TELKOM
NIKA
, Vol.13, No
.2, June 20
15
, pp. 597 ~ 6
0
3
ISSN: 1693-6
930,
accredited
A
by DIKTI, De
cree No: 58/DIK
T
I/Kep/2013
DOI
:
10.12928/TELKOMNIKA.v13i2.1171
597
Re
cei
v
ed
No
vem
ber 2
6
, 2014; Re
vi
sed
March 28, 20
15; Accepted
April 15, 201
5
Iris Image Recognition Based on Independent
Component Analysis and Support Vector Machine
Muhammad Fachru
rro
zi*
1
, Muhammad Mujtahid
2
Dep
a
rtment of Comp
uter Scie
nce,
Univ
esitas
Sri
w
i
j
a
y
a, In
do
nesi
a
Kampus Indr
al
a
y
a, Sumatera
Selata
n 30
662
Indon
esi
a
,
Pho
ne: +
62 71
1 58
016
9
*Corres
p
o
ndi
n
g
author, e-ma
i
l
: obetsob
e
ts@
g
mail.c
o
m
1
; mujtah
id.h
akim
@gmai
l
.com
2
A
b
st
r
a
ct
T
he iris
has
a
very un
iq
ue te
xture an
d p
a
ttern, differe
nt for eac
h i
ndiv
i
d
ual
an
d the
pa
ttern w
i
l
l
remain sta
b
le,
maki
ng p
o
ssi
ble w
hat bi
ometric techno
lo
g
y
call iris rec
o
gniti
on. In this
paper, 1
50 ir
is
imag
es fro
m
th
e De
part
m
e
n
t
of Co
mputer S
c
ienc
e, Pa
l
a
ck
y
Univ
ersity
i
n
Olomouc
iris d
a
tabas
e are us
ed
for iris reco
gniti
on b
a
sed
on i
n
dep
en
dent co
mp
on
ent an
al
y
s
is and s
u
p
port
vector machi
n
e. T
here ar
e th
ree
steps for dev
el
opi
ng this r
e
se
arch n
a
m
ely, i
m
a
ge
prepr
oc
essin
g
, feature
extraction
and
recog
n
itio
n. T
h
e
first step is i
m
age
pre
p
roces
s
ing
in
order to
get the
ir
is re
gio
n
fro
m
the
eye i
m
age. T
h
e seco
nd
is fe
atu
r
e
extraction
by
usin
g in
de
pe
n
dent co
mpo
n
e
n
t ana
lysis
in
order to
get t
he feat
ure fro
m
th
e iris
i
m
a
ge.
Supp
ort vector mach
in
e (SVM) is used for iris classifi
cati
o
n
and rec
o
g
n
iti
on. In the end
of this experi
m
ent
,
the i
m
p
l
e
m
e
n
t meth
od w
i
l
l
b
e
eva
l
uat
ed b
a
sed
up
on Ge
nui
ne Acce
pta
n
ce R
a
te (GAR). Experi
m
e
n
t
a
l
results sh
ow
that the r
e
co
gni
sed rat
e
fro
m
t
he var
i
at
i
on
of traini
ng
data
i
s
52%
w
i
th on
e dat
a trai
n, 7
3
%
w
i
th tw
o data
trains an
d 90
% three d
a
ta
trains. F
r
om
t
he exp
e
ri
ment
al resu
lt, it also show
s that this
techni
qu
e prod
uces a go
od p
e
rformanc
e.
Ke
y
w
ords
: Iri
s
Rec
ogn
itio
n, Bio
m
etric, Ind
epe
nd
ent C
o
m
pon
ent A
nalys
i
s
, Supp
ort Ve
ctor Mach
ine,
Iris
Processi
ng
1. Introduc
tion
Huma
ns h
a
ve uniqu
e and
distinctive ch
ara
c
teri
stics
su
ch a
s
face,
fingerp
r
int, voice, iri
s
and g
e
stu
r
e
s
,
with the
s
e
chara
c
te
risti
c
s able to
be u
s
ed a
s
recogni
tion or
cla
s
sifying of hum
a
n
s,
this is
kno
w
n as bi
ometric re
co
gnition
[1].
