a new texture analysis approach for iris recognition
TRANSCRIPT
AASRI Procedia 9 ( 2014 ) 2 – 7
Available online at www.sciencedirect.com
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 Institutedoi: 10.1016/j.aasri.2014.09.002
ScienceDirect
Abst
One apersmethnext convencoblockconsithe pcompmeth © 20Sele Keyw
1. In
Cprob
* E
2014
A New
a
tract
of the most imson for the authhod. The propos
neighbor to enverted into a deding process tok is calculatedidered as featur
performance of pared to the LBhod gives promi
014.Publishedection and/or p
words : Iris Recog
ntroduction
Computer visioblems. Patten r
Corresponding aE-mail address:sao
4 AASRI C
w Textur
aLRIA Laboratory
mportant authenhentication.In tsed method, Nencode it by 1 icimal number t
o obtain a rotati. After, the vare descriptor ofthe proposed IR
BP method. Expising results.
d by Elsevier Bpeer review un
gnition System; N
on is one of threcognition is
author. Tel.: [email protected].
onference o
re Analy
Izem Hamry/Computer Scien
ihamouc
ntication approathis paper, we eighborhood-baif it is greater to construct theion-invariant imariations of thef the iris. In theRS. The experimperimental resu
B.V. nder responsib
Neighborhood-bas
he most impor mainlyused t
321247187; fax: +
on Circuit an
ysis App
mouchenea,nce Department, [email protected],
aches is the irispropose a new
ased Binary Pator 0 if it is lo
e NBP image. Image. This image means are ene evaluation parments demonstrults show also t
bility of Amer
sed Binary Pattern
rtant areas of to recognize au
+21321247187.
nd Signal P
proach fo
, Saliha AoUSTHB [email protected]
s recognition syw iris recognitiottern, comparesower than the cIn order to dealge is subdivide
ncoded by a birt, the CASIA irate that the prothat the propos
rican Applied
n; Local Binary P
research, whiutomatically d
Processing (
or Iris R
ouata,* ity, Algiers,16111
dz
ystem (IRS), won system usins each neighborcentral pixel. Tl with the rotat
ed into several binary code. Theiris database [1oposed method sed system espe
Science Rese
Pattern; Texture a
ch provides edifferent entiti
CSP 2014)
ecogniti
1, Algeria
which is based g a novel featur of the central The obtained btion problem, wblocks and the e obtained bin0] has been usegives better rec
ecially the featu
earch Institute
analysis; Mean va
fficient solutiies from an im
ion
on the iris of ure extraction pixel withthe
binary code is we propose an mean of each
nary matrix is ed to evaluate cognition rate ure extraction
ariation.
ons to many mage. The
© 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
3 Izem Hamouchene and Saliha Aouat / AASRI Procedia 9 ( 2014 ) 2 – 7
secuhum
Mfunddevepassor crare physneedspeaSystpropdue broth
Tidencounthe iwork
Aloca
Tprepregiosegmto pdimeare noute
Fig.1
Tcoorillusiris
urity field hasman has particuModern securitdamental probelopment of rswordare not sracked. For thbased on bi
siological chad of reliable anaker recognitiotem (IRS) is posed by Flomto the stabilityhers or twins
The IRS is a ntification. Thntries use IRSiris identificatks of John Da
A classical irislization and n
The acquisitioprocessing proon of interesmentation procroduce an invensional.In fanot co-centricer boundaries.
. Typical iris reco
This method (rdinate(r, ) sstrated in red pimage in the
Feat
s shown a reaular propertiesty sciences usblems in securecent and effsatisfactory inhese weaknessometrics tech
aracteristics sund secure syston have been the most effi
m and Aran [1y of the huma[4]. high accurac
he IRS has beS in order to ition theory w
augman [7]. recognition s
normalization)on process is ocess allowstost (ROI), whcess consists tvariant iris ar
act, the inner a. Daugman [7
ognition system.
(integro-differspace [4]. Thepoint in Fig 2
e Cartesian co
Acquisition
ture Database
al interest in cs from others se these differeurity field. Tficient authentn many applicses, the recenthnology [1].Buch as fingerptems has involwidely studie
ficient and rel12]. Recent suan iris, its inva
cy verificationeen wildly stuimprove their
was started ear
system include), feature extra
to get an iro remove the hich is only to isolate the irea.It transforand outer bou7] proposed to
rential) is usee center of th2.We can notioordinates an
computer visisuch as shapeences to contrhe increasingtication system
cation areas. Tt science is intBiometric ideprints, iris (Figlved the emer
ed. Among allliable system
urveys of iris rariant over tim
n technology udied [4] and r security suchrlier, most im
es a series of saction and matris image frouseless inforthe iris. Th
iris ring from rms the circulundary can apo apply the Int
ed to transformhe pupil is illuice that the tw
nd its rectang
Segmentat
Similarity value
Matching
ion particularle, size etc…rol the access g need of thems.Old appro
These conventterested in autentification isg 1),and behargence of the bl the biometric [2] for authrecognition al
me and its uniq
[5],and a vaused especia
h as in airportmportant recen
steps: image atching steps [4
om a person rmation from his includes the iris image
lar iris regionpproximately btegro-differen
m the image ustrated by a
wo boundariesgular transform
tion
ly for the ide
to restricted pe security fieoaches of identional methodtomatic systems subdivided avior characterbiometric systc recognition
henticity checklgorithms canqueness for ea
alid biometrically in the sects and govern
nt works [6] h
acquisition, iri4]. These stepusing a specthe iris imagsegmentation
e. The normaln into a rectabe taken as citial operator t
from the Cargreen point.
