TELKOM
NIKA
, Vol.11, No
.4, Dece
mbe
r
2013, pp. 78
3~7
9
0
ISSN: 1693-6
930,
accredited
A
by DIKTI, De
cree No: 58/DIK
T
I/Kep/2013
DOI
:
10.12928/TELKOMNIKA.v11i4.1273
783
Re
cei
v
ed Ap
ril 26, 2013; Revi
sed
Jul
y
9, 2013; Accept
ed Septem
be
r 14, 2013
Separability Filter for Localizing Abnormal Pupil:
Identification of Input Image
Retno Supri
y
anti*
1
,
El
vin
Pranata
1
, Yogi Ramadh
ani
1
, Tutik Ida Rosanti
2
1
Electrica
l
Eng
i
ne
erin
g De
pt, Jend
eral S
oed
i
rman Univ
ersit
y
, Pur
w
ok
erto, Indon
esi
a
2
Medical Scie
nce De
pt, Jend
eral So
edirm
an
Universit
y
, Pur
w
o
k
erto, Indo
n
e
sia
*Corres
p
o
ndi
n
g
author, e-ma
i
l
: retno_su
p
ri
yanti@
unso
ed.a
c
.id, elvin
@
uns
oed.ac.i
d,
yo
gi.ram
adh
an
i@uns
oe
d.ac.i
d,
tutik.rosanti
@
uns
oed.ac.i
d
Abs
t
rak
Metode ya
ng
cukup h
a
n
dal
untuk p
end
ektesia
n
pu
pil a
d
a
la
h metod
e
separ
abi
lity filte
r
, namu
n
sela
ma ini
metode i
n
i h
any
a ditera
pkan
dal
am
mend
eteks
i
pup
il p
a
d
a
mata nor
ma
l saj
a
. Di lai
n
pi
ha
k
,
pad
a
mat
a
a
b
nor
mal
se
perti
pa
da
pe
nd
erita katar
a
k at
au
gl
uko
m
a
me
mi
liki
karakter
i
s
tik yan
g
b
e
rb
ed
a
den
ga
n mat
a
n
o
rmal. Pen
e
liti
an in
i me
nco
b
a
untuk
men
e
r
apka
n
sep
a
rab
ility filter dal
a
m
men
deteks
i
p
upil
pad
a
mata
a
b
nor
mal. M
a
sa
l
ah y
ang
d
i
ha
dap
i a
dal
ah
p
e
rbe
daa
n
ukur
an,
bent
uk d
an w
a
rn
a p
u
p
il,
sehi
ngg
a p
e
rlu
untuk
me
ng
i
m
p
l
e
m
e
n
tasik
a
n transfo
r
m
as
i
hou
gh, b
l
ob
area, d
an k
e
c
e
rah
an
pad
a c
i
tra
mas
u
ka
n se
be
lu
m
me
ng
gu
na
kan s
e
p
a
rab
ilit
y filter. H
a
sil
e
ksperi
m
e
n
me
nun
jukka
n
ba
h
w
a pen
a
m
b
a
h
a
n
pen
gol
ah
an aw
al citra da
pat me
ni
ngkatka
n
unj
uk kerj
a d
e
t
e
ksi pu
pil p
a
d
a
mata a
bnor
ma
l hin
gga
95.65
%.
Ka
ta
k
unc
i:
se
para
b
il
ity filter, hou
gh transfor
m
, blo
b
are
a
, kecera
han, d
e
te
ksi pup
il
A
b
st
r
a
ct
Sep
a
rab
ility fil
t
er meth
od
is a
relia
bl
e
metho
d
for pu
pil
dete
c
tion. How
e
ver
,
so far this me
thod
i
s
imple
m
ente
d
for detectin
g
p
upil
of nor
mal
eye, w
h
ile for abn
or
mal ey
e
such as catar
a
ct and gl
auco
m
a
patie
nts;
they have differe
nt
character
i
stics
of pup
il s
u
ch
a
s
color, sh
ap
e
and r
adi
us si
ze of p
upi
l. In this
pap
er
w
e
pro
pose
to
use
separ
abi
lity fi
lter for
det
ecti
ng pu
pil
of a
bnor
mal pati
e
nts
w
i
th
d
i
fferent
character
i
stics. W
e
faced
a pr
obl
e
m
a
bout r
a
dius s
i
z
e
, sh
ap
e an
d co
lor
of pup
il; ther
efore
w
e
imple
m
ent
e
d
Hou
gh T
r
ansf
o
rm, Blo
b
ar
ea
and Br
ight
ness
for ide
n
tifying
inp
u
t i
m
ag
es b
e
fore a
pply
i
n
g
separ
abi
lity filt
er.