The method of ident
ification ba
se
d on bio
m
etri
c
cha
r
a
c
teri
stics is preferre
d
over traditi
onal pa
sswo
rds a
nd PIN base
d
meth
ods for va
rio
u
s
rea
s
on
s
su
ch
as the fa
ct t
hat the pe
rso
n
to be
id
enti
f
ied is
requi
red to be
phy
sically pre
s
e
n
t
a
t
the time-of-i
d
entification. I
dentificatio
n based on
biometri
c te
ch
nique
s obvia
tes the ne
ed
to
remem
b
e
r
a
password o
r
carry a
to
ken
.
A biom
et
ric
system
is e
s
sentially
a p
a
ttern recogniti
on
system
whi
c
h ma
ke
s a
person
a
l ide
n
tification by
determi
ning
the auth
enti
c
ity of a spe
c
ific
physiol
ogical
or b
ehavio
ura
l
ch
ara
c
te
risti
c
p
o
sse
s
sed
by the u
s
e
r
[2
]. Biometric te
chn
o
logie
s
are
thus define
d
as the "auto
m
ated metho
d
s of identif
ying or authe
nticating the id
entity of a living
person b
a
sed
on a physiol
ogical or be
h
a
vioural
cha
r
acteri
stic" [3].
Eyes a
r
e
one
of the
imp
o
rt
ant hu
man
senses.
Stimul
ation of
light-sen
sitive
re
ceptors i
n
the eye
(ph
o
tore
cepto
r
s)
raise
s
th
e sen
s
e
of si
g
h
t [4
]. As sh
own i
n
Figu
re
1 th
e eye
stru
ctu
r
e
con
s
i
s
ts of e
y
e scle
r
a, iris, pupil and eyelid. In
biome
t
ric syste
m
s
use
d
fo
r the identificatio
n and
detectio
n
in a
case study, the stru
ctu
r
e
o
f
the eye most often used i
s
the iri
s
.
Figure 1. Eye anatomy
Evaluation Warning : The document was created with Spire.PDF for Python.
ISSN: 16
93-6
930
TELKOM
NIKA
Vol. 13, No. 2, June 20
15 : 597 – 60
3
598
The i
r
is
ha
s a
uniqu
e p
a
ttern and
texture
in the
huma
n
eye an
d
can
not be t
r
an
sfe
rre
d o
r
faked
whi
c
h
make
s the i
r
i
s
mo
re
se
cure than oth
e
r
biometri
c
systems. The i
r
is pattern al
so
ha
s
a marvello
us
and g
r
eat st
ructure an
d m
u
ltiplies the
t
e
xture to re
cogni
se pe
rso
nal identification.
This
pap
er i
m
pleme
n
ts t
he
combi
nati
on bet
wee
n
i
ndep
ende
nt
comp
one
nt a
nalysi
s
(I
CA) as
feature extra
c
tion an
d su
p
port vecto
r
m
a
chi
n
e
s
(
SVMs) a
s
the
cl
assificatio
n
method to eval
uate
the pe
rform
a
nce
of the
G
A
R. Thi
s
pa
p
e
r al
so
co
nsi
s
ts
of thre
e
step
s: imag
e
pre
p
ro
ce
ssi
n
g
,
feature extra
c
tion by u
s
in
g indep
end
en
t compo
nent
analysi
s
then
the sup
port
vector ma
chi
ne is
us
ed for iris
class
i
fic
a
tion and rec
o
gnition.
2. Rese
arch
Metho
d
Iris re
co
gniti
on is an a
u
tomated m
e
thod of bi
ometri
c iden
tification tha
t
uses
mathemati
c
al
pattern
-reco
gnition te
chni
que
s on vi
d
e
o
image
s of
the iride
s
of an individ
ual'
s
eyes: these, compl
e
x ran
d
o
m pattern
s a
r
e uniq
ue an
d
can be
see
n
from so
me di
stan
ce [5].
Millions of p
eople from several
cou
n
trie
s have
be
en en
rolled i
n
an iri
s
re
cognition
system
s with
a variety of purpo
se
s, such a
s
pa
ssport-f
r
ee a
u
tomated bo
rd
er-crossin
g, and
some n
a
tion
al ID syste
m
base
d
on
iris re
co
gni
tion. Therefo
r
e, t
he deve
l
opment of iris
recognitio
n
is a biom
etri
c tech
nolo
g
y that has t
he p
o
t
ential to be
d
e
velope
d. Fig
u
re
2 sho
w
s the
stru
cture of an iris im
a
ge reco
gnition
system.
