Indonesi
an
Journa
l
of El
ect
ri
cal Engineer
ing
an
d
Comp
ut
er
Scie
nce
Vo
l.
9
, No
.
3
,
Ma
rch
201
8
,
pp.
667
~
672
IS
S
N:
25
02
-
4752
,
DOI: 10
.11
591/
ijeecs
.
v9.i
3
.
pp
667
-
672
667
Journ
al h
om
e
page
:
http:
//
ia
es
core.c
om/j
ourn
als/i
ndex.
ph
p/ij
eecs
Person
ali
ty
P
rediction
Based
on Iris Positi
on
Classif
icat
i
on
Using Su
pp
or
t Ve
ctor Ma
chines
So
f
ea
Ramli
,
Sha
ri
fa
li
ll
ah
No
r
din
Facul
t
y
of
Com
pute
r & Ma
the
m
a
ti
c
al
Sc
ie
n
ce
s,
Univer
siti
T
eknologi
MA
RA,
400
00
Shah
Alam,
Sela
ngor
,
Ma
lay
si
a
Art
ic
le
In
f
o
ABSTR
A
CT
Art
ic
le
history:
Re
cei
ved
Sep
2
4
, 201
7
Re
vised
Dec
2
7
, 2
01
7
Accepte
d
Ja
n
1
6
, 2
01
8
Predic
ti
ng
per
so
nal
ity
gen
erall
y
invol
ves
per
sona
l
int
erp
re
tations
of
a
per
son
which
m
ake
s
th
e
cur
r
ent
m
et
ho
ds
for
per
sonalit
y
pre
di
ct
ion
pr
oce
ss
le
ss
ade
qua
te,
ti
m
ely
and
te
dious.
Th
us,
a
sim
ple
y
e
t
eff
icient
al
t
ern
ative
m
et
ho
d
is
proposed
in
thi
s
project
for
det
e
ct
ing
iri
s
positi
ons
which
are
used
in
Neuro
-
Li
nguist
i
c
Program
m
ing
as
cl
ues
for
the
hum
a
n
int
ern
al
re
pre
sent
at
ion
al
s
y
stem
and
m
ent
al
a
ctivity
.
Thi
s
stud
y
set
ou
t
to
det
ermin
e
seve
ra
l
posit
ion
s
of
the
iri
s
of
a
per
son
base
d
on
the
E
y
e
Ac
c
essing
C
ues.
The
design
and
t
he
developm
ent
of
a
complete
s
y
stem
will
be
und
ert
ak
en
as
for
the users
to
u
se
as
a
m
edi
um
t
o
pre
di
ct
the
ir
pe
rsonali
t
y
b
ase
d
o
n
the
i
r
ir
is
positi
on.
Sever
al
pre
-
proc
essing
te
chni
qu
es
were
exe
cu
te
d
to
each
of
the
dat
a
bef
ore
run
in
to
t
he
te
sting
and
t
ra
ini
ng
activiti
es
for
a
cc
ur
acy
ga
ini
ng.
Th
e
al
gorit
hm
used
f
or
cl
assifi
ca
t
ion
of
the
positi
ons
is
Support
Vec
t
or
Mac
hine
which
b
y
ta
k
ing
re
ctangl
e
cro
p
of
an
e
y
e
with
9
000
pixe
ls
as
in
puts.
Radial
Basis
Functi
on
i
s
used
for
the
ker
nel
par
a
m
e
te
r
o
f
the
proposed
m
et
hod.
The
cl
assifi
ca
t
ion
wi
ll
th
en
m
ap
in
to
the
t
y
pe
of
a
p
erson
with
th
e
l
ists
of
his
per
sonalit
y
base
d
on
Visual
,
Au
dit
or
y
and
Kina
esthe
tic
the
or
y
.
The
r
esult
o
f
the
cl
assifi
ca
t
ion
of
the i
ris
posi
tions
is
cur
re
n
tly
84.
9%
a
cc
ur
at
e
which
in the
future
m
ight
be
inc
re
ase
d
b
y
tuning seve
ra
l
othe
r paramete
rs t
ha
t
consiste
d
in
Support
Vec
tor
Mac
hine.