T
he ex
peri
m
en
t results sh
ow
that w
e
can
in
crease
per
for
m
ance
of p
upi
l
d
e
tection
for a
b
nor
mal
ey
e to
b
e
95.65
%.
Ke
y
w
ords
: se
para
b
il
ity filter, hou
gh transfor
m
, bl
o
b
are
a
, b
r
ightn
e
ss, pup
il
detection
1. Introduc
tion
Rapi
d develo
p
ments of Informatio
n Technolo
g
y
espe
cially in the field of Digital Imag
e
Processin
g
g
i
ve very big i
m
pact in
ma
ny area
s. Fa
ce recognitio
n
is a p
a
rt of
fields in di
gi
tal
image
processing
techniq
u
e
s i
n
whi
c
h
iri
s
a
n
d
pu
pil
d
e
tection
a
s
o
ne of
its
pa
rts is a to
pic wi
d
e
ly
discu
s
sed in
variou
s rese
arch a
r
ea
s
such
as in me
dical fiel
d a
r
ea. In this fie
l
d, usu
a
lly p
upil
detectio
n
is i
m
pleme
n
ted for ea
rly detection of eye disea
s
e
s
such as catara
ct [1
-6].
There are m
any resea
r
ch
es that discuss
abo
ut iris and p
upil
detectio
n
. Arnia [7]
prop
osed a method that use
d
only one fourth of fu
ll normali
zed i
r
is si
ze to achieve re
cog
n
ition
rate. Putra [8] propo
se
d a low co
st b
a
se
d for eye
gaze tra
c
ki
ng system.
Cie
s
la [9] did a
resea
r
ch for l
o
cali
zin
g
pupi
l using
Web
c
am. In his re
search
He em
pha
sizes fo
r comp
ari
ng three
algorithms for locali
zing pupil. First i
s
Cumula
tive
Distribution F
u
n
c
tion (CDF
) al
gorithm,
se
co
nd
is Proje
c
tion
Functio
n
alg
o
rithm
and
th
ird i
s
E
dge
Analysi
s
. Koo
s
hke
s
tani
[10]
propo
se
d n
e
w
pupil localiza
t
ion for findin
g
the iris inn
e
r bou
nda
ry based on wa
velet transfo
rm and analyt
ic
geomet
ry rel
a
tions. Ali [1
1] did a re
se
arch fo
r lo
cali
zing in
ne
r bo
unda
ry and o
u
ter bo
und
ary of
iris u
s
in
g two regi
on p
r
o
pertie
s
Ecce
ntricity
and
Area
without
usin
g any i
t
erative meth
od.
Zhaofen
g [1
2
]
pro
p
o
s
ed
a
novel i
r
is lo
ca
lization
meth
od b
a
sed
on
a spri
ng fo
rce
-
drive
n
ite
r
ati
o
n
scheme
.
Ya
n
g
[13] propo
sed a
novel
C
aborEye
mod
e
l for eye
localizatio
n. Based on
the
spe
c
ial
gray distri
but
ion in the cye-and
-b
ro
w regio
n
,
prop
er Gab
o
r ke
rnel is ad
apt
ively chose
n
to
convol
ute with the face im
age to highli
ght the
eye-a
nd-b
r
o
w
re
gi
on, whi
c
h ca
n be exploite
d to
segm
ent the
two
pupil
region
s
efficie
n
tly. Wild
e
s
[14] propo
se
d a
metho
d
for q
u
ickly a
n
d
Evaluation Warning : The document was created with Spire.PDF for Python.
ISSN: 16
93-6
9
30
TELKOM
NIKA
Vol. 11, No. 4, Dece
mb
er 201
3: 783
– 790
784
robu
stly local
i
zing th
e iri
s
and p
upil b
o
unda
rie
s
of a
huma
n eye i
n
clo
s
e
-
up
i
m
age
s. Such
an
algorith
m
ca
n be critical
for iris id
en
tification,
or
for appli
c
atio
ns that mu
st determin
e
the
subj
ect’
s ga
ze directio
n, e.g., human
-co
m
puter i
n
tera
ction o
r
d
r
iver attentiveness dete
r
min
a
tion.
Mase
k devel
oped an ope
n
so
urce
i
r
i
s
reco
gnition sy
st
em in
order to verify both
the uni
q
ue
n
e
ss
of the hu
ma
n iri
s
an
d al
so it
s pe
rformance a
s
a
biometri
c [15
]. Fukui
and
Yamagu
chi
[16]
proposed
poi
nt feature ext
r
action
such as nostril
and pupil using separab
ility filter.