Figure 2. The
stru
cture of
iris image
re
co
gnition sy
ste
m
Based
on Fi
gure
2, there are three
main
ste
p
s f
o
r this
resea
r
ch
namely, image
s
prep
ro
ce
ssin
g, feature extractio
n
and
re
cog
n
ition.
2.1. Image Preproc
essin
g
Image pro
c
e
ssi
ng is a form of treatment or pro
c
e
s
sing the
imag
e as input an
d transfo
rmin
g it
into anothe
r image a
s
out
put with ce
rta
i
n techniq
u
e
s
. Image processing is
con
ducte
d to fix
the
image si
gnal
data errors caused by sig
nal acq
u
isit
io
n and tran
sm
issi
on, as wel
l
as to impro
v
e
the quality of the appeara
n
ce of the image to be mo
re ea
sily interp
reted by the huma
n
visual
system to pe
rform well a
n
d
also an
al
ysin
g the manipul
ation of the image.
In ord
e
r to
g
e
t the be
st p
a
rt of the i
r
is im
age, ima
g
e
processing
is n
e
cessa
r
y to se
parate t
he
image of the
iris from th
e
information t
hat is
not ne
eded. Vari
ou
s metho
d
s fo
r se
parating t
he
image
of the
i
r
is from
the
e
y
e image
h
a
ve be
en
co
ndu
cted. In
this study, there
a
r
e fou
r
step
s f
o
r
image p
r
ep
ro
ce
ssi
ng in order to get the
best iri
s
regi
on for the be
st recog
n
ition result:
1)
Conve
r
t to grayscale: co
nverting RGB image to
grayscale in orde
r to facilitate the next stage
.
2)
Histo
g
ra
m eq
ualisation: a t
e
ch
niqu
e for
adju
s
ting the
image inte
nsi
t
ies to en
han
ce
cont
ra
st.
The pu
rpo
s
e
of this techni
que is to prod
uce a u
n
iform
image histo
g
r
am.
Evaluation Warning : The document was created with Spire.PDF for Python.
TELKOM
NIKA
ISSN:
1693-6
930
Iris Im
age Recognition Based on Indepe
ndent Com
ponent .... (Muham
m
ad Fachrurrozi)
599
3)
Edge
dete
c
tion u
s
in
g Pre
w
itt ope
rato
r: a di
sc
reet differentiation
ope
rato
r
to
comp
ute an
approximatio
n of the gradi
ent of
the image inten
s
ity functio
n
.
4)
Conve
r
t to
Carte
s
ia
n po
lar: to get a prope
r iris
area. Carte
s
i
an into pola
r
coo
r
din
a
te
conve
r
si
on is
requi
re
d.
2.2. Featur
e Extrac
tion u
s
ing Indepe
ndent
Comp
onent
Analy
s
is (ICA)
Feature extra
c
tion i
s
the
p
r
oces
s
of extractin
g
info
rmation o
r
im
portant fe
atures of
an
image. As m
entione
d abo
ve, indepen
d
ent com
pone
nt
analysi
s
(I
CA) is a
pplie
d as a featu
r
e
extraction m
e
thod. One of the ICA algo
rithm w
ill be u
s
ed in this re
search, nam
el
y FastICA.
FastICA is a
popula
r
algo
rithm for inde
pend
ent com
pone
nt analysis
cre
a
ted b
y
Aapo
Hyvärine
n, Helsin
ki University of Tech
n
o
logy.
The al
gorithm i
s
ba
sed
on a fixe
d-poi
nt iterati
on
scheme
maxi
mising
non
-Gau
ssi
an a
s
a mea
s
u
r
e
of statistical i
ndep
ende
nce
.
It can also
b
e
derived a
s
an
approxim
atively Newto
n
iteration.
.Here i
s
the
algorith
m
of fastICA ba
se
d Fast an
d Rob
u
st Fixed
-
Point Algorit
hms for
Indepe
nde
nt Comp
one
nt Analysi
s
[6]:
1)
Cente
r
ing:
centerin
g of the input d
a
ta
x
i
s
don
e
by cal
c
ulati
ng the ave
r
age of ea
ch
comp
one
nt o
f
x
then
x
i
s
red
u
ced
b
y
the me
an.
