TELKOM
NIKA Indonesia
n
Journal of
Electrical En
gineering
Vol.12, No.5, May 2014, pp
. 3399 ~ 34
0
6
DOI: http://dx.doi.org/10.11591/telkomni
ka.v12i5.4951
3399
Re
cei
v
ed O
c
t
ober 2
4
, 201
3; Revi
se
d Decem
b
e
r
4, 2013; Accepte
d
De
cem
ber
22, 2013
An Effe
ctive Iris Recognition System
Hsiau Wen
L
i
n
1
, H
w
e
i
Je
n
Lin*
2
, Yue Sheng Li
2
1
Departme
n
t of Information M
ana
geme
n
t,
Chihl
ee i
n
stitute
of
T
e
chnol
og
y
Ne
w
T
a
ipei City
, T
a
iw
an, R.O.C.
2
Departme
n
t of Computer Sci
ence a
nd Info
r
m
ation En
gi
ne
erin
g, T
a
mkang Univ
ersit
y
Ne
w
T
a
ipei City
, T
a
iw
an, R.O.C.
*Corres
p
o
ndi
n
g
author, e-ma
i
l
: 0862
04@m
a
i
l
.tku.edu.t
w
A
b
st
r
a
ct
An iris r
e
cog
n
i
t
ion syste
m
us
es the ir
is to d
i
stin
g
u
ish t
he i
dentity of
a p
e
r
son us
ing
the
rich ir
i
s
texture feature.
To effectively remove noise
and pr
ec
is
ely s
e
gm
ent the stable ir
is region is a crucial stage
prior to
reco
gn
i
t
ion. Most n
o
is
es o
n
iris
i
m
a
g
e
s are
ca
used
by occl
usi
on
of eyel
ids
or ey
el
ashes
in
certai
n
areas. In this
pap
er, w
e
propose a
n
iris r
e
cog
n
itio
n system w
h
ic
h prec
isely l
o
cates a
nd seg
m
ents i
r
is
regi
ons. W
e
e
x
tract the iris feature fro
m
a
relativ
e
ly
re
lia
b
l
e p
o
rtion
of the iris re
gi
on u
s
ing
a DoG filt
er.
Experimental r
e
sults show t
hat the proposed iris reco
gnition system
has
satisfacto
ry results in ter
m
s
of
time effici
ency
and rec
o
g
n
itio
n rate.
Ke
y
w
ords
:
bi
o
m
etric rec
o
g
n
iti
on, iris reco
gnit
i
on, iris se
g
m
e
n
tation, iris n
o
r
m
a
l
i
z
a
t
i
on, feat
ure extractio
n
.
Copy
right
©
2014 In
stitu
t
e o
f
Ad
van
ced
En
g
i
n
eerin
g and
Scien
ce. All
rig
h
t
s reser
ve
d
.
1. Introduc
tion
The biom
etri
c re
cog
n
ition
techniq
ue is a scie
n
tific solutio
n
to re
cog
n
izi
ng ind
i
viduals
based upo
n biologi
cal
ch
ara
c
teri
stics
l
i
ke app
earan
ce
, i
r
is, fa
ce,
finge
rpri
nts,
voice, a
n
d
h
and
geomet
ry [1-3]. Voice and
hand g
eome
t
ry recognitio
n
are too
un
stable at their
pre
s
ent
stage
of
techn
o
logi
cal
develop
men
t. Iris recogn
ition is th
e
most p
r
e
c
i
s
e
of all bi
ome
t
ric id
entificat
ion
sy
st
em
s [
4
-1
0]
.
The iri
s
i
s
a
circula
r
stru
cture
with texture in th
e ey
e between th
e co
rne
a
an
d len
s
,
controlling
th
e si
ze
of th
e
pupil
and
the
light rea
c
hing
the
eye. In 1
987, L
eon
ard
Flom
and
Ara
n
Safir [1] con
ducte
d cli
n
ical experi
m
en
ts whi
c
h
sh
o
w
ed th
at irid
es
will no l
o
nger
ch
ang
e
from
birth to one y
ear to on
e an
d a half year l
a
ter. Ac
cordi
ng to statisti
cs, the probabi
lity of two irides
having the sa
me feature
s
is about
.