Ke
yw
or
d
s
:
Accessin
g
c
ues
Audito
ry
Ir
is
Kinaest
hetic
Neur
o
-
l
in
gu
ist
i
c
p
rogr
am
m
ing
Suppor
t
v
ect
or
m
achines
Visu
al
Copyright
©
201
8
Instit
ut
e
o
f Ad
vanc
ed
Engi
n
ee
r
ing
and
S
cienc
e
.
Al
l
rights re
serv
ed
.
Corres
pond
in
g
Aut
h
or
:
So
f
ea Ra
m
li
,
Faculty
of Com
pu
te
r
& Mat
hem
atical
Sciences
,
Un
i
ver
sit
i Te
knol
og
i M
ARA
,
40000 S
hah A
l
a
m
, S
el
ango
r,
Ma
la
ysi
a
.
Em
a
il
:
so
feara
m
li
@yaho
o.
c
om
1.
INTROD
U
CTION
Pers
on
al
it
y
is
t
he
dy
nam
ic
organ
iz
at
io
n
wit
hin
the
in
div
i
dual
of
t
ho
se
psy
choph
ysi
cal
s
yst
e
m
s
tha
t
determ
ine h
is characterist
ic
b
e
hav
i
our
an
d
th
ought”
(
Allp
or
t, 1
961). T
he M
err
ia
m
-
W
e
bst
er d
ic
ti
on
a
ry de
fines
per
s
onal
it
y
as
t
he
set
of
em
oti
on
al
qual
it
ie
s
a
nd
al
so
the
wa
ys
of
be
hav
i
ng
that
m
ake
a
p
erson
dif
fe
ren
t
fr
om
oth
e
r
pe
op
le
.
Ther
e
a
re
m
an
y
theor
ie
s
us
e
d
in
the
co
uns
el
li
ng
sessio
n
in
orde
r
to
pre
dict
or
rea
d
a
per
s
on’s
per
s
onal
it
y.
This
pr
oject
ad
op
te
d
a
neur
o
-
li
nguisti
c
pro
gr
a
m
m
ing
(N
LP
)
m
od
el
that
us
es
the
visu
al
,
a
udit
or
y
and
ki
naestheti
c
(VA
K)
cl
assifi
cat
ion
s
t
o
pr
e
dict
one’s
pe
rs
on
al
it
y
w
hich
i
n
par
ti
cular
f
oc
us
es
on
a
person’
s
le
arn
in
g
sty
le
s.
Neur
o
-
li
nguisti
c
program
m
ing
was
fi
rst
introd
uced
by
John
Gr
i
nd
e
r
an
d
Ri
chard
Ba
ndle
r
w
hose
backg
rou
nd
w
as
in
li
ng
uisti
cs
and
m
at
he
m
a
ti
cs
and
gestal
t
therap
y
res
pec
ti
vely
.
The
pur
po
s
e
of
hav
i
ng
NLP
is
to
m
ake
ex
plici
t
m
od
el
s
of
hum
an
exce
ll
ence.
Neuro
-
l
inguist
ic
pro
gra
m
m
ing
c
onsist
s
of
the
th
re
e
m
os
t
influ
e
ntial
co
m
po
nen
ts
in
volve
d
in
pro
duci
ng
hu
m
an
ex
per
ie
nce
wh
ic
h
are
ne
urolo
gy,
la
ng
uag
e
an
d
pro
gr
am
m
ing
.
The
neur
ology
syst
e
m
con
tr
ol
s
the
f
unct
ions
of
ou
r
bo
die
s
w
her
eas
the
la
nguag
e
act
s
as
the
m
edium
to
co
m
m
un
ic
at
e
with
oth
e
r
people
and
the
pro
gra
m
m
ing
par
t
de
te
rm
ines
the
m
od
el
s
of
th
e
w
or
l
d
we
create
.
Ba
s
ic
al
ly
,
neu
r
o
-
li
ng
uisti
c
pr
og
ram
m
ing
def
i
nes
the
fundam
ent
al
dynam
ic
s
betwee
n
m
ind
and
la
nguag
e
as
we
ll
as show
i
ng the e
ff
ect
s
of th
ei
r
interact
io
n
t
o our
body a
nd
b
e
hav
i
our (Dil
ts, 19
99).