Origi
n
al
ly,
sep
a
ra
bility filter
wa
s p
r
op
ose
d
a
s
a m
e
thod fo
r
edg
e extra
c
ting
[17]. In this m
e
thod
an
edg
e i
s
defined a
s
a
boun
dary not
as a poi
nt in whi
c
h the
inte
nsity cha
nge
d dra
s
tically. Ho
wever, du
e to
the sepa
ra
bili
ty filter using
size of circle radi
u
s
a
s
a main parame
t
er, then this method re
qui
r
es
same
size of pupil ra
diu
s
o
f
input image
s relatively.
Almost all
wo
rks a
bove u
s
i
ng no
rmal
eye as
an
obje
ct. While
in f
a
ct, no
rmal
e
y
e has
different
cha
r
acteri
stics
wit
h
abn
orm
a
l e
y
e esp
e
ci
ally
in color,
sh
a
pe o
r
radiu
s
size of p
u
pil.
In
this pape
r we empha
si
ze
for applying sep
a
ra
bility fi
lter to localize pupil are
a
for abn
orm
a
l eye
su
ch
a
s
cata
ract and glau
coma
patient
automatic
all
y
. We study
to con
s
id
er f
o
r u
s
ing
Ho
u
gh
transform, bl
ob area and
brightness value to compar
e performance of usi
ng
separability filter to
locali
ze pupil
on abnorm
a
l
eye. T
he problem
will ari
s
e when
usi
n
g separability filter to locali
ze
pupil o
n
ab
n
o
rmal
pupil
d
e
tection
syst
em are radi
u
s
size, sh
ape
and col
o
r of
pupil.
Due
to the
fact that inp
u
t image
s o
f
abnormal
pupil dete
c
ti
on sy
stem i
s
taken un
d
e
r un
co
ntroll
ed
illumination a
nd also different pupil re
spon
se bet
we
en abn
ormal
and no
rmal e
y
e cau
s
e to the
differen
c
e rad
i
us si
ze, shap
e and color of
input image
s significantly.
Acco
rdi
ng
to
this ca
se, we
need
to so
rt
i
nput
im
age
s i
n
o
r
de
r to
im
prove
pe
rformance
of
pupil dete
c
tio
n
based sepa
rability filter. Image
s sort
in
g are d
one b
y
identifying chara
c
te
risti
c
s of
input ima
g
e
s
refer to the
m
a
in pa
ram
e
te
r of
sep
a
ra
bili
ty filter. In this
ca
se
we
co
nsid
er fo
r u
s
i
n
g
Hou
gh tran
sf
orm for gettin
g
fix size of pupil radi
us, bl
ob are
a
and
brightn
e
ss for segme
n
ting
eye
area.
2. Rese
arch
Metho
d
In this re
se
a
r
ch
we
used
prima
r
y dat
a in whi
c
h
we ta
ke dat
a indep
e
nd
e
n
tly. Data
acq
u
isitio
n is
done by ta
kin
g
photog
ra
ph
of patient
using co
mpa
c
t digital ca
mera. A criteri
o
n
of
input imag
e
is focused o
n
the whole
of patient
’s f
a
ce
witho
u
t backg
rou
nd
as d
e
scribed
i
n
Figure 1.
Figure 1. An example of in
put im
age (so
u
rce: privat
e documentatio
n)
The aim fo
r usin
g face
area only is fo
r re
duci
n
g
co
mputation ti
me. Also we
have to
con
s
id
er
abo
ut dista
n
ce b
e
twee
n
came
ra a
nd
patie
nt, therefo
r
e
there i
s
no
b
l
ur in
the i
n
p
u
t
image e
s
pe
ci
ally in an eye area. Th
e good level of
light focuses also an im
p
o
rtant facto
r
for
getting an ide
a
l input imag
e. Figure 2
sh
ows a flow ch
art of our research.
In the first step, input image will b
e
comp
re
ssed
for savin
g
computation ti
me and
facilitating
se
gmentation
p
r
ocess th
at will be d
one i
n
the next ste
p
. Origi
nal
size
of input im
a
ge
is quite large;
therefore we
comp
re
ss o
u
r
input
imag
e
to be n x 160 pixel size. n is image
wi
dth
that appro
p
ri
ate to
aspe
ct ratio
of origi
n
al image
As we
discu
s
sed a
bove,
input imag
e
has fa
ce a
r
ea only, therefore; by u
s
i
ng this
c
h
ar
ac
te
r
i
s
t
ic w
e
d
e
t
er
mine
re
gion
of in
terest
(ROI)
of eye a
r
ea
i
s
lo
cate
d in
the
cente
r
of
an
input image a
fter image wa
s divided into
three ho
ri
zont
al parts a
s
de
scribe
d in Fig
u
re 3.