10
/
1
78
More
over, iris texture is not he
reditary; even for twins,
their iri
d
e
s
a
r
e n
o
t the
same. In a
ddi
tion, an i
r
is i
s
extre
m
ely
difficult to co
py becau
se i
t
is
locate
d in
sid
e
the
eye a
nd
the si
ze
is
ch
ange
d by
li
gh
t. Therefo
r
e,
an i
r
is offers
uniqu
ene
ss a
n
d
stability for recognition.
The
fi
rst a
u
tomatic i
r
i
s
reco
gnition
sy
stem
wa
s d
e
velope
d by
Da
ugma
n
[
4
-6]. Thi
s
system u
s
e
d
an integ
r
odiff
erential o
p
e
r
a
t
or (IDO)
to d
e
tect the inn
e
r
and o
u
ter b
ound
arie
s of the
iris. In the n
o
rmalizatio
n ste
p
, the iris
regi
on is
re
sampl
ed to a recta
ngula
r
imag
e
to comp
en
sat
e
for the vario
u
s si
ze
s of d
i
fferent eye image
s. Fi
nall
y
, a 2D Gab
o
r filter was
use
d
for feat
ure
extraction. M
ehrot
ra propo
sed
se
ctor-ba
s
ed n
o
rm
a
lization [11] to elimi
nate the range mo
st likely
to be occlud
e
d
by eyelash
e
s an
d eyelid
s.
Iris detectio
n
is a crucial
stage
for
a succe
ssf
ul
iri
s
re
cog
n
ition
system
[11-1
7
]. In this
study we
propo
se a me
thod of ir
is
feature extra
c
tion, whi
c
h
con
s
ist
s
of
four mod
u
l
e
s:
segm
entation
,
normali
zati
on, noise re
moval,
and feature
extra
c
tion. Thi
s
method p
r
e
c
i
s
ely
locate
s a
sta
b
le iri
s
re
gion
to extract reliable iri
s
featu
r
es fo
r recog
n
ition. Before
extraction
of iri
s
feature
s
, it i
s
essential
to remove the
n
o
ise,
su
ch
a
s
in the
eyelid
s a
nd
eyela
s
hes re
gion,
si
nce
the noise factors taken as
a part of the iris texture will
seri
ously dimini
sh the recognit
i
on
accuracy rate. This study
focuses o
n
locatin
g
the iris region a
n
d
sele
cts o
n
ly non-o
ccl
ud
ed
regio
n
s to extract featu
r
e
s
, in hope
s
of improvin
g re
cognition p
e
rfo
r
man
c
e.
The
re
st of t
he p
ape
r i
s
orga
nized
as follows: Se
ction 2
de
scri
bes the
prop
ose
d
i
r
i
s
recognitio
n
method. So
me expe
rim
ental results and
compa
r
iso
n
with ot
her
metho
d
s are
descri
bed in
Section 3, an
d the last
se
ction provide
s
our con
c
lu
sio
n
s.
Evaluation Warning : The document was created with Spire.PDF for Python.
ISSN: 23
02-4
0
46
TELKOM
NI
KA
Vol. 12, No. 5, May 2014: 3399 – 34
06
3400
2. Iris Recognition S
y
ste
m
The p
r
op
ose
d
iris
re
cog
n
i
tion system
includ
es th
e followin
g
steps: inn
e
r b
o
und
ary
detectio
n
, out
er b
ound
ary
detectio
n
, iri
s
norm
a
lizat
io
n, feature
extractio
n
an
d f
e
ature
matchi
ng
.
Each of the st
eps i
s
de
scrib
ed belo
w
.