Evaluation Warning : The document was created with Spire.PDF for Python.
IS
S
N
:
2502
-
4752
Ind
on
esi
a
n
J
E
le
c Eng &
Co
m
p
Sci,
Vol
.
9
,
No.
3
,
Ma
rc
h
201
8
:
667
–
672
668
Am
on
g
al
l
t
he
pe
rsonali
ty
th
eor
ie
s
t
hat
are
bein
g
ap
plied
tod
ay
,
vi
su
al
,
au
ditor
y
a
nd
kin
aest
hetic
(VAK) lear
ning sty
le
s ar
e t
he
ones that
were
f
ou
nd m
os
t su
it
able for
the
r
e
search
un
der
ta
king. Acc
ordin
g
to
a
stud
y
c
onduct
e
d
by
S
winburn
e
U
niv
e
rsity
of
Tech
nolo
gy,
t
his
co
rrel
at
ion
is
pro
bab
ly
be
cause
VAK
c
onveys
the
per
s
on
al
it
y
that
le
ads
to
t
he
way
a
per
s
on
le
arn
s
best.
Ther
e
is
no
rig
ht
or
w
r
ong
in
te
rm
s
of
the
sty
le
s
of
le
arn
in
g
of
a
pe
rson
bu
t
in
ta
king
this
oppo
rtu
nity
to
act
ually
help
tho
se
po
te
ntial
reade
rs
to
unde
rstand
t
he
ty
pe
of
le
ar
ning
that
woul
d
work
best
f
or
them
and
th
us
righ
tf
ully
adopt
that
sty
le
as
their
own
pre
ferred
le
arn
in
g
sty
le
.
In
this
ne
uro
-
li
nguisti
c
pr
ogra
m
m
ing
,
the
c
oncept
of
visu
al
acce
ssin
g
c
ue
s
is
intr
od
uced
for
a
n
app
li
cat
io
n
f
or
ey
e
featu
re
a
naly
sis
(Bra
nd
le
r
&
Gr
i
nder,
1999
).
T
he
posit
ion
s
of
t
he
iris
ca
n
be
use
d
as
ind
ic
at
ors
f
or
the
inter
nal
re
pr
ese
ntati
onal
syst
e
m
fo
r
whic
h
pa
rt
of
the
br
ai
n
is
act
iv
e
durin
g
t
he
m
ental
process
(Vrâ
nc
eanu,
Flo
rea,
F
lorea,
&
Ve
rta
n,
2015)
.
F
ur
th
erm
or
e,
bio
m
et
ric
-
base
d
m
eth
od
of
i
den
ti
fi
cat
ion
has
t
he
lo
west
error rate
th
at
leads t
o
a
good
reli
abili
ty
f
or
i
ris r
ec
ogniti
on
(S
hi
et al
., 2
009).
2.
RESEA
R
CH MET
HO
D
Dev
el
op
m
ent
of
a
syst
em
create
s
and
al
te
r
s
the
syst
e
m
u
sing
se
ver
al
pr
ocesses
,
te
chn
i
qu
e
s,
m
od
el
s,
pr
act
ic
es
an
d
m
et
ho
dolo
gies.
Fo
r
this
pro
je
ct
,
Su
pp
o
rt
Vecto
r
Ma
chine
(SVM)
as
on
e
of
the
m
achi
ne
le
arn
in
g
m
et
hod
is
us
e
d
to
de
velo
p
the
sys
tem
to
cl
assify
pa
rtic
ularly
th
e
posit
ion
s
of
the
iris
to
assi
st
the
pr
e
dicti
on
s
of
per
s
onal
it
y.
Figure
1
dem
on
s
trat
es
r
oughly
on
ho
w
the
syst
e
m
will
be
operate
d.
It
st
ar
ts
wit
h
data
colle
ct
ion
fr
om
the
us
ers.
The
data
wi
ll
be
in
te
r
m
of
ey
e
i
m
ages.
Th
us
it
s
su
ppose
d
to
be
cal
le
d
as
featur
e
s.