Evaluation Warning : The document was created with Spire.PDF for Python.
TELKOM
NIKA
ISSN:
1693-6
9
30
Separability F
ilter for Lo
cali
zing Abn
o
rm
al Pupil:
Identification of Input Im
age (R Supri
y
anti)
785
2.1. Separabilit
y
F
ilter
The main purpose of using sepa
rability filter is to determin
e the center of eye roundness
by performi
n
g convolution of separ
abilit
y filter algorithm to ROI of input images
in whi
c
h the filter
is expre
s
sed
by Equation 1
.
Figure 2. Flow ch
art of re
search
0.
.
1
0
.
.
1
(1)
This filter al
so can b
e
expressed by two
vect
ors a
s
de
scribe
d in Eq
uation 2 an
d Equation 3.
0
..
1
0.
.
1
(2)
Such that for
each (I,j)
([0…(width -1)],[0……(height-1)])
∗
(3)
If the filter is sep
a
ra
ble, th
e convol
ution
operat
ion m
a
y be perfo
rmed u
s
ing o
n
l
y (width +
he
ight)
multiplications for each output pixel.
Applyi
ng the separability
filter
to Equation 2 becomes
Equation 4.
Figure 3. Reg
i
on of Interest
ROI
Inpu
t im
ag
e
Collecting c
h
aracteristics
in
fo
rm
atio
n
Pr
ocessing
and An
alysis
Co
ng
ru
en
ce iden
tificatio
n
C
l
assi
fi
cat
i
on of
i
n
put
im
ages base
d
on
congrue
nce le
vel
Id
en
tificatio
n Process
Testin
g
of
co
ng
ru
en
ce
lev
e
l
Evaluation Warning : The document was created with Spire.PDF for Python.
ISSN: 16
93-6
9
30
TELKOM
NIKA
Vol. 11, No. 4, Dece
mb
er 201
3: 783
– 790
786
(4)
This
can be
simplified to be Equation 5.
(5)
To apply th
e
sep
a
ra
ble
co
nvolution, first apply G
row
as thou
gh it were a
width
b
y
filter.
Then apply G
col
as tho
ugh
i
t
were a
by height filter.
In
ou
r resea
r
ch, the result
of this
step i
s
a
sep
a
ra
blity m
ap
im
age
that contai
ns t
he po
ssi
ble value
s
of
each
pixel as the cente
r
of circl
e
.
The greater v
a
lue of
separability in each pixel will
be the best candidate of
the circle center. In
the
se
pa
rabili
ty m
a
p
image, values
distri
bution in
whi
c
h
the greatest separabilit
y value appears
as a bri
ght do
ts as de
scrib
ed in Figu
re 4
.
Figure 4. Image re
sult
of separability filter
2.2. Non-M
a
ximum Suppression
Accordi
ng
to section 2.
1 by using
separability f
ilter we abl
e
to determine a
cent
er of eye.
The center i
s
a point that has greatest
separa
bility value. Howev
e
r the maxim
u
m separabili
ty
value i
s
di
stri
buted
around
eye
cente
r
. I
n
o
r
de
r to
re
d
u
ce
di
stribut
e
d a
r
ea
to b
e
maximum val
ues
only, therefore we imple
m
ent
non-m
a
xi
m
u
m suppression
. The
re
sult is descri
b
e
d
in Figure 5.
Figure 5. Non
-
maximum
su
ppre
s
sion im
age
2.3 Hough Tr
ansform
Hou
gh tran
sf
orm meth
od i
s
used for
cl
assifying pro
c
e
ss a
c
co
rdi
n
g to the sim
ilarity to
sep
a
ra
bility filter that u
s
u
a
lly usin
g fix radi
us. Eq
uation 6
de
scrib
ed
gene
ral equ
a
tion f
o
r
detectin
g
a ci
rcle in a
n
ima
ge.
r
2
= (
x-
a
)
2
+ (
y-b
)
2
(6)
(a,b) i
s
a ci
rcl
e
cente
r
and
r is ra
diu
s
.
The m
a
in
pu
rpo
s
e
of u
s
in
g Houg
h T
r
a
n
sform
i
s
to
get eye
shap
e, mark an
d
find the
cente
r
point o
f
eye shape.
Then
we hav
e to find co
rrelation bet
we
en eye ci
rcl
e
from se
pa
rabi
lity
separability m
a
p
image
B
r
i
ght
d
o
t
B
r
i
ght
d
o
t
After implem
enting no
n-m
a
ximum su
pp
ressio
n
Cen
t
er
Cen
t
er
Evaluation Warning : The document was created with Spire.PDF for Python.
