Available online at www.sciencedirect.com
ScienceDirect
AASRI Procedia 9 (2014) 2 – 7
20144 AASRI Conference on
o Circuit annd Signal Processing
P
(CSP 2014)
A New
w Texturre Analyysis Appproach fo
or Iris Recognitiion
Izem Ham
mouchenea, Saliha Ao
ouata,*
a
LRIA Laboratory
ry/Computer Sciennce Department, USTHB Universiity, Algiers,161111, Algeria
ihamoucchene@usthb.dz, saouat@usthb.d
dz
Absttract
One of the most im
mportant authenntication approaaches is the iriss recognition sy
ystem (IRS), which
w
is based on the iris of
t
paper, we propose a new
w iris recognitio
on system using a novel featuure extraction
apersson for the authhentication.In this
methhod. The propossed method, Neeighborhood-baased Binary Patttern, comparess each neighborr of the central pixel withthe
next neighbor to enncode it by 1 if
i it is greater or 0 if it is loower than the central
c
pixel. The
T obtained bbinary code is
convverted into a decimal number to
t construct thee NBP image. In
I order to deall with the rotattion problem, w
we propose an
encoding process too obtain a rotatiion-invariant im
mage. This imagge is subdivideed into several blocks
b
and the mean of each
blockk is calculated. After, the vaariations of thee means are enncoded by a biinary code. Thee obtained binnary matrix is
consiidered as featurre descriptor off the iris. In thee evaluation parrt, the CASIA iris
i database [10] has been useed to evaluate
the performance
p
of the proposed IR
RS. The experim
ments demonstrrate that the pro
oposed method gives better reccognition rate
comppared to the LB
BP method. Expperimental resuults show also that
t
the propossed system espeecially the featuure extraction
methhod gives promiising results.
©
The Authors.
Published
by Elsevier
B. V. This is an open access article under the CC BY-NC-ND license
d by
Elsevier
B
B.V.
© 2014
20014.Published
(http://creativecommons.org/licenses/by-nc-nd/3.0/).
Seleection and/or peer
p review unnder responsibbility of Amerrican Applied Science Reseearch Institute
Peer-review under responsibility of Scientific Committee of American Applied Science Research Institute
Keyw
words : Iris Recoggnition System; Neighborhood-bas
N
sed Binary Patternn; Local Binary Pattern;
P
Texture analysis;
a
Mean vaariation.
1. In
ntroduction
C
Computer
visioon is one of thhe most imporrtant areas of research, which provides efficient solutions to many
probblems. Patten recognition
r
is mainlyused to
t recognize auutomatically different
d
entitiies from an im
mage. The
* Corresponding author.
a
Tel.: +213321247187; fax: +21321247187.
+
E
E-mail
address:saoouat@usthb.dz.
2212-6716 © 2014 The Authors. Published by Elsevier B. V. This is an open access article under the CC BY-NC-ND license
(http://creativecommons.org/licenses/by-nc-nd/3.0/).
Peer-review under responsibility of Scientific Committee of American Applied Science Research Institute
doi:10.1016/j.aasri.2014.09.002
Izem Hamouchene and Saliha Aouat / AASRI Procedia 9 (2014) 2 – 7
secuurity field hass shown a reaal interest in computer
c
visiion particularlly for the ideentification. Inndeed, every
hum
man has particuular propertiess from others such as shapee, size etc…
M
Modern
securitty sciences usse these differeences to contrrol the access to restricted places,
p
which is one of the
funddamental probblems in secuurity field. The increasingg need of thee security fieeld has given rise to the
deveelopment of recent
r
and effficient authenttication system
ms.Old appro
oaches of idenntification succh as key or
passswordare not satisfactory
s
inn many appliccation areas. These
T
conventtional methodds can be forgootten, stolen
or crracked. For thhese weaknessses, the recentt science is intterested in auttomatic system
ms of identificcation which
are based on biometrics techhnology [1].B
Biometric ideentification iss subdivided into two m
main classes:
physsiological chaaracteristics suuch as fingerpprints, iris (Figg 1),and behaavior characterristics such ass voice. The
needd of reliable annd secure systtems has invollved the emerrgence of the biometric
b
systtems. Fingerprrint,face and
speaaker recognitioon have been widely studieed. Among alll the biometricc recognition systems, Iris Recognition
Systtem (IRS) is the most effi
ficient and relliable system [2] for auth
henticity checkk. Iris identiffication was
propposed by Flom
m and Aran [112]. Recent suurveys of iris recognition
r
allgorithms cann be found in [13]. This is
due to the stabilityy of the humaan iris, its invaariant over tim
me and its uniq
queness for eaach person. Evven between
brothhers or twins [4].