T
he
n,
the
f
eat
ur
e
s
will
be
up
l
oaded
int
o
the
sy
stem
us
ing
the
de
velo
ped
gr
aph
ic
al
us
er
in
te
rf
ace
(GUI) to
start t
he pre
-
pr
ocessi
ng
pr
ocedur
e
s.
Figure
1
.
Syst
em
A
rch
it
ect
ur
e
of Iris P
os
it
io
n
Cl
assifi
cat
io
n Usin
g
S
VM
Data
pr
e
-
proce
ssing
i
nclu
des
firstly
,
the
syst
e
m
will
con
ve
rt
the
i
m
age
into
gr
ay
-
scal
e
i
m
age
that
will
then
pro
ceed
to
the
e
ye
detect
ion
proce
dure
us
i
ng
Ca
scade
Object
detect
or
unde
r
Viol
a
-
Jone
s
al
gorithm
s.
At
first,
these
fe
at
ur
es
we
re
re
siz
ed
int
o
160
b
y
460
w
hich
ga
ve
73600
pi
xels
in
t
otal
f
or
eac
h
data.
U
nfo
rtu
na
te
ly
,
wh
en
the
se
huge
da
ta
w
ere
ab
ou
t
to r
un
into
the
cl
ass
ifie
r,
it
too
k
ho
ur
s f
or
the
syst
e
m
to
act
ually
finish
it
s
cl
assifi
cat
ion
process
.
T
hi
s
al
so
le
a
ds
t
o
low
syst
e
m
pe
rfor
m
ance
bec
ause
i
t
was
t
oo
tim
e
Evaluation Warning : The document was created with Spire.PDF for Python.
Ind
on
esi
a
n
J
E
le
c Eng &
Co
m
p
Sci
IS
S
N:
25
02
-
4752
Perso
nali
ty
Pred
ic
ti
on
Ba
sed
on Iris P
os
it
io
n
Cl
as
sif
ic
atio
n Usin
g S
uppo
rt
Vect
or
M
ac
hi
nes
(
Sofe
a Ra
mli
)
669
consum
ing
.
T
he
refor
e
,
in
ord
er
to
overc
om
e
this
prob
le
m
,
the
im
ages
were
resized
int
o
s
m
al
le
r
siz
e
of
75
by
120
with
9000
pix
el
s
i
n
t
otal
per
data
sam
ples,
w
hic
h
the
n
increase
s
the
perform
ance
of
the
syst
em
in
te
rm
s
of
the
cl
assifi
cat
ion
pro
ces
s
us
ing
S
VM.
Fu
rt
her
m
or
e,
f
or
feat
ur
e
e
xtracti
on,
the
syst
e
m
will
be
us
in
g
histo
gr
am
equal
isa
ti
on
f
or
i
m
age
en
ha
nce
m
ent
an
d
sm
oo
thin
g.
T
he
de
te
ct
ion
of
t
he
ey
es
ed
ges
wa
s
us
i
ng
Sobel
Edge
De
te
ct
ion
and
a
fterw
a
rds
is
the
i
m
age
Bi
nar
iz
at
ion
an
d
la
stl
y
t
he
m
or
phologi
cal
filt
ering
process.
This
filt
erin
g
e
nquires
a
n
im
a
ge
us
i
ng
a
sm
al
l
sh
ape
or
te
m
pla
te
na
m
el
y
structu
rin
g
el
e
m
ent
that
is
us
ed
to
def
i
ne
the
inter
est
r
egi
on or t
he
n
ei
gh
bourh
ood ar
ound t
he pixel.
Af
te
r
the
featu
res
hav
e
be
en
save
d
int
o
te
xt
fi
le
s,
th
ey
wil
l
then
be
trai
ni
ng
an
d
te
sti
ng
us
in
g
SV
M
cl
assifi
er.
It
sta
rts
with
trai
nin
g
the
data
colle
ct
ed
wh
ic
h
a
re
the
iris
i
m
ag
es.
The
featu
re
s
that
hav
e
bee
n
pr
e
-
processe
d
are
store
d
in
te
rm
s
of
bina
ry
valu
es
with
9000
pi
xels
in
total
of
nu
m
ber
of
al
l
21
5
data
sam
ples.
These
feat
ur
es
wer
e
the
n
c
onver
te
d
into
ta
bl
es
for
the
pur
pose
of
trai
ni
ng
us
in
g
S
VM.