s are not co-cmation result
Normalization
Feature extra
entification. In
places, which eld has given ntification sucds can be forgoms of identific
into two mristics such astems. Fingerprsystems, Iris k. Iris identif
n be found in ach person. Ev
cs approach fcurity fields. nment buildinhave been insp
is preprocessinps are illustratecifically sense and to extr
n and normalization proceangular regionircles.But, theto detect the in
rtesian (x,y) The center o
centric.Fig 2 i.The third is
n
action
ndeed, every
is one of the rise to the
ch as key or otten, stolen cation which
main classes: s voice. The rint,face and Recognition fication was [13]. This is ven between
for personal Thus, many gs.Although pired by the
ng (includes ed in Fig 1.
sor [4]. The act only the
alization.The ss is applied n with fixed e two circles nner and the
to the polar of the iris is illustrates an
the feature
4 Izem Hamouchene and Saliha Aouat / AASRI Procedia 9 ( 2014 ) 2 – 7
extraon thsignThe
Fig.2
Fwhitiris cbelothe iidenprop
Inmethmadfolloexpe
2. Pr
Inmeth1 exsummwas analyneigBina
Bthe fto ex
2.1.
Tneigto 1 methequacons
action step. Thhe rectangulars of the real agenerated bin
. Normalization u
Fig. 2 illustratete line and thecode and stor
ow a thresholdiris image. H
ntification, Daposed by Daugn order to exhod. The prop
de to capture tows: In sectioerimental resu
roposed meth
n this sectionhod. So, the Lxplains our Nmarizes the pproposed by ysis window.
ghbor is codedary code is obt
By studying thfeature extractxtract the loca
Neighborhoo
The NBP extrghbor (starting
if its gray vahod. The first al to 0. Aftersidered as the
Iris ima
he goal is to cr iris. He usedand the imaginnary code is ca
using Daugman’s
es an examplee negatives byred iris code id. Daugman haHe used the Haugman fixed gman is often xtract the releposed method the local inforon 2, our propults.The conclu
hod
, we will detaLBP method isNBP method,roposed archiOjala and PieThe neighbor
d by 1 if its vatained from th
he iris recognition. A novel
al features from
d-based Binar
racts the binag from the topalue is greater
neighbor (4) r that, the obvalue of the
age
capture the reld a multiscale nary part of thalled ‘Iris Cod
s rubber sheet mo
e of an iris imay a black lineis calculated. as considered
Hamming distathe threshold
used and is a evant informatis inspired byrmation and dposed methodusion of this p
ails the propos briefly expla,Part 2 illustritecture of theetikainen in 19rhood of the calue is above ohe analysis wiition system dtransformatio
m the texture.
ry Pattern
ary pattern by-left neighborr than the nexhas a gray le
btained binarycentral pixel.
Rect
levant informquadrate meth
he Gabor coefde’ (Fig 2).
odel and extractio
age and its iris. For the MatThe query irithe quality of
ance (HD) to d of the matcbasis of all cution from an
y the Local Bindescribe betted for iris textpaper is given
osed texture aained. After thrates the inve IRS will be 996 [9].And ucentral pixel ior equal than indow and condetailed earlieron, called Neig
y thresholdinr and going clxt neighbor, 0evel value lessy code (1101 Finally, the
tangular iris
ation from imhod on the obfficients. Each
n of the Iris code
s code. The potching step, a is is consideref the iris imag
calculate theching around urrent iris reco
iris image, wnary Pattern (
er the iris textture analyses in section 4.
analysis methohat, the proposariance to roalso given.Th
used in recent s thresholded the central pix
nverted into a r, we can notighborhood-ba
ng each neighlockwise). The0 otherwise.Fis than its next10) is convercentral pixel
mage. Daugmatained image.
h pixel is enco
e.
ositive coefficdistance meaed authentic i
ge, and extracte distance bet0.34 [8]. Th
ognition systemwe propose aLBP) methodture. This papis explained.
od. This methsed method is otation of NBhe Local Binar
works [3][11by the value
xel value, anddecimal numbice that the mased Binary Pa
hbor of the ce binary valueig 3 illustratest neighbor (6)rted into a din the origina
I
an applied the This method
oded by two b
cients are represure between if the distanceted a matchingtween two irie iris recognims.
a novel featurd. Some improper has been o
Section 3illu
hod is inspireddetailed on tw
BP. Anillustrary Pattern (LB].This methodof the central
d encoded by ber LBP numb
main step of reattern (NBP),
central pixel be of one neighs an example ). Thus, its binecimal numb
al image (5) w
Iris code
Gabor filter encodes the
binary codes.
esented by a a generated
e measure is g mask from ises. For the ition system
re extraction ovements are organized as ustratessome
d from LBP wo parts.Part ation which BP) operator d uses a 3x3 l pixel. Each 0 otherwise. ber.
ecognition is is proposed
by the next hbor is equal
of the NBP nary code is er (26) and
will have the
5 Izem Hamouchene and Saliha Aouat / AASRI Procedia 9 ( 2014 ) 2 – 7
valu3.