2.1. Inner Bo
undar
y
Dete
ction
The inn
e
r
bo
unda
ry is lo
cated bet
wee
n
the pu
pil a
nd iri
s
. Chou
[10] use
d
conne
cted
comp
one
nt a
nalysi
s
and
a
n
ellipse fitting algo
rithm to find the pu
pil. However,
ellipse fitting
is
time co
nsumi
ng an
d n
o
t suitable fo
r ey
e imag
es
wit
h
too
mu
ch
eyelashe
s-co
vered are
a
. We
improve the
pupil dete
c
tio
n
method p
r
opo
sed by
Chou. First, a threshold val
ue T for ima
g
e
binari
z
atio
n is determi
ned
by (1), where
is the lowest intensity value and
is the averag
e
intensity of the image.
2
sw
T
(1)
And
then, co
nne
cted com
pone
nt
anal
y
s
is is used to
discar
d the
re
gion
s which a
r
e eith
e
r
too lar
ge
or
too sm
all. Fr
om the
rem
a
ining
r
egi
on
s, the on
e cl
o
s
e
s
t to the
i
m
age
ce
nter
is
sele
cted a
s
the pupil
can
d
i
date, as sho
w
n in Figu
re
1.
(a)
(b)
Figur
e 1. (a)
eye image, (b
) re
ctang
ular
r
egio
n
on the
binari
z
e
d
eye
image obtain
ed by
con
n
e
c
ted co
mpone
nt anal
ysis.
In ord
e
r to
re
move noi
se
caused by
eye
l
ash
e
s,
m
o
rp
hologi
cal ope
ning (er
o
si
on followe
d
by dilation) is
perfo
rmed. A result of
this oper
ation is
shown in Figu
r
e
2.
Figur
e 2. Re
sult of Openin
g
Oper
ation
In so
me
bina
rize
d eye
ima
ges,
hole
s
mi
ght be
cau
s
e
d
by
cam
e
ra
reflectio
n
, a
s
sho
w
n
in Figure 3. To fill up the holes, morphologi
cal cl
osing (a dilation followed by an erosi
o
n) is
perfo
rmed, a
s
sh
own in Figure 4.
Evaluation Warning : The document was created with Spire.PDF for Python.
TELKOM
NIKA
ISSN:
2302-4
046
An Effective Iris Recognitio
n
System
(Hsiau We
n Lin)
3401
(a)
(b)
Figure 3. (a)
Eye Image wi
th Reflectio
n
, (b)
Re
sult after Ope
n
ing
with holes in th
e pupil area.
Figure 4. Re
sult of Closin
g Operation
Ho
wev
e
r,
i
n
s
o
me
ca
se
s,
t
he r
e
ct
a
ngle
i
n
clu
d
e
s
pa
rt
of the eyelid,
as
sh
own in
Figure 5.
To refine the
recta
ngle, we
use the hori
z
ontal
p
r
oje
c
t
i
on and vertical proje
c
tion
of the enclo
sed
image
a
s
sho
w
n i
n
Fi
gure
6(a
)
, an
d find
the
po
siti
ons with
proje
c
tio
n
value
s
sm
al
ler th
an
a giv
e
n
threshold to form the bo
un
darie
s of a ne
w re
ctangl
e, as sho
w
n in
Figure 6(b
)
.
(a)
(b)
Figure 5. (a)
Pupil Image
Con
n
e
c
ted wi
th Eye
lid, (b) Re
sult of Morpholo
g
ical Proce
ss
(a)
(b)
Figure 6. (a)
Hori
zo
ntal Projectio
n
and
Vertic
al Proje
c
tion, (b
) A More Suitabl
e Enclo
s
ing
Re
ctangl
e Ob
tained u
s
ing
Proje
c
tion Re
sults
2.2. Outer
Boundary
Detection
The oute
r
bo
unda
ry is lo
cated between
the scl
er
a a
nd iri
s
. In mo
st ca
se
s, the
cente
r
of
the iri
s
in
ne
r
boun
dary i
s
close
to the
ce
nter of
the i
r
is outer bo
und
ary. We d
e
si
gn two ma
sks to
Evaluation Warning : The document was created with Spire.PDF for Python.
