Govan and Buggy [] proposed eective matching
solution that addresses security and privacy issues. is tech-
nique eliminates t he requirement to release biometric tem-
plate data into an open environment which uses embedded
applications such as smart cards. e eective disturbance
rejection methodology which is able to dierentiate between
equivalent and insignicant structure models was discussed.
Nain et al. [] proposed an algorithm to classify n-
gerprint images into four dierent classes using High Ridge
Cur vature (HRC) algorithm involving two stages. In the rst
stage, HRC region was extracted, which avoids core point
detection. In the second stage, ridges inside HRC region were
considered for matching. e global distribution structure
and the local matching similarities [ ] between ngerprints
were considered for matching using Hidden Markov Model
(HMM) []. Nikam and Agarwal [] proposed spoof n-
gerprint detection using ridge let transform. e comparisons
of individual ridgelet energy and cooccurrence signatures
were analysed and also testing was done using diverse
classiers. Masmoudi et al. [] proposed an algorithm which
used the rotation invariant measured as local phase and was
combined with Linear Binary Pattern Features to improve
the performance accuracy. Stewart et al. [] proposed the
test technique to determine the eects of outdoor and cold
weathereectsonchipversusopticalngerprintscanner,
ngerprint recognition quality, and device interaction. e
results suggested that performance has no dependence on
temperature and humidity. Cao and Dai [] proposed nger-
print segmentation for online process using frame dierence
technique. Further the segmented foreground was used for
identication.
Umamaheswari et al. [] proposed ngerprint classica-
tion and recognition using neuro-nearest neighbour based
method which improves classication rate. is consists of
dierent stages such as image enhancement, line detector
base feature extraction, and neural network classication
using back propagation networks. e results have shown
the accurate estimation of orientation and ridge frequency
which helps in better recognition. C onti et al. [] proposed
pseudo-singularity points based ngerprint recognition. is
technique uses additional parameters such as their relative
distance and orientation around standard singularity points
(core and delta) which enhances the matching performance.
Ahmed et al. [] proposed Compound Local Binary Pattern
(CLBP) for rotation invariant texture classication. is com-
bines magnitude information of the dierence b etween two
grey values with original LBP pattern and provides robust-
ness. Paulino et al. [] proposed an alignment algorithm
(descriptor-based Hough t r ansform) for latent ngerprint
matching. is technique measures similarity between n-
gerprints by considering both minutiae and orientation eld
information. e comparison was done between proposed
and generalized Hough transform for large database.
Feng et al. [] proposed a technique using orientation
eld estimation base d on prior knowledge of ngerprint
structure. e dictionary of reference for orientation patches
was constructed using a true set of orientation elds. e
approach was applied to the overlapped latent ngerprint
database to achieve better performance compared to conven-
tional algorithms.
Contribution. e contribution and novel aspects of the
proposed techniques are listed as follows: (i) the computation
of the novel matching score for CLBP and DWT features; (ii)
thematchingscorevalueswhicharevariedbasedonchar-
acteristics of images, that is, the values which are computed
adaptively based on characteristics of the images; (iii) the
fusion of matching scores with improvement factor; (iv) the
implementation of FSM based VLSI architecture to improve
the hardware performance.
2. Proposed Fingerprint Recognition System
An ecient ngerprint recognition model using histogram of
CLBPscores,DWTfeaturescores,andfusionofbothscores
is given in Figure .
2.1. Fingerprint Database. e DB of FVC ngerprint
database [] is considered for performance analysis. e
size of each ngerprint image is  × with  dpi. e
ngerprint samples of ten dierent persons are shown in
Figure .

VLSI Design 
CLBP
CLBP
Histogram
Histogram
DWT
DWT
DWT
features
DWT
features
DWT
matching
CLBP
matching
Fusion
Match score
DWT
Match score
CLBP
Match/
mismatch
Training samples
Test samples
Recognition stage
Training stage
Image
resizer
(256 × 256 )
Image
resizer
(256 × 256 )
F : Block diagram of the proposed ngerprint recognition system.
2.2. Preprocessing. e original ngerprint image of size 
× is resized to  ×, which is suitable for hardware
implementation.
2.3. Complete Local Binary Pattern (CLBP). It is an extension
of the Local Binary Pattern (LBP) [] texture operator.
e CLBP operator gives both sign CLBP
and magnitude
components CLBP
 for each pixel from its neighbouring
pixels. If isthenumberofneighboursofacentrepixel,then
CLBP operator uses 2bits to code centre pixel. e rst 
MSB bits represent sign and the next  LSB bits represent
magnitude.
e binary bit patterns are generated for sign and mag-
nitude components for each pixel. e ngerprint image is
scanned from le to right and top to bottom and considering
each pixel which is surrounded by  neighbouring pixels, that
is,  ×  matrix. e centre pixel intensity value is 
𝑐
and
surrounded neighbouring pixel intensity values, say, 
𝑝
.e
sign bit patterns for  × matrices are generated using

(

)
=



0, 
𝑝
−
𝑐
≤0
1, 
𝑝
−
𝑐
>0.
()
e magnitude bit pattern is generated using

(

)
=



0, 
𝑝
−
𝑐
≤
avg
1, 
𝑝
−
𝑐
>
avg
,
()
where 
avg
= |1|+|2|⋅⋅⋅|8|/8 and 1to 8are the
magnitude values of the dierence between respective 
𝑝
and