T
he
co
nversi
on
of
th
e
featur
e
s
was
m
ade
us
i
ng
the
c
la
ssific
at
ion
le
arn
e
r
ap
plica
ti
on
t
hat
we
re
al
read
y
been
buil
t
in
the
MAT
LAB.
O
ut
of
the
ke
r
nels
th
at
co
ns
i
st
in
SV
M,
Ra
dial
Ba
sis
Fu
nc
ti
on
or
al
so
know
n
as
Stan
da
rd
Gaussi
an
Kernel
is
us
e
d
f
or
the
ac
tual
im
ple
m
entat
ion
of
this
sy
stem
.
Ther
e
w
ere
fi
ve
cl
asses
in
total
for
t
he
featu
res
sam
ples
to
be
cl
assifi
e
d.
By
us
in
g
on
e
t
o
m
any
for
th
e
m
ult
ic
la
ss
m
et
hod
as
well
as
fixi
ng
the
t
otal
box
co
ns
t
raint
of
value fi
ve
, th
e
perform
ance o
f
the trai
ned d
at
a that ha
ve
b
ee
n
acc
ur
at
el
y cl
assifi
ed was
op
tim
u
m
.
In
th
e
featu
re
extracti
on
ph
ases,
the
im
a
ges
are
giv
e
n
the
associat
e
d
la
bels
to
ea
ch
of
them
.
Ther
e
f
or
e,
afte
r
done
tr
ai
ning
,
if
t
he
im
age
si
m
i
la
r
in
la
be
l,
it
cl
assifi
es
as
a
gro
up
or
a
cl
ass.
If
t
he
im
a
ge
has
diff
e
re
nt
in
la
bel,
in
will
be
include
d
into
ot
he
r
cl
ass.
The
n,
the
cl
assifi
cat
i
on
m
et
ho
d
ap
pl
ie
s
te
st
ing
act
ivit
y.
Test
ing
gi
ves
the
resu
lt
of
accuracy
an
d
e
ven
t
ually
sh
ows
the
syst
e
m
per
f
orm
ance.
Pr
oc
eedi
ng
to
the
la
st
sta
ge
is
the
cl
assifi
cat
ion
of
the
i
m
ages
them
se
lves
based
o
n
the
iris
posi
ti
on
s
us
i
ng
the trained
S
VM
m
et
ho
d.
Af
te
r
the
po
sit
ion
has
been
c
la
ssifie
d,
the
ty
pe
of
pe
rs
on
and
the
li
sts
of
the
per
s
on
al
it
y
will
be
retrie
ved
t
o
the users
.
3.
RES
ULT
S
A
N
D ANAL
YS
IS
Ther
e
a
re
seve
ral
factor
s
th
at
influ
e
nce
the
accuracy
of
th
e
trai
nin
g
a
nd
te
sti
ng
sect
ions
of
the
dat
a
sam
ples
in
the
pro
j
ect
.
The
fa
ct
or
s
m
igh
t
al
s
o
le
ad
to
cha
nges
in
the
ov
e
rall
resu
lt
s
of
the
syst
e
m
.
The
factors
include
dif
fer
e
nces
in
pr
e
-
pr
oc
essing
te
c
hn
i
qu
e
s
w
hich
by
us
in
g
Hist
ogr
a
m
Equ
al
isa
ti
on,
im
ages
sh
ow
m
or
e
so
li
d
im
age
of
an
ey
e
rathe
r
than
us
in
g
Ad
a
pte
d
Histo
gr
am
Equ
al
isa
ti
on
f
or
im
age
enh
a
ncem
ent.
This
m
et
ho
d
balanc
es
the
c
olour
distrib
utions
of
it
s
RGB
c
ha
nn
el
s
so
a
s
to
pro
du
ce
m
or
e
ey
e
-
cat
chin
g
c
olou
r
s
rather
tha
n
to
com
par
e
tha
n
the
oth
er
aut
o
-
l
evel
m
et
ho
ds
(
Brindha
et
al
.
,
2017)
.