T IRS is a high accuraccy verificationn technology [5],and a vaalid biometriccs approach ffor personal
The
idenntification. Thhe IRS has beeen wildly stuudied [4] and used especiaally in the seccurity fields. Thus, many
counntries use IRS
S in order to improve
i
theirr security suchh as in airportts and governnment buildings.Although
the iris
i identificattion theory was
w started earrlier, most im
mportant recen
nt works [6] have
h
been insppired by the
workks of John Daaugman [7].
A classical iris recognition system
s
includees a series of steps:
s
image acquisition,
a
iriis preprocessinng (includes
localization and normalization)
n
), feature extraaction and mattching steps [4
4]. These stepps are illustrateed in Fig 1.
T acquisitioon process is to get an irris image froom a person using a speccifically senssor [4]. The
The
prepprocessing proocess allowstoo remove the useless inforrmation from the iris image and to extract only the
regioon of interesst (ROI), whhich is only the iris. Thhis includes segmentationn and normaalization.The
segm
mentation proccess consists to
t isolate the iris
i ring from the iris imagee. The normallization process is applied
to produce an invvariant iris arrea.It transforrms the circullar iris region
n into a rectaangular regionn with fixed
a outer bouundary can appproximately be
b taken as ciircles.But, thee two circles
dimeensional.In faact, the inner and
are not
n co-centric. Daugman [77] proposed too apply the Inttegro-differential operator to
t detect the innner and the
outeer boundaries.
Acquisition
Segmentattion
Normalizationn
Matching
Featture Database
Feature extraaction
Similarity
value
Fig.1. Typical iris recoognition system.
T
This
method (integro-differ
(
rential) is useed to transform
m the image from the Carrtesian (x,y) to the polar
coorrdinate(r, ș) space
s
[4]. Thee center of thhe pupil is illuustrated by a green point. The center oof the iris is
illusstrated in red point
p
in Fig 2.We
2
can notiice that the tw
wo boundariess are not co-ccentric.Fig 2 iillustrates an
iris image in thee Cartesian cooordinates annd its rectanggular transform
mation result.The third is the feature
3
4
Izem Hamouchene and Saliha Aouat / AASRI Procedia 9 (2014) 2 – 7
extraaction step. Thhe goal is to capture
c
the rellevant information from im
mage. Daugmaan applied the Gabor filter
on thhe rectangularr iris. He usedd a multiscale quadrate methhod on the obtained image. This method encodes the
signs of the real and
a the imaginnary part of thhe Gabor coeffficients. Each
h pixel is encooded by two bbinary codes.
The generated binnary code is caalled ‘Iris Codde’ (Fig 2).
Iris imaage
Recttangular iris
I code
Iris
Fig.2. Normalization using
u
Daugman’ss rubber sheet moodel and extraction of the Iris codee.
F 2 illustratees an examplee of an iris imaage and its iriss code. The po
Fig.
ositive coefficcients are repreesented by a
whitte line and thee negatives byy a black line. For the Mattching step, a distance measure between a generated
iris code
c
and storred iris code is
i calculated. The query iriis is considereed authentic if
i the distancee measure is
beloow a thresholdd. Daugman haas considered the quality off the iris imag
ge, and extractted a matchingg mask from
the iris
i image. He
H used the Hamming
H
distaance (HD) to calculate thee distance bettween two iriises. For the
idenntification, Daaugman fixed the thresholdd of the matcching around 0.34 [8]. The iris recogniition system
propposed by Dauggman is often used and is a basis of all cuurrent iris reco
ognition system
ms.