Fig.3
2.2.
Inthe rrelatproc
Fig.4
Inis demeanothe
Fig.5
Fobtathe vwhic
Fsimi
ue 26 in the NB
. Extraction of th
Rotation inva
ndeed, a smallrotation probltively from thcess gives the
. Rotation-invaria
n order to desecomposed inns are encod
erwise. A bina
. Mean variations
Fig. 5 illustrateained NBP imvariation of thch is greater th
For the matchiilarity distance
O
BP image. NB
he NBP pattern.
ariant Neighbo
l rotation of thlem of the N
he higher neigsame code. Th
ant NBP method.
scribe the NBPnto several bloed. One bloc
ary matrix of th
s encoding proces
es the variatioage is decomphe mean of eahan its right nng step, the ine between the
Original imag
BP gets the re
orhood-based
he same analyNBP method, wghbor of the anhis encoding p
P image, we pocks. The me
ck is encodedhe variation m
ss.
on encoding pposed into 2*ach block is eeighbor 24. Tntersection simtwo features
NBP
ge
Var
elative connec
d Binary Patte
ysis window gwe proposed nalysis windoprocess is illu
proposed to uean of each bd by 1 if its means are extr
process. First,4 blocks in re
encoded. For Therefore, the milarity measumatrixes extra
NBP C
iation of mean
tion between
rn
generates a difan encoding
ow.Thus, evenustrated in Fig4
se a decompoblock is calcumean are gre
racted and use
, the NBP meed. The mean example, the first variationure is used. Inacted are calcu
Code
000110
26
ns
each neighbo
fferent NBP cprocess. This
n if the pattern4.
osing architectulated. After teater than its ed as template
ethod is applieof each blockmean of the f is coded by 1n order to comulated followi
010
NBP image
Mean
or of the centra
code. In order s encoding prn is rotated, th
ture.First, the that, the varia
rightneighboof the iris(Fig
ed on the iris k is calculatedfirst block is
1. mpare two irising (1).
e
ns
al pixels Fig
to deal with rocess starts he encoding
NBP image ations of the or’s mean, 0 g5).
image. The d. After that, equal to 26,
s images, the
(1)
6 Izem Hamouchene and Saliha Aouat / AASRI Procedia 9 ( 2014 ) 2 – 7
WM1 airis i
3. E
InmosthavemethimaghistobetwThe majoexpe
Fig.6
The
Fig.7
Where M1, M2and M2 are thimage. If the d
xperimental
n order to evat used benchme been taken ahod, twenty pges are used ogram is extraween the query
obtained hamority vote of erimental exam
. Example of the
recognition ra
. Recognition rate
Query
100
80
60
40
20
01
2 are the varihe same, 0 othdistance Dis is
results
aluate our systmark in the irias a referencepersons taken as references
acted and the my’s feature and
mming distancthe top three
mple is illustra
recognition proc
ates (R.R) of t
e of each person
y iris
1 2 3 4 5
ation binary cherwise. Nb is s above a thre
tem performans recognition
e and four as trandomly fro
and 80 as a mean variatiod the featureses are sorted e is considerated in Fig 6.
ess
the two metho
using LBP and N
5 6 7 8 9
codes of the twthe number oshold, the two
nce, we have uresearch.In thtest. In order om the databa
test images. n of the NBP extracted frofrom the most
red and the q
ods are illustra
NBP
NBP im
Feature database
DecisioPerson
10 11 12 13 1
wo iris imageof blocks whico irises are con
used the publihe evaluation pto compare thase are used Each test imimage is extrm all referenct similar to th
query iris is
ated in Fig 7. T
Major
mage
e
on 1
14 15 16 17 1
es. S is equal tch depends onnsidered of th
ic iris databasprocess, three he LBP methoin the experim
mage is considracted. After thce images of the dissimilar cclassified fol
The global rat
rity vote
8 19 20
to 1 if the ith n the decompohe same person
se CASIA [10images from
od and the promental procesdered as queryhat, the hammthe database iscomparing to tlowing the m
teof the LBP m
Sort
Feature vect
Hammingdistance
Person 1 Person 1 Person 2
Using LB
Using NB
blocks from osition of the n.
], which is a each person
oposed NBP ss. Thus, 60 y. The LBP
mingdistance s calculated. the query. A
majority. An
method is
ed top 3
tor
g
BP method
BP method
7 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.