ISSN: 23
02-4
046
TELKOM
NI
KA
Vol. 12, No. 5, May 2014: 3399 – 34
06
3402
detect the lef
t
most positio
n and the rig
h
tmost po
si
tion, respe
c
tively, of the ou
ter boun
dary,
as
sho
w
n
in
Fig
u
re
7. Ea
ch
of the t
w
o m
a
sks is
ap
pli
ed
o
n
a
spe
c
ific re
ctan
gu
lar regi
on of the
image. T
he
cente
r
of
the
re
ctangl
e fo
r dete
c
ting
t
he leftmo
s
t p
o
sition
an
d t
he
cente
r
of
the
detecte
d pu
pi
l boun
dary a
r
e of the
sam
e
heig
h
t. Mo
reover, the
distance
s
from the left bou
nd
of
pupil to the right bou
nd and the left boun
d of the rectan
gle are denoted b
y
mi
n
and
max
,
r
e
spec
tively.
The values
of
mi
n
and
ma
x
ca
n be use
d
represent a
lower bo
und
and an upp
er
boun
d for the
radii diffe
ren
c
e b
e
twe
en p
upil and i
r
i
s
. The hei
ght of
the re
ctangl
e
is taken a
s
t
he
lowe
r 2/3 of
the height of
the detecte
d pupil bo
un
dary, as
sho
w
n in Fig
u
re
8. Con
s
ide
r
the
recta
ngle
for detectin
g
th
e leftmost p
o
sition
of
th
e oute
r
bo
u
ndary, in
wh
ich e
a
ch pix
e
l is
convolve
d wit
h
the m
a
sk
shown in
Figu
re 7
(
a
)
.
Rows of
convolve
d value
s
in
the recta
ngle
are
then sum
m
e
d
up, and the location
with the maximum accum
u
la
ting value is rega
rd
ed as
the
leftmost po
sit
i
on of th
e o
u
ter b
oun
da
ry. The
right
mo
st positio
n of t
he o
u
ter
bou
ndary
of the i
r
i
s
can b
e
determined in the
same m
ann
er.
(a)
(b)
Figure 7. Masks fo
r Dete
cti
ng (a
) Leftmo
s
t Posi
tion an
d (b)
Rightmo
st Position of
the Outer
Bounda
ry
Figure 8. Filter Pro
c
e
ssin
g
Ran
g
e
With both th
e leftmost po
sition an
d th
e rightmo
st p
o
sition of the
outer b
ound
ary, the
diamete
r
can
be determin
ed
by dra
w
in
g
a hori
z
o
n
ta
l line
seg
m
e
n
t, throug
h t
he
cente
r
of
the
pupil, from th
e leftmost po
sition to the ri
ghtmo
st p
o
sit
i
on, and con
s
eque
ntly, the
outer bo
und
a
r
y
is
formed, as
s
h
own in Figure 9.
Figure 9. Iris
Segmentatio
n Re
sult
2.3 Iris Normaliz
ation
After obtaini
ng the iri
s
region
I
(
x,
y
),
we n
o
rm
alize it by tran
slating it from
an xy-
coo
r
din
a
te system into a
polar coo
r
di
nate sy
ste
m
using the rubbe
r-sh
eet method [4-6], as
r
e
pr
es
e
n
t
ed
b
y
(
2
)
,
w
h
ere
γ
and
d
enote the ra
dial coo
r
di
nat
e and the an
gular
coo
r
din
a
te,
respec
tively, with
0
γ
6
4
and 0
<
360.
Evaluation Warning : The document was created with Spire.PDF for Python.
TELKOM
NIKA
ISSN:
2302-4
046
An Effective Iris Recognitio
n
System
(Hsiau We
n Lin)
3403
(
(
,)
,
(
,)
)
(
,)
Ix
y
I
(2)
That is, for pi
xel
,
in the ne
w plan
e, its value is th
e value of pixel (
,
,
,
) in
the ra
w plan
e, whe
r
e
,
and
,
can be d
e
t
ermine
d by (3) and
(4
), where
(
,
)
a
nd
(
,
) are the poi
nts lying on the inne
r bou
ndary an
d
ou
ter boun
da
ry, resp
ectively, as sh
own in
Figure 10. An example of the mappi
ng i
s
sh
own in Figure 1
1
.