TELKOM
NIKA
ISSN:
1693-6
930
Separability F
ilter for Lo
cali
zing Abn
o
rm
al Pupil:
Identification of Input
Im
age (R Supri
y
anti)
787
filter and eye circle from
Hough T
r
an
sfo
r
m. If we
get the same p
o
sition of circl
e
cente
r
betwe
en
both method
s, therefore
we
can jud
ge th
at this is an a
ppro
p
ri
ate inp
u
t image.
3. Results a
nd Analy
s
is
3.1. Initial Pe
rformance
In this
secti
on, we use
both of
met
hods Se
parability Filter i
n
order to
get initia
l
perfo
rman
ce
of this metho
d
when
we u
s
e for lo
calizi
ng pupil in ab
norm
a
l pupil. The re
sult
s are
sho
w
n in Ta
b
l
e 1.
Table 1. Perf
orma
nce of Separability Filter
Image (*.
jp
g)
Separabili
t
y
Fil
t
er
Success
Failed
a v
aa v
b v
bb v
c
v
cc
v
d
v
dd v
e v
ee
v
f v
ff
v
g v
gg v
h v
hh v
i
v
ii
v
j v
jj
v
k
v
kk
v
l v
Acco
rdi
ng to Table 1, whe
n
we impl
em
ent S
eparabil
i
ty filter to localize a
bno
rm
al pupil
dire
ctly, the
perfo
rman
ce
is 73.9
1
%. In othe
r ha
nd
, Houg
h T
r
a
n
sform meth
od is a fam
ous
method for
d
e
tecting a
sh
ape. The
r
efo
r
e, we in
cl
ude
this method
as a fa
ctor in
identifying the
approp
riate i
n
put ima
ge i
n
t
he im
pleme
n
tation of
sepa
rability filter.
We
have
to n
o
te he
re
that
we
impleme
n
t ori
g
inal alg
o
rith
m whi
c
h is
d
e
velope
d by Fukui [1
6-1
7
], but using
a
bnormal pu
pil
as
an obje
c
t. The result sho
w
s as in Ta
ble
1that we
ha
ve to improve algorith
m
b
y
modifying this
method u
s
in
g
colo
r an
d brightne
ss. Also we h
a
ve to
con
s
ide
r
a
b
out Ho
ugh T
r
ansfo
rm a
s
will
descri
bed in t
he followi
ng subsectio
n
.
3.2. Compari
s
on of Separabilit
y
Value
and Local Maximum
The main re
aso
n
for com
parin
g se
parability value and lo
cal ma
ximum becau
se there
are some in
p
u
t images in
whi
c
h an en
d
point is dete
c
ted a
s
a local maximum, in fact this is
not
desi
r
ed
eye p
o
int. This is
caused by sep
a
rabilit
y valu
e of eye poin
t
less tha
n
su
rro
undi
ng are
a
.
Separability will be
a
see
k
er while lo
cal maximu
m
will be
an eli
m
inator. O
n
separability, sought
the greate
s
t value in an are
a
, store this v
a
lue and
cont
inue for findin
g
sep
a
ra
bility value in other
area until getting the great
est value of
separab
ility. While for non-maxi
mum
suppression as a
method for finding l
o
cal maximum,
separability values
of
each
region is comp
aring, therefore the
greatest value will be assumed as an
eye point ca
ndidate. According to the fact that between
sep
a
ra
bility value
and
lo
cal maximum
alway
s
su
stai
nable
an
d h
a
v
e the
sam
e
value, the
r
ef
ore
both meth
od
s could
be u
s
ed fo
r id
enti
f
ying input
i
m
age in
the
impleme
n
tation of sepa
ra
bility
filter.
Evaluation Warning : The document was created with Spire.PDF for Python.
ISSN: 16
93-6
9
30
TELKOM
NIKA
Vol. 11, No. 4, Dece
mb
er 201
3: 783
– 790
788
3.3. Colour
In this step,
we emp
h
a
s
ize our id
entificat
ion of colo
ur elem
ent o
n
the analysi
s of
blob
(
b
in
ar
y la
rg
e o
b
j
ec
t)
. Blob
is
a co
lle
ctio
n
o
f
pixels that have a
neigh
bor rel
a
tionship. Blo
b
cal
c
ulatio
n p
r
oce
s
s
can
be
don
e by
an
al
yzing
neig
h
b
o
ring
pixel
s
.
Neig
hbo
ring
pixels
at a
pix
e
l is
determi
ned a
s
pixels
withi
n
one of the o
r
iginal pixel
s
.