Ne
xt
f
or
feat
ur
e
e
xtra
ct
ion
,
i
m
ages
ru
n
th
r
ough
Sobel
Ed
ge
Detect
ion
e
m
ph
asi
ze
the
s
ign
ific
a
nt
li
nes
wh
ic
h
m
ade
easi
er
for
the
li
nes
of
the
iris
t
o
be
de
te
ct
ed
if
t
o
c
om
par
e
with
Ca
nn
y
Ed
ge
Dete
ct
ion
.
Last
ly
,
im
ages
resize
d
into
sm
al
le
r
siz
e
of
755
by
120
that
re
su
lt
9000
pix
el
s
i
n
t
ot
al
per
data
s
a
m
ples
increa
se
the
pe
rform
ance
of
t
he
SV
M
cl
assifi
cat
ion
.
Table
1
pr
ese
nts
the
dif
fer
e
nces
of
the
ac
cur
acy
i
nf
lue
nc
ed
by
di
ff
e
re
nt
rati
o
of
the
trai
ning
a
nd
te
sti
ng
s
plit
pe
rcen
ta
ge.
Thes
e
trai
ning
a
nd
te
sti
ng
act
ivit
ie
s
wer
e
co
n
duct
ed
us
i
ng
Ra
dial
Ba
sis
F
unct
ion
(RBF)
ke
rn
el
f
or
SV
M
data
t
rainin
g.
T
he
sp
li
t
per
ce
nta
ge
s
are
of
20
:
80%,
25:7
5%,
30
:
70%
a
nd
35:6
5%
rati
os
.
H
ow
e
ve
r,
am
ong
al
l
the
s
plit
rati
os
,
25:7
5%
gaine
d
the
hi
ghest
per
ce
ntage
of
it
s
cl
assifi
cat
ion
wit
h
84.9%.
Wh
il
e,
T
able
2
belo
w
sh
ows
t
he
num
ber
of
data
f
or
each
cl
assifi
cat
ion
ob
ta
ine
d
in
the
data
colle
c
ti
on
for
the
s
plit
r
at
io w
it
h t
he hi
ghest
acc
ur
acy
,
25
:
75%.
Table
1
.
Sp
li
t r
at
io for
data te
sti
ng
a
nd trai
nin
g
Sp
lit Ratio
(
%)
Accurac
y
(
%)
Er
ror
(
%)
Testin
g
Tr
ain
in
g
20
80
7
9
.1
2
0
.1
25
75
8
4
.9
1
5
.1
30
70
8
2
.8
1
7
.2
35
65
8
0
.0
2
0
.0
Table
2
.
Sam
pl
e d
at
a
for
trai
nin
g an
d
te
sti
ng
with
rati
o 25
:
75%
Po
sitio
n
Top
Rig
h
t
Top
L
ef
t
Cen
tre
Cen
tre
Rig
h
t
Cen
tre
Left
Total
Tr
ain
in
g
34
37
34
30
27
162
Testin
g
11
13
11
10
8
53
Total
45
50
45
40
35
215
Evaluation Warning : The document was created with Spire.PDF for Python.
IS
S
N
:
2502
-
4752
Ind
on
esi
a
n
J
E
le
c Eng &
Co
m
p
Sci,
Vol
.
9
,
No.
3
,
Ma
rc
h
201
8
:
667
–
672
670
Table
3
disp
la
ys
the
sa
m
ple
data
that
ha
ve
been
pr
e
-
proce
ssed.
T
he
feat
ur
es
wer
e
pr
e
s
ented
based
on their
posit
io
n of
t
he
iris
w
hi
ch
are t
op r
i
ght, top l
eft, ce
ntr
e, cen
t
re r
i
gh
t
and cent
re left.
Th
ese
featu
res
will
then
be
trai
ne
d
an
d
te
ste
d
w
hich
e
ven
t
ually
cl
assifi
ed
usi
ng
S
VM
to
obta
in
their
pos
it
ion
an
d
ret
rieve
th
e
resu
lt
of p
e
rs
onal
it
y.
T
able
3
.