Thi
s
h
a
s th
e effect
of
makin
g
e
a
ch
comp
one
nt having a ze
ro
mean.
(1)
2)
Whiteni
ng:
Whiteni
ng da
ta involves a
linear tra
n
sf
ormatio
n
of the data so that the ne
w
comp
one
nts
are un
co
rrela
t
ed and have
varian
ce on
e.
(2)
In this e
quat
ion,
is
the centralised o
b
se
rvation si
gnal,
z
is o
b
se
rvation
si
gnal afte
r
whiteni
ng tre
a
tment,
Λ
an
d
U
a
r
e ei
ge
nvalue mat
r
ix and eig
enve
c
tor m
a
trix of
x
covar
i
ance
matrix
, re
spe
c
tively. Eigenvalue m
a
tri
x
and
eige
n
v
ector matrix
is
obtain
e
d
according to
PCA method.
3)
Ran
domly ge
nerate
d
orth
o
gonal mat
r
ix
w-w is di
re
cte
d
toward no
rmalise
d
z.
4) Perform
the
calcul
ation
:
(3)
Where, E {...} is
the average of all c
o
lumn vec
t
ors
x
matrix.
5) Perform
n
o
rm
alisatio
n:
(4)
6)
If not converg
ed, repe
at ste
p
4.
2.3. Recog
n
ition using Support Ve
ctor Machine
SVM will b
e
u
s
ed
a
s
a te
ch
nique fo
r i
r
is i
m
age
cla
s
sification
or re
co
gnition. Th
e
workin
g
prin
ciple
of SVM is essentially only able to
h
a
ndle t
w
o-
cla
ss
cla
s
sif
i
cat
i
on [
7
]
.
Ho
w
e
v
e
r,
techni
que
s h
a
ve been de
veloped an
d multicla
ss
su
pport vecto
r
machi
n
e
s
are
able to classify
two or more
classes. Therefore,
in this
thesi
s
will use the mult
iclass SVM as
cl
assifiers for the
iris image.
There are two option
s
for impleme
n
ting the mu
lticlass SVM by combining several binary
SVM or com
b
ining all of
the data that
con
s
ist
s
of multiple cla
s
se
s into an
optimal form
of
probl
em
s. Ho
wever, the
seco
nd ap
pro
a
ch fo
r the o
p
timisation p
r
oblem to be
solved i
s
mu
ch
more
com
p
licated. He
re is a comm
on
method u
s
e
d
to implemen
t the multicla
ss SVM with
the
firs
t approac
h:
Evaluation Warning : The document was created with Spire.PDF for Python.
ISSN: 16
93-6
930
TELKOM
NIKA
Vol. 13, No. 2, June 20
15 : 597 – 60
3
600
1)
One-agai
nst
-
all method: By using this m
e
t
hod, SVM model
s build
binary k
(k i
s
the numbe
r
of
clas
se
s).
2)
One-agai
nst
-
one metho
d
: By using this meth
o
d
, build k (k-1) / 2 pieces of bina
ry
cla
ssifi
cation
model
(k i
s
t
he n
u
mb
er of
cla
s
se
s).
Th
ere
are
seve
ral meth
od
s to
pe
rform
the
test after the whol
e k (k-1)
/ 2 classificati
on model i
s
b
u
ilt. One is the method of voting
[8]
.
The
step
s of
the multi-cla
s
s SVM is
use
d
to cl
assify iris im
ages in
this research are a
s
follows
:
1)
Available da
ta denote
d
as
whe
r
e
a
s
ea
ch
lab
e
l den
oted
for
, where
l
is the numb
e
r of data. Assume
d two
cla
s
ses
-1 a
nd +1 can
b
e
compl
e
tely se
parate
d
by the hyper pla
n
e
dimensi
o
n
d
,
which is defi
ned:
(5)
2)
If all the training data sati
sf
y the const
r
ai
nts, then:
(6)
(7)
and the di
sta
n
ce b
e
twe
en
the tw
o hyperplane
s is exp
r
esse
d as:
(8)
3)
No
w, by com
b
ining (6) a
n
d
(7) into a si
n
g
le co
nstraint
, we get:
(9)
In the traini
ng pha
se, t
he main
go
al is to find
the SV that maximise
s the margi
n
of
s
e
pa
r
a
tion
,
d
. The la
rg
est
margin
coul
d be
foun
d b
y
maximisin
g
the valu
e of
the di
stan
ce
betwe
en the
hyperpl
ane a
nd the clo
s
e
s
t point, which
is
.