On the Blo
b
filter process,
t
he filter is
ba
sed
on h
e
ight
and
width of
the Blob. Blo
b
with
height
or wi
dth bel
ow the
minimum
value
will be
removed from t
he object
map. Then bl
ob
was
detecte
d is l
abele
d
. The
aim for
usi
n
g Blob i
s
to
get Blob o
n
the eye a
r
ea
seg
m
entatio
n;
therefo
r
e fin
a
lly we
ca
n g
e
t eye p
o
siti
on to
be
det
ected
later.
F
i
gure
6
sh
ows a
n
exam
pl
e of
Blob are
a
se
gmentation.
Figure 6. Re
sult of Blob area se
gmentat
ion
In this
step, i
dentificatio
n
pro
c
e
s
s adju
s
ts
eye
po
siti
on segm
ente
d by Blob a
r
e
a
and
eye
position detected by
separability filter. Table 2
shows
result
s
of identification process
of thi
s
step.
Table 2. Re
sult of eye position detect
e
d
by separabili
ty filter and bl
ob are
a
Image
E
y
e
posi
t
io
n de
tecte
d
b
y
Suitabil
it
y
Separabilit
y
Filte
r
Blob Area
a right
right
V
aa right
left
X
b right
right
V
bb left
left
V
c -
left
X
cc left
right
X
d right
left,
right
V
dd left
left
V
e left
left
V
ee left
left
V
f right
right
V
ff right
right
V
g right
right
V
gg right
right
V
h right
right
V
hh left
left
V
i left
left,
right
V
ii left
left,
right
V
j left
left
V
jj right
right
V
k left
left
V
kk right
right
V
l right
right
V
Acco
rdi
ng to Table 2, only
3 of 23 images h
a
ve different eye p
o
s
ition. The
r
ef
ore Blob
area h
a
s a
ccura
cy 86.95%
.
3.4 Brightne
ss
The m
a
in
rea
s
on
why we
u
s
e
brig
htne
ss as an i
m
ag
e
element fo
r id
entification
is
base
d
on the fact that characteri
stic of
separability filter is area-bas
ed i
n
formation. While other edge
detectio
n
me
thod usi
ng i
n
formatio
n o
f
point as a
n
edge. In t
h
is expe
rime
nt, we ch
an
ge
brightn
e
ss va
lue for e
a
ch
pixel on inp
u
t
image until
we g
e
t new
brightn
e
ss va
lue. Brightn
e
ss
value cha
nge
process is o
n
ly perform
ed
on the i
nput image that previously
faile
d to detect using
Evaluation Warning : The document was created with Spire.PDF for Python.
TELKOM
NIKA
ISSN:
1693-6
930
Separability F
ilter for Lo
cali
zing Abn
o
rm
al Pupil:
Identification of Input
Im
age (R Supri
y
anti)
789
C
i
t
r
a s
eparabi
l
i
t
y
m
a
p
C
i
t
r
a s
e
pa
r
a
b
i
l
i
t
y
m
a
p
separability filter as described in T
a
ble 1.
Figure 7 descri
bes an
exampl
e of changing
brightn
e
ss va
lue in an ima
ge.
Figure 7. An Example of changi
ng bri
g
h
t
ness value
Refer to the
re
sult of thi
s
step, befo
r
e we
chan
g
e
bri
ghtne
ss value, di
stri
bution of
sep
a
ra
bility value ten
d
mo
re a
nd lo
ok li
ke n
o
ise, al
so local maxi
mum value
o
b
tained i
s
not
eye
point candi
da
te. While
after we
chan
ge
brightn
e
ss va
lue, dist
ributi
on of
sep
a
ra
bility value le
ss
visible a
nd te
nd to gath
e
r i
n
to one
edg
e
and a
point.
Local maxim
u
m value
obt
ained i
s
a
n
e
y
e
p
o
i
n
t
c
a
nd
id
ate
.
T
a
b
l
e 3
de
sc
r
i
be
s th
e
r
e
su
lt o
f
pu
p
i
l lo
ca
liz
ing
pe
r
f
o
r
ma
nc
e
afte
r
we
c
h
a
nge
brightn
e
ss va
lue.
Table 3. Perf
orma
nce of Separability F
ilter after ch
an
ging bri
ghtne
ss valu
e
Image
Brightness Chan
ging
Implementation o
f
separability
filter
Increase Decrease
Before
After
Success
F
a
iled
Success
F
a
iled
c v
v
v
v
d
v
v
v
ee v
v
v
ff v
v
v
i
v
v
v
k v
v
v
Refer to the
Table
2, by
cha
ngin
g
in
te
nsity, we
ca
n
improve
pe
rforman
c
e
of
sep
a
ra
bility fil
t
er to locali
ze
abnormal pu
pil from 73.91
% to 95.65%.