Inn order to exxtract the releevant informattion from an iris image, we
w propose a novel featurre extraction
methhod. The propposed method is inspired byy the Local Binnary Pattern (LBP) methodd. Some improovements are
madde to capture the
t local inforrmation and describe
d
betteer the iris textture. This papper has been oorganized as
folloows: In sectioon 2, our propposed methodd for iris textture analyses is explained. Section 3illuustratessome
expeerimental resuults.The concluusion of this paper
p
is given in section 4.
2. Prroposed meth
hod
Inn this section, we will detaails the propoosed texture analysis
a
metho
od. This methhod is inspiredd from LBP
methhod. So, the LBP
L method iss briefly explaained. After thhat, the propossed method is detailed on tw
wo parts.Part
1 exxplains our NBP
N
method,,Part 2 illustrrates the invariance to ro
otation of NB
BP. Anillustraation which
summ
marizes the proposed archiitecture of thee IRS will be also given.Th
he Local Binarry Pattern (LB
BP) operator
was proposed by Ojala and Pieetikainen in 19996 [9].And used
u
in recent works [3][11].This methodd uses a 3x3
c
pixel is thresholded by the value of the centrall pixel. Each
analyysis window. The neighborrhood of the central
neigghbor is codedd by 1 if its vaalue is above or
o equal than the central pix
xel value, andd encoded by 0 otherwise.
Binaary code is obttained from thhe analysis wiindow and connverted into a decimal numbber LBP numbber.
B studying thhe iris recogniition system detailed
By
d
earlierr, we can notiice that the main
m
step of reecognition is
the feature
f
extracttion. A novel transformatioon, called Neigghborhood-baased Binary Paattern (NBP), is proposed
to exxtract the locaal features from
m the texture.
2.1. Neighborhood-based Binarry Pattern
T NBP extrracts the binaary pattern byy thresholdinng each neigh
The
hbor of the central
c
pixel bby the next
neigghbor (startingg from the top-left neighborr and going cllockwise). Thee binary valuee of one neighhbor is equal
to 1 if its gray vaalue is greaterr than the nexxt neighbor, 0 otherwise.Fiig 3 illustratess an example of the NBP
methhod. The first neighbor (4) has a gray leevel value lesss than its nextt neighbor (6)). Thus, its binnary code is
equaal to 0. Afterr that, the obbtained binaryy code (110110) is converrted into a decimal number (26) and
conssidered as the value of the central pixel. Finally, the central pixel in the originaal image (5) w
will have the
5
Izem Hamouchene and Saliha Aouat / AASRI Procedia 9 (2014) 2 – 7
valuue 26 in the NB
BP image. NB
BP gets the reelative connection between each neighboor of the centraal pixels Fig
3.
26
NBP
000110
010
Fig.3. Extraction of thhe NBP pattern.
2.2. Rotation invaariant Neighboorhood-basedd Binary Pattern
Inndeed, a smalll rotation of thhe same analyysis window generates
g
a diffferent NBP code.
c
In order to deal with
the rotation
r
probllem of the NBP
N
method, we
w proposed an encoding process. Thiss encoding prrocess starts
relattively from thhe higher neigghbor of the annalysis windoow.Thus, even
n if the patternn is rotated, thhe encoding
proccess gives the same code. Thhis encoding process
p
is illuustrated in Fig4
4.
NBP Code
C
Fig.4. Rotation-invariaant NBP method.