(10
)
This problem
can
be
solve
d
by a
vari
ety of
comp
utational te
ch
niqu
es, in
clu
d
ing
the L
agrang
e
Multiplier.
(11)
whe
r
e
is sim
u
ltaneo
usly
minimised wit
h
re
spe
c
t to
w
and
b
and
maximise
d wi
th
respec
t to
.
4)
Finally, the deci
s
ion b
oun
dary ca
n be d
e
rived a
s
follows:
(12
)
5)
If the data p
o
ints are not
sepa
rabl
e by a li
near se
paratin
g hyp
e
r plan
e, a set of slack o
r
relaxation va
riable
s
is intro
duced with
such t
hat
(
9
) b
e
com
e
:
(13
)
The sl
ack variable
s
mea
s
u
r
e the deviati
on of t
he dat
a points f
r
om
the margi
nal
hyper pla
ne.
The ne
w obje
c
tive function
to be minimised be
come
s:
Evaluation Warning : The document was created with Spire.PDF for Python.
TELKOM
NIKA
ISSN:
1693-6
930
Iris Im
age Recognition Based on Indepe
ndent Com
ponent .... (Muham
m
ad Fachrurrozi)
601
(14
)
whe
r
e
C
is t
he u
s
er-d
efin
ed pe
nalty p
a
ram
e
ter th
a
t
penali
s
e
s
a
n
y violation
of the safety
margi
n
for all the training d
a
ta.
6)
In orde
r to ob
tain a nonli
n
e
a
r de
ci
sion
b
ound
ary, we
repla
c
e th
e in
ner p
r
od
uct
of (12
)
with a nonli
n
e
a
r ke
rn
el
and
get:
(15
)
There are three types of ke
rnel
s that can be u
s
e
d
to deal with cases which
are not
linear, it can
use the h
e
lp
of a wide vari
ety
of kernel f
unctio
n
s a
s
shown in Tabl
e I.
Table 1. Kern
el Functio
n
for SVM
Kernel
K
Linear
Poly
nomial
Radial Basis Function
exp
3. Results a
nd Discu
ssi
on
Before we de
scribe the ex
perim
ents p
e
r
forme
d
to ass
e
s
s
our proposed methods
,
firs
t,
the data
b
a
s
e
employed
in
the a
s
se
ssme
nt is bri
e
fly in
trodu
ced
a
n
d
then
the
act
ual exp
e
rim
e
nts
with the co
rre
s
po
ndin
g
re
sults are p
r
e
s
e
n
ted.
3.1. Iris Data
base
The hum
an i
r
is ima
g
e
s
that are u
s
e
d
in this stud
y are not ta
ken di
re
ctly but use
secondary
data that is obtai
ned
f
r
om http://phoenix.inf.upol.cz Dep
t. Computer Sci
ence, Palacky
University in Olomou
c [8]. Iris imag
e d
a
ta is
offline
with si
ze 2
0
0
x 200 pixels and RGB (red,
gree
n, blue) f
o
rmat con
s
ist
i
ng of 25 peo
ple with six p
hotos for e
a
ch individual. The imag
es
were
taken at
di
ff
e
r
ent ho
urs a
n
d
days. T
he i
m
age fo
rmat
use
d
is P
N
G
(Portabl
e Network
Gra
phi
cs).
Figure 3 sh
o
w
s o
ne exam
ple of an eye
image that is
use
d
as in
put
data.
Figure 3. Sample imag
e from iris d
a
tab
a
se
3.2. Experimental Setting
s
Iris imag
es th
at are alre
ady
input will go
throu
gh the im
age preproce
ssi
ng sta
ge in
orde
r
to se
pa
rate t
he i
r
is imag
e
from the
eye
image.
T
h
e
i
m
age
prep
ro
ce
ssi
ng process will pro
d
u
c
e a
norm
a
lised iris image wit
h
size 5
80 x 35 pixels
. Figure 4 sho
w
s the result of
the image
prep
ro
ce
ssin
g stage. After that, the pixel val
ues from norm
a
lized iris im
age
will be use
d
to
obtain
th
e ch
ara
c
teri
stics by
usi
ng Fa
stICA
algo
rith
m
,
and th
e la
st
stage
is iri
s
i
m
age
re
co
gni
tion
by usi
ng SV
M. The
expe
ri
ment i
s
d
one
by usi
ng a
va
riation on
th
e numbe
r of
im
age
s
of different
Evaluation Warning : The document was created with Spire.PDF for Python.