64
(,
)
(
)
(
)
64
64
io
xx
x
(3)
64
(,
)
(
)
(
)
64
64
io
yy
y
(4)
Figure 10. Ru
bber-she
et Method
Figure 11. Ru
bber-she
et Method (a
) Detected Iri
s
Re
gion, (b
) the Tran
sfo
r
m Re
gion (3
60*6
4
)
(a)
(b)
Figure 12. Re
moving Pupil
by Linear Inte
rpolatio
n
Ho
wever, the
iris
regio
n
de
termine
d
in th
e above m
a
n
ner i
s
very likely to includ
e
part of
the pupil, sin
c
e the pupil i
s
not a perfe
ct circle,
as shown in Figu
re 11(a). As
sho
w
n in Fig
u
re
64
0
0
0
Evaluation Warning : The document was created with Spire.PDF for Python.
ISSN: 23
02-4
046
TELKOM
NI
KA
Vol. 12, No. 5, May 2014: 3399 – 34
06
3404
11(b
)
, in the
portion of the pupil di
st
ributed
al
ong
the upper
part of the r
e
ctan
gle afte
r
norm
a
lization
,
the bounda
r
y
between p
upil and iri
s
i
n
the recta
n
g
l
e can be d
e
tected u
s
in
g an
edge dete
c
tio
n
method, an
d then t
he portion of the pupil can b
e
r
e
moved by performi
ng line
a
r
interpol
ation
on the
ra
dial
of th
e r
e
ct
angle
alon
g t
he a
ngula
r
coordi
nate
axis. As
sh
own
in
Figur
e 12(
a), in colum
n
of
the recta
ngle,
the det
ected
pupil boun
da
ry is at
=
, a
nd s
o
the
lowe
r pa
rt of length 64
-
is o
c
cupied
b
y
the iris, whi
c
h
i
s
then
scaled by facto
r
64/(64
-
) to
fill up the
whole column.
T
he port
ion
of the pupil i
n
th
e rectangle
can be
removed by removi
ng
the po
rtion of
pupil in
ea
ch
colu
mn in thi
s
ma
nne
r, an
d at the
sam
e
time, the re
sulting r
e
cta
n
g
l
e
is furthe
r nor
malize
d
.
Mehr
otra [11]
prop
osed
se
ctor
-ba
s
e
d
n
o
rmali
z
atio
n to exclud
e so
me un
reliabl
e
regio
n
s,
whi
c
h
are
fre
quently o
c
clu
ded
by the
ey
elids an
d
eye
l
ash
e
s.
Ho
we
ver, many
un
reliabl
e
regi
o
n
s
still remai
n
, as shown in Figure 13.
Figur
e 13. Re
gion
s Rem
o
ved by Se
ctor
-
based Norm
a
lization [11]
To rai
s
e the r
e
co
gnition r
a
te, we rem
o
ve
more un
relia
ble regi
on
s, 35
146,
225
31
6, and
6
1
γ
64,
as sho
w
n
in Fi
g
u
re
14. Fin
a
ll
y, the remai
n
i
ng r
egio
n
s
ar
e mer
ged
int
o
a
smalle
r re
cta
ngle, as
sho
w
n in Figur
e 15
.
Figur
e 14. Re
gion
s Rem
o
ved by Our Me
thod
Figur
e 15. No
rmali
z
ed Iri
s
Image
2.4. Featur
e Extrac
tion
To enhance the contrast of
iris
texture caused by
we
ak illumination, histogram
equali
z
ation i
s
perfo
rme
d
on the nor
mal
i
zed ima
ge, a
s
sh
own in Figure 1
6
.
Figur
e 16. Re
sult of Histo
g
r
a
m Equali
z
ati
on on Figu
re
15
Evaluation Warning : The document was created with Spire.PDF for Python.