Inn order to desscribe the NBP
P image, we proposed
p
to use a decompo
osing architectture.First, the NBP image
is deecomposed innto several bloocks. The meean of each block
b
is calcu
ulated. After that,
t
the variaations of the
meanns are encoded. One blocck is encodedd by 1 if its mean are greeater than its rightneighboor’s mean, 0
otheerwise. A binaary matrix of thhe variation means
m
are extrracted and useed as template of the iris(Figg5).
NBP imagee
O
Original
imagge
Variation of meanns
Meanns
Fig.5. Mean variationss encoding processs.
F 5 illustratees the variatioon encoding process.
Fig.
p
First,, the NBP meethod is applieed on the iris image. The
obtaained NBP image is decompposed into 2*4 blocks in reed. The mean of each blockk is calculatedd. After that,
the variation
v
of thhe mean of eaach block is encoded.
e
For example, the mean of the first
f
block is equal to 26,
whicch is greater thhan its right neighbor 24. Therefore,
T
the first variation is coded by 1.
1
F the matching step, the inntersection sim
For
milarity measuure is used. In
n order to com
mpare two iriss images, the
simiilarity distancee between the two features matrixes extraacted are calcu
ulated followiing (1).
௡௕
‫ݏ݅ܦ‬ሺ‫ͳܯ‬ǡ ‫ʹܯ‬ሻ ൌ ෍ ܵሺ݅ሻ‫ ݁ݎ݄݁݁ݓ‬൜
௜ୀଵ
ܵሺ݅ሻ ൌ ͳ݂݅݅ ‫ͳܯ‬ሺ݅ሻ ൌ ‫ʹܯ‬
ʹሺ݅ሻ
ܵሺ݅ሻ ൌ Ͳ݂݅݅ ‫ͳܯ‬ሺ݅ሻ ് ‫ʹܯ‬
ʹሺ݅ሻ
(1)
6
Izem Hamouchene and Saliha Aouat / AASRI Procedia 9 (2014) 2 – 7
Where M1, M22 are the variation binary codes
W
c
of the tw
wo iris imagees. S is equal to
t 1 if the ith blocks from
M1 and
a M2 are thhe same, 0 othherwise. Nb is the number of
o blocks whicch depends onn the decompoosition of the
iris image.
i
If the distance
d
Dis iss above a threshold, the twoo irises are con
nsidered of thhe same personn.
3. Experimental results
Inn order to evaaluate our systtem performannce, we have used
u
the publiic iris databasse CASIA [10], which is a
mostt used benchm
mark in the iris recognition research.In thhe evaluation process,
p
three images from each person
havee been taken as
a a referencee and four as test.
t
In order to compare th
he LBP methood and the prooposed NBP
methhod, twenty persons
p
taken randomly froom the databaase are used in the experim
mental processs. Thus, 60
imagges are used as references and 80 as a test images. Each test im
mage is considdered as queryy. The LBP
histoogram is extraacted and the mean
m
variation of the NBP image is extrracted. After thhat, the hamm
mingdistance
betw
ween the queryy’s feature andd the features extracted from all referencce images of the
t database iss calculated.
The obtained ham
mming distances are sorted from the mostt similar to th
he dissimilar comparing
c
to tthe query. A
q
iris is classified following the m
majority. An
majoority vote of the top threee is considerred and the query
expeerimental exam
mple is illustraated in Fig 6.
Queryy iris
NBP im
mage
Feature vecttor
Hammingg
distance
Sorted top 3
Feature
databasee
Decisioon
Person 1
Majorrity vote
Person 1
Person 1
Person 2
Fig.6. Example of the recognition process
The recognition raates (R.R) of the
t two methoods are illustraated in Fig 7. The
T global ratteof the LBP m
method is
100
Using LB
BP method
80
60
Using NB
BP method
40
20
0
1 2 3 4 5 6 7 8 9 10 11 12 13 14
1 15 16 17 18 19 20
Fig.7. Recognition ratee of each person using LBP and NBP
N
Izem Hamouchene and Saliha Aouat / AASRI Procedia 9 (2014) 2 – 7
58.75% and the NBP method is 76.25%.We can notice, from the results, that the NBP method is better than
the classical LBP method.Because in fact, the neighborhood of the central pixel is thresholded by its value
using the LBP method.So, LBP get the relationship between each neighbor and the central pixel.However, the
NBP method describes each pixel by the relationship of its neighborhoods.Experimental results demonstrate
that it is more interesting to capture the relative pertinent information between the neighbors.Thus, the results
have shown the robustness and the efficiency of the proposed NBP method.