𝑐
.
Each neighbourhood pixel is represented by two bits;
that is, MSB bit represents sign and the LSB bit represents
magnitude. Each centre pixel is represented by eight sign
bits and eight magnitude bits. e example for CLBP is as
shown in Figure . e arbitrary values for  × matrix
are considered in Figure (a). e values of neig hbouring
pixels are subtracted from centre pixel value and are given
in Figure (b). e sign of each co ecient in Figure (b) is
representedinFigure(c)assigncomponentofCLBP.e
magnitudecomponentsofCLBPareshowninFigure(d)
by considering only magnitude of Figure (b). e average
value of the CLBP magnitude component is computed and
is compared with neighbouring CLBP magnitude coecient
values and assigns binary values using () to generate CLBP
magnitude p attern given in Figure (f). e numbers of
centrepixelsavailableforimagesize×  are 
using  × window matrix. e binary eight bits of sign and
magnitude of each pixel are converted into decimal values for
feature extraction. If the CLBP sign and magnitude coecient
features are considered directly for an image size of  ×
, the algorithm requires  for sign and  for
magnitude; that is, total number of features are .
2.3.1. Histogram of CLBP Features. e features obtained
directly from CLBP are large in number and hence increase in

 VLSI Design
F : Sample images of ten dierent persons of FVC database.
matching processing time and are a disadvantage in hardware
implementation. e histogram on CLBP produces only 
features for each sign and magnitude. Hence the number of
features is reduced from  to , that is, approximately
.% features compared to CLBP. e advantage of histogram
on CLBP is that the number of features reduces and also
features are more unique. e histograms of original nger-
print, sign, and magnitude components of CLBP are shown
in Figure .
2.3.2. Proposed CLBP Matching Score. e CLBP histograms
oftestanddatabaseimagesarecomparedcomponentwiseto
compute CLBP match score .eabsolutesigncomponent
dierence CLBP
 
𝑖
betweensigncomponentCLBP  
𝑖
of
test ngerprint and sign component CLBP
 
𝑗𝑖
of nger-
print images in the database is computed using
CLBP
 
𝑖
=





CLBP
 
𝑖
−CLBP  
𝑗𝑖





, ()
where is intensity values ( to ); =numberofpersons
in the database ×number of images per person.
e CLBP sign histogram coecients match 
𝑖
is com-
puted based on threshold sign dierence value ( for best
match) given in

𝑖
=



1, CLBP
 
𝑖
<14
0, otherwise.
()
e absolute magnitude component dierence
CLBP
 
𝑖
between magnitude component CLBP  
𝑖
of
test ngerprint and magnitude component CLBP
 
𝑗𝑖
of
ngerprint images in the database is computed using
CLBP
 
𝑖
=





CLBP
 
𝑖
−CLBP  
𝑗𝑖





, ()
where is intensity values ( to ); =numberofpersons
in the database ×number of images per person.
e CLBP magnitude histogram coecient match 
𝑖
is
computed based on threshold magnitude dierence value (
for best match) given in

𝑖
=



1, CLBP

𝐷
𝑖
<18
0, otherwise.
()

VLSI Design 
912 34
10 25 28
99 64 56
−1 −11
1
−1
111
16 13 9
15 3
313974
74 39 31
3
−15
−16 −13
00 00 11
00 11
11 11 11
9
91234
10 25 28
99 64 56
(e) (f)
(d)(c)
(a) (b)
CLBP
operator
F : CLBP operator: (a)  × sample block; (b) local dierence; (c) sign components; (d) magnitude components; (e) original matrix;
(f) CLBP matrix.
e overall CLBP match count by considering both sign and
magnitude histogram coecients is
CLBP
Match count
=



Match count +1, if 
𝑖
⋅
𝑖
=1
Match count, otherwise.
()
e CLBP match score is computed using CLBP match count
and number of histogram levels using
%CLBP
Match score 
=
CLBP
Match count ∗100
Number of Histogram Levels
.
()
e rst and eighth samples of same person are considered
as database and test image. e original ngerprint, CLBP
magnitude component, and CLBP sign component images of
database and test image are shown in Figures (a)–(c) and
(d)–(f) respectively. e CLBP
Match score is computed
betweendatabaseimageandtestimageofthesameperson,
which yields high value, that is, .%.
e rst and eighth samples of dierent person are con-
sidered as database and test image. e original ngerprint,
CLBP magnitude component and CLBP sign component
imagesofthedatabaseandtestimageareshowninFigures
(a)–(c) and (d)–(f) respectively. e CLBP
Match score
is computed between t he database image and test image of the
dierent person, which yields low value, that is, .%.
2.4. DWT Algorithm. e DWT [] provides spatial and
frequency characteristics of an image. It has an advantage
over Fourier transform in terms of temporal resolution where
it captures both frequency and location information. e
signal is translated into shied and scaled versions of the
mother wavelet to generate DWT bands. e ngerprint
image is decomposed into multiresolution representation
using DWT. e LL subband gives overall information of
the original ngerprint image, the LH subband represents
horizontal information of the ngerprint image, HL gives
vertical characteristics of the ngerprint image, and HH gives
diagonal details.
e Haar wavelets are orthogonal and have simplest
useful energy compression process. e Haar transformation
on one-dimension inputs leads to a -element vector using


(
1
)
,
(
2
)

=
(

(
1
)
,
(
2
))
,
()
where  = (1/

2)
11
1−1
 is the Haar operato r and (1)
and (2)are the sum and dierence of (1)and (2)which
producelowpassandhighpassltering,respectively,scaled
by 1/

2 to preserve energy. e Haar operator  is an
orthonormal matrix since its rows are orthogonal to each
other(theirdotproductsarezero)andhaveunitlengths;
therefore 
−1
=
𝑇
. Hence we may recover from using
(

(
1
)
,
(
2
))
=
𝑇

(
1
)
,
(
2
)
.
()
For D image, Let  be  × matrixofanimage;the
transformation is obtained by multiplying columns of by