TELKOM
NIKA
ISSN:
2302-4
046
An Effective Iris Recognitio
n
System
(Hsiau We
n Lin)
3405
Finally, we
e
x
tract the i
r
is feature f
r
om
the imag
e o
b
tained
after
perfo
rming
hi
stogram
equali
z
ation
by co
nvolving
it with
the
Di
fference
of G
aussia
n
(Do
G
)
as defin
ed
in
(5),
wh
ere
(u,
v)
determine
s
the po
sition
of the filter
center,
σ
i
s
th
e stan
da
rd d
e
rivation
and
K is the
rati
o of
two
stand
ard
de
rivation
u
s
ed
in
ea
ch
Gau
s
sian
bl
ur. Th
e i
r
is
cod
e
i
s
set
to 1
whe
n
t
he
convolve
d value is la
rge
r
th
an ze
ro, an
d
otherwise
set to 0. The fea
t
ure code of t
he imag
e given
in Figure 10 i
s
sh
own in Figure 1
7
.
22
2
2
2
2
2
()
/
(
2
)
()
/
(
2
)
22
2
11
(,
,
)
e
x
p
e
x
p
22
uv
u
v
K
fu
v
K
(5)
Figure 17. Fe
ature Extra
c
tion Re
sult
2.5. Featur
e Matchin
g
The fe
ature
s
of the
test i
m
age
are
co
mpar
ed
with
the featu
r
e
o
f
each im
age
in th
e
databa
se u
s
i
ng Ham
m
ing
distan
ce, a
s
sho
w
n in (6),
where the A and B are two compa
r
e
d
iris
cod
e
s, N i
s
the size of each iris code, a
nd
denote
s
the exclu
s
ive
OR op
erato
r
.
1
1
()
N
ii
i
HD
A
B
N
(6)
3. Experimental Re
sults
In this sectio
n, we
will
ev
aluate
our p
r
opo
sed
meth
od fo
r iri
s
re
cog
n
ition
on
two d
a
ta
sets,
CAISA V1.0 and
CA
ISA V3.0, from the p
ubli
c
databa
se
CA
SIA [18], wh
ere
CAISA V1.0
contai
ns
756
iris imag
es from
108
different ey
e
s
(i.e
., 108 cl
asse
s)
and
CAIS
A V3.0 co
nta
i
ns
2639 i
r
is im
a
ges fo
rm 39
5 different ey
es (i.e., 39
5 cla
s
ses). All image
s in bo
th dataset
s a
r
e
320x24
0 in
size in g
r
ayscale. In
CAISA V1.0, each
cla
ss
ha
s 7 imag
es.
90 cl
asse
s
were
rega
rd
ed a
s
legal u
s
ers an
d the re
st as i
m
posto
rs
, the first 6 image
s from ea
ch l
egal cl
ass we
re
taken
a
s
train
i
ng
sampl
e
s
and th
e la
st o
ne
wa
s
k
ept f
o
r te
sting. In
CAISA V3.0,
each
cla
s
s h
a
s
1 to 26 image
s. We re
gard the 286 cla
sses with mo
re
than 5 image
s as leg
a
l users a
nd the rest
as imp
o
sto
r
s, and trai
n th
e system
by usin
g t
he first 5 image
s. The pe
rform
ance indi
ce
s
are
cho
s
e
n
as e
qual erro
r rat
e
(EER), wh
ere the fals
e accept rate
(FAR) an
d the false rej
e
ct
rate
(FRR) a
r
e e
q
ual.
In the first e
x
perime
n
t, we attempt to verify whethe
r the relia
ble
portion
we
cho
s
e i
s
more effe
ctive than that ch
ose
n
by Mehr
otra [11] or th
e whol
e iris
region.
Table 1. Re
sults of Our M
e
thod Based
on Differe
nt
Iris Re
gion Po
rtions Te
sting
on CASIA V1.0.
Whole region
Portion chosen b
y
Mehr
otra
Portion chosen b
y
our
method
EER(%)
3.81
2.35
0
FRR(
FAR=0
%
)
6.66
3.33
0
Table 2. Co
m
pari
s
on of Ou
r Method
with
the
Method
Propo
se
d by Mehrotra Te
sted on CASIA
V3.0
Mehtotra
Our
m
e
thod
FAR(
%)
4.58
3.39
FRR(
%)
3.85
3.9
EER(%)
4.23
3.73
Evaluation Warning : The document was created with Spire.PDF for Python.