4. Conclusio
n
Acco
rdi
ng to our expe
rime
nt result
s, we
con
c
lude th
at for abno
rmal pupil, se
para
b
ility
filter
metho
d
and Ho
ugh
transfo
rm ha
s simila
rities
in use
of radi
us
value. Hou
g
h
tra
n
sfo
r
m a
l
so
can be
use as
sorting process for i
nput images
in
order to detect
pup
il based on separability
filter by givin
g
accu
ra
cy a
bout 95%. In
this re
search
we
cho
o
se fo
r u
s
ing
Ho
ug
h tran
sform a
s
a
sortin
g metho
d
becau
se se
para
b
ility filte
r
ha
s ch
ara
c
t
e
risti
c
for u
s
i
ng fix radiu
s
value and al
so it
has reliability to chang
e im
age inform
ation. We al
so note here tha
t
brightne
ss e
l
ement influe
nce
to sepa
rabilit
y value on separability filter an
d
bloob
area
segm
e
n
tation also can b
e
use to
segm
ent eye area in o
r
d
e
r
to detect pupi
l base
d
on se
para
b
ililty filte
r
.
Ackn
o
w
l
e
dg
ment
This
work is supp
orte
d by research
grant
Hi
bah
Bersai
ng Di
recto
r
ate G
e
neral of
Hig
h
e
r
Educatio
n Fiscal Yea
r
201
3 unde
r co
ntract numb
e
r 2
740/UN2
3
.10
/
PN/2013.
Referen
ces
[1]
R Supriy
anti, Hitoshi Habe, Ma
satsug
u Ki
dod
e, Satoru Nag
a
ta.
A Simp
le a
nd Ro
b
u
st Method to
Screen
Catar
a
ct usin
g S
pec
ular
Refl
ectio
n
App
eara
n
ce
.
SPIE Medic
a
l
Imagin
g
C
onfe
r
ence, S
a
n
Dieg
o
, Cal
i
forn
ia. 200
8.
[2]
R Supri
y
a
n
ti,
Hitoshi
Ha
be,
Masatsug
u K
i
do
de, Sator
u
Nag
a
ta. Catar
a
ct Screen
ing
b
y
S
pec
ular
Reflecti
on a
nd
T
e
xture Anal
ys
is, Comm
unic
a
tions of SIW
N
,
200
9. 6: 59-64.
B
e
fo
re
c
h
an
ge bri
ght
ness val
u
e
Aft
e
r
cha
n
ge
b
r
i
g
ht
ness
val
u
e
-
1
5
0
Evaluation Warning : The document was created with Spire.PDF for Python.
ISSN: 16
93-6
930
TELKOM
NIKA
Vol. 11, No. 4, Dece
mb
er 201
3: 783
– 790
790
[3]
R Supriy
anti, Hitosh
i H
a
b
e
, Masatsug
u Kid
o
de an
d Sator
u
Nag
a
ta.
Extracting App
ear
anc
e Informatio
n
insid
e
th
e P
u
p
il for
Catar
a
ct Scree
n
in
g
. IAPR Co
nfere
n
c
e
o
n
Mac
h
in
e
Visio
n
A
ppl
ic
ation. T
o
k
y
o,
Japa
n. 200
9: 342-3
45.
[4]
R Supri
y
a
n
ti,
Y Ramad
h
a
n
i,
Hitoshi
Hab
e
,
Masatsugu
Kido
de.
Perfor
ma
nce of Var
i
ous D
i
gita
l
Ca
meras for
Cataract Scre
eni
ng T
e
ch
niq
ues bas
ed o
n
Digita
l
Ima
g
e
s
. International Seminar of
Electrical
Po
w
e
r, Electron
ics, Commu
nicati
o
n
s, C
ontrol and Informatics (E
ECCI
S), Malang, East Java,
Indon
esi
a
, Dec
e
mber, 20
10.
[5] R
Supriy
anti
, Y R
a
ma
dh
an
i
.
T
he Achi
eve
m
e
n
t of Various S
hap
es of Spec
ular R
e
flecti
on
s for Cataract
Screen
ing
System Bas
ed
on
Digita
l
I
m
a
ges
.
Internati
o
n
a
l
Confer
ence
o
n
Biome
d
ic
al E
n
gin
eeri
ng
an
d
T
e
chnolog
y (IC
BET
). Kualalu
m
pur, Mala
ys
ia
. 2011:
[6]
R Supri
y
a
n
ti,
B Setia
w
a
n
, H
B
W
i
dodo, E
Murd
ya
nt
oro.