4. Conclusion
In this paper, we have proposed a new IRS system. This system used a new feature extraction method NBP.
The NBP method extracts the relative connection between neighbors of pixels. Each neighbor of each pixel is
thresholded by the next neighbor and encoded.After that, the NBP image is decomposed into several blocks.
The mean of each block is calculated. Then, the variations of the meanare encoded. The resulted binary matrix
is used as a feature descriptor of the iris. In the experimentation, the CASIA iris database is used. Good
performance has been obtained for the proposed IRS. We can summarize that the proposed NBP method is
interesting since it gets the relative relevant information between the neighbors of pixels.In future works, we
will study the combination of the NBP method with other approaches like Gabor transform.
References
[1] B.Fang,Y.Y.Tang, “Elasticregistrationforretinal images based on reconstructedvascular trees”, IEEE Trans.
on BiomedicalEngineering,vol. 53, no. 6, pp. 1183–1187, June 2006.
[2] R. Szewczyk, K. Grabowski,M. Napieralska, W. Sankowski, M. Zubert, A. Napieralski, “A reliable iris
recognition algorithm based on reverse biorthogonal wavelet transform”, Pattern RecognitionLetters, Volume
33, Issue 8, Pages 1019–1026,2012.
[3] Shu-Ren Zhou, Jian-Ping Yin, Jian-Ming Zhang,“Local binary pattern (LBP) and local phase quantization
(LBQ) based on Gabor filter for face representation”. Neurocomputing, Vol.116 ,pp. 260-264, 2013.
[4] J.G. Daugman, “High Confidence Visual Recognition of Persons by a Test of Statistical Independence”,
IEEE Trans. on Pattern Analysis and Machine Intelligence, vol. 15, no. 11, pp.1148-1161,1993.
[5] J.Daugman,C.Downing, “Epigenetic randomness, complexity, and singularity of human iris patterns”,
Proceedings of the Royal Society (London) Series B: Biological Sciences,vol. 268, pp.1737–1740. 2001.
[6] Abidin, Z.Z.; Manaf, M.; Shibghatullah, A.S.; Anawar, S.; Ahmad, R., "Feature extraction from epigenetic
traits using edge detection in iris recognition system," IEEE International Conference on Signal and Image
Processing Applications (ICSIPA), pp.145-149, 2013.
[7]J. Daugman, “How iris recognition works”, IEEE Trans. on Circuits and Systems for Video Technology,
vol. 14, no.1, pp. 21-30,2004.
[8] J. Daugman, “New methods in iris recognition”, IEEE Trans. on System, Man, and Cybernetics, vol. 37,
no. 5, pp. 1167-1175, 2007.
[9] J. Daugman “The importance of being random: statistical principles of iris recognition”, Pattern
recognition, vol. 36, no.2, pp. 279-291.2003.
[10]CASIA iris image database (v1.0), The National Laboratory of Pattern Recognition (NLPR), Institute of
Automation, Chinese Academy of Sciences (CAS), 2006.
[11] Hamouchene I. Aouat S, Lacheheb H. “New segmentation architecture for texture matching using the
LBP method”. IEEE Technically Co-Sponsored SAI Conference, London UK, 2013.
[12] L. Flom and A. Safir, “Iris Recognition System”, U.S. Patent 4,641,349, 1987.
[13] Kevin W. Bowyer, Karen P. Hollingsworth, Patrick J. Flynn, “A Survey of Iris Biometrics Research:
2008–2010”, Handbook of Iris Recognition, pp. 15-54, 2013.
7