ISSN: 23
02-4
046
TELKOM
NI
KA
Vol. 12, No. 5, May 2014: 3399 – 34
06
3406
In the se
co
nd
experim
ent, we
comp
are the pe
rform
a
n
c
e of o
u
r m
e
thod
with the
method
prop
osed by Mehrotra [11]
, where the result of
Mehrotra’s meth
o
d
tested on
CAISA V3.0
wa
s
taken from [11], as sho
w
n i
n
Table 2.
4. Conclusio
n
In this pa
per, we propo
sed an i
r
is
re
cog
n
ition
system whi
c
h p
r
eci
s
ely lo
cat
e
s a
n
d
segm
ents i
r
is regio
n
s. We
extract iri
s
feature
s
fro
m
a more
relia
ble po
rtion of
the iris regi
on
usin
g the
Do
G filter. Expe
rimental
re
sul
t
s sho
w
that
the feature e
x
tracted f
r
om
the remaini
n
g
reliabl
e portion is
still able to di
stinguish the identity of a pers
on.
Moreover, it
not only saves
time for detecting eyelashe
s and eyeli
d
s,
but also imp
r
oves the re
co
gnition rate.
In
our
fut
u
re work, we will take
into
a
c
count
mo
re
no
n-ide
a
l ima
g
e
s
, such a
s
iri
s
image
s
with in-pl
ane
rotation o
r
no
n-o
r
thog
onal
view iris im
ag
es.
Referen
ces
[1]
L F
l
om, A Safir. Iris recognitio
n
s
y
stem. U. S. Patent
no. 464
134
9, 198
7.
[2]
P atricia M
e
li
n, Victor Herr
era
,
Dann
iel
a
R
o
mero
, F
e
vrier
Vald
ez, Oscar
Castill
o, Geneti
c
Optimizatio
n
of Neura
l
Net
w
o
r
ks for Pers
on Rec
o
g
n
itio
n
based o
n
the
Iris.
T
E
LKOMNIKA Indon
esi
an Jour
nal
of
Electrical E
ngi
neer
ing
. 2
012:
10(2): 30
9–
320
.
[3]
Jian
bo Y
ao, T
ao Z
h
a
ng. Bi
ometric Cr
yp
t
o
s
y
stem B
a
sed
Energ
y
Attac
k
Anal
ys
is.
TEL
K
OMNIKA
Indon
esi
an Jou
r
nal of Electric
al Eng
i
ne
eri
n
g
.
2012; 1
0
(5): 1
130
–1
136.
[4]
J Daugman. High con
fi
denc
e visual
rec
o
g
n
iti
on of
p
e
rsons
b
y
a
test of
sta
t
istical
ind
e
p
e
n
denc
e.
IEEE
T
r
ans. on Pattern Ana
l
ysis a
nd Mach
ine Int
e
lli
ge
nce
. 19
93
; 15: 1148-
116
1.
[5]
J Dau
g
ma
n. H
o
w
ir
is rec
o
g
n
it
ion
w
o
rks.
IEE
E
T
r
ans. o
n
C
i
r
cuits a
nd Syst
ems
for Vi
de
o
T
e
chno
logy
.
200
4; 14: 21-3
0
[6]
J Da
ugma
n
. N
e
w
m
e
tho
d
s i
n
iris rec
o
g
n
itio
n.
IEEE Trans.
on System
s, M
a
n, and
Cyber
netics, Part B:
Cyber
netics
. 2
007; 37: 1
167-
117
5.
[7]
RP W
ild
es, JC
Asmuth, GL
Green, SC
Hs
u, RJ Ko
lcz
y
n
ski, JR Mate
y,
SE McBrid
e.
A system
f
o
r
auto
m
ate
d
iris recog
n
itio
n
. Pr
ocee
din
g
s
of the S
e
co
nd IE
EE W
o
r
ksho
p
on A
ppl
icati
o
n
s
of C
o
mpute
r
Visio
n
. 199
4; 121-1
28.
[8] RP
Wildes.
Iris recog
n
iti
on: a
n
e
m
er
gi
ng
bi
ometric tech
no
logy
. Proc
ee
di
ngs of th
e IEE
E
. 199
7; 8
5
:
134
8-13
63.