Detectin
g Pup
i
l and Iris u
n
d
e
r
Uncontro
lle
d
Illumin
a
tio
n
usi
ng F
i
xed-
Hou
g
h
Circle T
r
ansform.
Internatio
nal Jo
urna
l of Sign
al Process
i
ng, Ima
g
e
Processi
ng an
d Pattern Rec
o
gniti
on (IJSIP)
. 2012; 5(4): 1
7
5
-18
8
.
[7]
F
Arnia, N Pramita. Enha
n
c
ement of Iris
Recog
n
itio
n
S
y
stem Bas
ed on Ph
ase
Phase Onl
y
Correlation.
T
E
LKOMNIKA T
e
leco
mmunic
a
tion C
o
mp
uti
ng Electr
onics
and C
ontro
l
. 2
011: 9(
2): 387-
394.
[8]
IKG Darma Putra, A Cahy
a
w
a
n
,
Y Per
d
a
na.
Lo
w
-
C
o
st Bas
e
d E
y
e
T
r
acking
an
d E
y
e
Gaze
Estimatio
n
.
T
E
LKOMNIKA T
e
leco
mmunic
a
tion C
o
mputi
n
g Electron
ics a
nd Co
ntrol
. 20
11; 9(2): 37
7-3
86.
[9]
Michal C
i
esl
a
, Przem
y
s
l
a
w
. E
y
e P
upi
l Loc
ati
on Usi
ng W
ebc
am. Jagie
llo
ni
a
n
Univ
ersit
y
. P
o
la
nd. 20
12.
[10]
Samira K
oos
h
k
estani, Mo
ha
mmad Po
o
y
a
n
,
Ha
me
d Sa
dje
d
i. A N
e
w
Method f
o
r Iri
s
Reco
gniti
on
S
y
stems Bas
e
d on F
a
st Pupi
l
Local
izatio
n.
L
e
cture Notes i
n
Computer Sci
ence
. 20
08. 50
72: 555-
56
4.
[11]
Haid
er Ali, Ah
mad Ali
2
, Riaz
Ul Husn
ain, R
o
man Kh
an, Mohsi
n
Kha
n
, Ihsan Ull
ah Kh
a
n
. Automatic
Loca
lizati
on of
Iris Using Reg
i
on Pro
perties.
Internation
a
l J
ourn
a
l of
Co
mputer Scie
nce
and N
e
tw
ork
Security
. 201
1:
11(7): 93-9
7
.
[12]
Z
haofen
g He, T
i
eni
u T
an,
Z
hena
n Sun.
Iris Loca
l
i
z
a
t
i
on vi
a Pulli
ng a
nd P
u
shi
n
g
. T
he 18
th
Internationa
l
Confer
ence
on
Pattern Reco
g
n
it
io
n. Hon
g
ko
ng. 200
6: 366-
369.
[13]
Peng Ya
ng, B
o
Du, Shig
ua
ng Sha
n
, W
en Gao.
A Novel Pupi
l Loc
ali
z
a
t
i
on Meth
od Base
d
o
n
Gaborey
e Mo
del
an
d R
adi
a
l
Sy
mmetry Operator
. Inter
n
ation
a
l
Conf
erence
on
Imag
e Proc
essin
g
(IClP). Singa
po
re. 2004: 6
7
-70
.
[14]
T
A
Camus, R Wildes.
Re
lia
ble
and F
a
st
Eye F
i
ndi
ng
in
Close-
up I
m
a
ges
. T
he 16
th
Internatio
na
l
Confer
ence
on
Pattern Reco
g
n
itio
n (ICPR). Can
ada. 2
002:
389-3
94.
[15]
Lib
o
r Masek. Reco
gniti
on of
Human Iris Patte
rns for Biometric Ide
n
t
ification. PhD
thesis,
T
he
Univers
i
t
y
of Western Austral
i
a
, 2003.
[16]
K Fukui. Ed
ge
Extracti
on M
e
thod
B
a
sed
o
n
Sep
a
rab
ilit
y
o
f
Image Feat
ures.
IEICE T
r
an
sactions
on
Information and System
s
. 19
95; E78(1
2
): 1
533-
153
8.
[17]
K F
u
kui, Y O
s
amu. F
a
cia
l
F
eature P
o
int
Extracti
on M
e
thod
Base
d
on C
o
mbi
nati
o
n of Sh
ap
e
Ex
traction and
Pattern Matching.
Systems a
nd Co
mputers i
n
Japa
n
. 199
8; 29(6): 49-5
8
.
Evaluation Warning : The document was created with Spire.PDF for Python.