ISSN: 16
93-6
930
TELKOM
NIKA
Vol. 13, No. 2, June 20
15 : 597 – 60
3
602
traine
rs. The
first test will use on
e train
i
ng data
with
three data tests, the se
cond test will
use
two training data with three data tests, and the la
st testing will be three training data with three
data tests.
Figure 4. Iris
norm
a
lized in
to polar coo
r
d
i
nates.
To evaluate
the algorith
m
, the performanc
e ca
n be mea
s
u
r
ed
by calcul
ating GAR
(Gen
uine
Acceptan
ce
Rate
). The
GA
R stated su
ccess rate of
verifi
cation of
a
system with
all t
h
e
training
data
and test d
a
ta is u
s
ed.
The high
er value indi
cates the hi
g
her GA
R sy
stem
verification
su
ccess rate. T
he eq
uation f
o
r calculat
ing
the value of
GAR
can b
e
see
n
in e
quat
ion
(16
)
.
(16
)
3.3. Experimental Result
The
re
sult of
these exp
e
riments
ha
s b
een
co
ndu
ct
ed u
s
in
g six
iris ima
ges
from 2
5
different pe
op
le, whe
r
e in
such a
n
expe
ri
ment ha
s bee
n ca
rrie
d
out
usin
g thre
e variation
s
o
n
the
amount
of tra
i
ning d
a
ta to t
he te
st data.
The first te
st
will u
s
e
one
trainin
g
ima
g
e
with th
ree
te
st
image
s, the se
con
d
test will use two trainin
g
im
age
s with thre
e test image
s, and the final
test
will use a three trai
ning i
m
age with t
h
ree te
st
image
s. Figure
5 sho
w
s th
e result of the
experim
ent
whe
r
e
we o
b
tained
52%
, 73% and
90% gen
ui
ne a
c
ceptan
ce
rate (GA
R
)
r
e
spec
tively.
Figure 5. Experime
n
tal re
sult with GAR
cal
c
ulatio
n
In orde
r to co
mpare ou
r m
e
thod with
other
exi
s
ting
method
s, sev
e
ral meth
od
s listed in
publi
s
hed p
a
pers are implemente
d
u
nder the
sa
me method
in feature e
x
traction o
r
the
recognitio
n
m
e
thod. Th
ese
method
s i
n
cl
ude
Roy [9],
Wan
g
[10], M
i
rda
w
ati [11],
Muslim
[12] a
nd
prop
osed I
C
A. From thi
s
data, we o
b
t
ained 9
8
.54
%
, 97.25%, 87.50%, 80.0
0
% and
90.0
0
%
Evaluation Warning : The document was created with Spire.PDF for Python.
TELKOM
NIKA
ISSN:
1693-6
930
Iris Im
age Recognition Based on Indepe
ndent Com
ponent .... (Muham
m
ad Fachrurrozi)
603
identificatio
n su
ccess rate
s resp
ectively. Table
2 sho
w
s the com
pari
s
on of the pro
posed metho
d
with som
e
m
e
thod
s in iris
recognitio
n
. These com
p
a
r
i
s
on
s indi
cate
our alg
o
rithm
is effective an
d
has a
n
eme
r
g
i
ng perfo
rma
n
ce in iri
s
recognition.
Table 2. Meth
ods p
e
rfo
r
ma
nce
comp
ari
s
on in iris
re
co
gnition
Me
t
h
od
s R
e
c
o
gn
i
t
io
n
R
a
te
Ro
y
[9]
98.54%
Wang [10]
97.25%
Mirda
w
ati [11]
87.50%
Muslim [12]
80.00%
Proposed
90.00%
4. Conclusio
n
The p
ape
r
h
a
s
examine
d
the d
e
velop
m
ent of
a
n
i
r
is-ba
s
e
d
recognition
sy
stem. The
Indepe
nde
nt Co
mpo
nent
Analysi
s
was i
m
plem
en
ted a
s
a fe
ature
extra
c
tion m
e
thod
while
Suppo
rt Vector Ma
chin
e
wa
s ad
opt
ed a
s
a
cla
ssifie
r
in o
r
der to d
e
vel
op an i
r
is-b
ase
d
recognitio
n
system. An
experime
n
tal study
u
s
ing the iri
s
image dat
aba
se from
the
Dep
a
rtem
ent.of Com
pute
r
Scien
c
e, P
a
lacky Univ
ers
i
ty in
Olomouc [4] was
c
a
rried out to
evaluate the
effectivene
ss of
the pro
p
o
s
ed
system.