[9]
DRS Kum
a
r, K
B
Raj
a
, N
Nut
han, B
Sin
d
h
u
j
a
, P Su
pri
y
a,
RK Ch
hotar
a
y
,
S Pattna
i
k.
Iris recog
n
iti
o
n
base
d
on D
W
T
and
P
C
A
. Procee
di
ngs
of 2
0
1
1
Int
e
rnat
i
ona
l Co
nferenc
e on Comp
utation
a
l
Intelli
genc
e an
d Commu
nicati
on Net
w
o
r
ks (CICN). 201
1; 489-4
93.
[10]
Chia-T
e Ch
ou
, Sheng-W
e
n
Shih, W
en-S
h
iu
ng
C
hen,
Victor W
.
Cheng, D
uan-Y
u
Chen. N
on-
orthog
on
al vie
w
ir
is reco
gn
iti
on s
y
stem.
IEEE Trans.
Circuits an
d Systems for Vi
de
o
T
e
chnol
ogy.
201
0; 20: 417-
430.
[11]
H Mehr
otra, GS Badri
nath, B
Majhi, P Gu
pta
.
An efficie
n
t iri
s
recog
n
itio
n u
s
ing
loca
l featu
r
e descr
iptor
.
Procee
din
g
s
of 20
09
16th
IEEE Internati
o
n
a
l C
onf
er
enc
e
on Ima
ge Pr
oc
essin
g
(ICIP). 200
9; 1
957-
196
0.
[12]
Z
haofen
g He, T
i
eniu
T
an,
Z
h
ena
n S
un,
Xi
a
n
cha
o
Qiu.
Ro
bust ey
eli
d
, ey
elas
h a
n
d
sha
d
o
w
loca
li
z
a
ti
on
for iris recogn
ition
. Proce
edi
n
g
s of 15th IEEE Internation
a
l
Confere
n
ce o
n
Image Proce
ssing. 20
08;
265-
268.
[13]
T
ae-Hong Min
,
Rae-Ho
ng
Park.
Co
mpar
ison of ey
eli
d
and ey
elas
h
detectio
n
al
gorith
m
s for
perfor
m
a
n
ce
i
m
pr
ove
m
ent of
iris r
e
co
gniti
o
n
. Proceedings
of 15th IEEE
In
ternational C
onference on
Image Process
i
ng. 20
08; 25
7-
260.
[14]
Namrata P
Jo
shi, Ro
op
al K
Lamb
a
, Dev
a
n
g
U S
hah, B
h
argav V
Gha
d
i
a
.
Imp
l
e
m
e
n
tat
i
on of
vari
ous
appr
oach
e
s fo
r iris
i
m
ag
e
nor
mal
i
z
a
ti
on
.
Proce
edi
ngs
of 2
0
1
1
N
i
rma Un
iversit
y
Internati
o
n
a
l
Confer
ence
on
IEEE Engine
e
r
ing (NU
iCON
E
). 2011; 1-5.
[15]
B
y
u
ng Ju
n Ka
ng, Kan
g
R
y
o
ung Park. A r
obust
e
y
e
l
as
h detectio
n
bas
e
d
on iris foc
u
s
assessment.
Pattern Recognition Letters.
200
7; 28: 163
0
-
163
9.
[16]
R C
hen,
XR
Li
n, T
H
Din
g. Iris se
gme
n
ta
tion
for no
n-coo
perativ
e recog
n
itio
n
s
ystem.
Im
age
Processi
ng, IET
.
2011; 5: 448
-456.
[17]
DM Monr
o, S
Rakshit,
D Z
h
ang. DCT
-based iris
r
e
cog
n
it
ion.
IEEE Trans. on P
a
ttern
Analys
is and
Machi
ne Intell
i
genc
e.
200
7; 29: 586-5
96.
[18]
Chin
ese
Acad
em
y of Sci
e
n
c
es
Institute of
Automation. CASIA
V1
.0 Iris Image Database.
http://
biometrics.idealtest.org/
Evaluation Warning : The document was created with Spire.PDF for Python.