Based
on o
b
tained
re
sul
t
s, the ICA a
n
d
SVM classifier produ
ce
s good recogni
ze re
sult
tha
t
can be me
asu
r
ed by calcul
ating G
A
R
(Gen
uine A
c
cepta
n
ce Ra
te). The re
sults sh
ow
th
at the combi
nation of ICA and SVM is a
promi
s
in
g an
d effective in iris-ba
s
e
d
re
cognition.
Referen
ces
[1]
Liu S, Mark S. A practical gui
de to biom
etric
securit
y
tec
h
n
o
lo
g
y
.
IT
Profession
al 3
. 20
01
; 1: 27-32.
[2]
Delac K, Mislav G.
A surv
ey of b
i
o
m
etri
c recog
n
iti
on
meth
ods. In E
l
ectron
ics in
Marine,
20
04
.
Procee
din
g
s El
mar 200
4.
46
th
Internatio
na
l Sy
mp
osi
u
m. IEEE. 2004: 18
4-1
93.
[3]
Santosa B. F
e
ature Se
lectio
n
w
i
t
h
Su
pport
Vector Mach
in
es App
lie
d o
n
T
o
rnado Detec
t
ion.
IPTEK
T
he Journ
a
l for
T
e
chnol
ogy a
nd Scie
nce 1
8
. 2007; 1.
[4]
Cor
w
in
EJ.
Buku Saku Pato
fi
siolo
g
i C
o
rw
in
. EGC. 2006.
[5]
Dau
g
man J. Ho
w
iris rec
o
gniti
on
w
o
rks.
Circuits and
Systems for Video Tech
n
o
lo
gy, IEEE
T
r
ansactio
n
s
. 200
4; 14(1): 21
-30.
[6]
H
y
v
ä
rin
en A,
Erkki O. Indepe
nd
ent com
pon
ent a
nal
ys
is: alg
o
rithms
and
app
licati
ons.
Ne
ura
l
netw
o
rks
. 2000
; 13(4): 411-4
3
0
.
[7]
Bos
w
e
ll
D.
Introducti
on to Su
pport Vector M
a
chi
nes
. 20
02.
[8]
Dep
a
rtement C
o
mputer Sci
e
n
c
e, Palack
y Un
iv
ersit
y
in Olo
m
ouc – http://pho
eni
x.i
n
f.upo
l.cz.
[9]
Roy
K, Prabir
B.
Iris recognit
i
on w
i
th sup
p
o
r
t vector mac
h
ines
. Adv
ance
s
in Biom
etric
s
. Spring
er
Berlin H
e
i
del
be
rg. 2005: 4
86-4
92.
[10]
W
ang Y, Jiu
QH.
Iris recog
n
itio
n usi
ng i
n
dep
en
dent co
mp
on
ent an
aly
s
is
. Machin
e
Lear
nin
g
a
n
d
C
y
ber
netics, 2
005. Proce
e
d
i
n
g
s of 200
5 In
ternatio
nal C
onfe
r
ence o
n
. IEEE. 2005: 7.
[11]
Muslim
L.
Pe
n
gen
ala
n
In
divi
du M
e
la
lui
Iris
Mata
me
ngg
un
akan
ko
mb
in
a
s
i Ind
epe
nd
ent
Co
mp
on
ent
Analys
is dan S
upp
ort Vector Machi
n
e
. Institut
T
e
knolo
g
i T
e
lkom. Ban
d
u
n
g
. 2009.
[12]
Mirda
w
at
i
S.
E
k
traksi Ciri Iris
Mata
men
g
g
u
naka
n
Meto
de
Indep
en
de
nt Co
mp
one
nt An
alysis (ICA)
.
Univers
i
tas Sri
w
i
j
a
y
a. P
a
lem
b
ang. 20
12.
Evaluation Warning : The document was created with Spire.PDF for Python.