Sam
pl
e of traini
ng
da
ta
To
p
Ri
ght
To
p
Le
ft
Ce
ntre
Ce
ntre
Ri
ght
Ce
ntre
Le
ft
Fo
r
pa
ram
et
er
s
involve
d
in
Suppor
t
Vecto
r
Ma
chin
e
cl
assifi
er
su
c
h
as
t
he
m
ai
n
key
is
the
ke
r
nel
us
e
d,
box
co
nst
raints,
scal
es
that
al
so
kn
own
as
gam
m
a
and
ot
her
s
.
Ba
s
ed
on
ta
ble
4
wh
ic
h
s
hows
s
ever
a
l
ty
pes
of
ke
r
ne
l
wer
e
us
e
d
in
the
trai
nin
g
proces
s
for
the
pur
po
se
of
co
m
par
ing
the
hi
gh
est
cl
assifi
cat
ion
accuracy
usi
ng
MATLAB
cl
a
ssific
at
ion
le
ar
ner
a
pp
li
cat
ion.
This
ap
par
e
nt
ly
rev
eal
ed
tha
t
by
us
ing
RB
F,
the
cl
assifi
cat
ion
proces
s wo
uld g
ive b
et
te
r
r
es
ul
ts t
han the
oth
e
rs.
Table
4
.
Acc
uracy
r
esults
bas
ed on ke
r
nels
Kernel
Accurac
y
(
%)
Er
ror
(
%)
Linear
8
3
.7
1
6
.3
Qu
ad
ratic
8
2
.8
1
7
.2
Cu
b
ic
8
1
.3
1
8
.7
RBF
8
4
.9
1
5
.1
Af
te
r
se
ver
al
t
rial
s
that
hav
e
been
done
us
in
g
dif
fer
e
nt
kind
of
par
am
et
ers,
it
can
be
co
nclu
ded
t
hat
Stand
a
r
d
Ga
ussi
an
or
RB
F
ke
rn
el
produces
bette
r
res
ults
in
te
rm
s
of
it
s
cl
assifi
cat
ion
accuracy
a
nd
l
ow
e
r
error
rate
on
t
he
da
ta
that
ha
s
bee
n
m
isc
la
s
sifi
ed
if
t
o
be
com
par
ed
to
F
ine
Ga
us
sia
n,
Me
diu
m
Gau
s
sia
n
or
Course
Ga
us
si
an.
The
refo
re,
ou
t
of
the
ke
rn
el
s
that
co
ns
ist
in
SV
M,
RB
F
will
be
us
ed
f
or
the
act
ual
i
m
ple
m
entat
io
n of t
his
pro
j
ec
t sy
stem
.
Ther
e
are
va
rie
ty
te
chn
iq
ues
i
n
Ar
ti
fici
al
I
nt
el
li
gen
ce
th
at
c
an
be
us
e
d
for
cl
assifi
cat
ion
proces
s.
T
he
te
chn
iq
ues
incl
ud
e
Ba
ck
pro
pa
gation
Neural
Netw
ork
(BP
N
N)
,
S
uppo
rt
Vector
Ma
chine
(S
VM)
,
K
-
Ne
arest
Neig
hbour
,
Pe
rcep
tr
on,
an
d
so
m
e
lot
oth
ers
m
or
e.
In
thi
s
par
ti
cular
proj
ect
,
the
m
ain
cl
assifi
er
ch
os
e
n
is
Supp
or
t
Vect
or
Ma
chi
ne,
an
d
for
t
he
c
omparis
on
pu
rpos
es,
BP
NN
is
s
el
ect
ed
to
c
ompare
the
res
ult
of
it
s
accuracy.
T
his
is
becau
se
ba
sed
f
ro
m
few
resea
rch
re
views
,
A
NN
a
nd
S
VM
we
re
two
m
os
t
con
sist
ent
m
achine
le
arn
i
ng
m
et
ho
ds
use
d
f
or
cl
assifi
cat
ion
proce
ss
and
one
of
th
e
exam
ple
was
a
facial
exp
r
ession
cl
assifi
cat
ion
f
or in
pu
t i
nto
e
m
ot
ion
us
in
g Ar
ti
fici
al
N
e
ural
N
et
w
orks (S
alm
a
m
et al
.,
20
17
).
These
F
ig
ur
e
s
of
2,
3,
a
nd
4
pr
ese
nt
t
he
r
esults
of
t
he
i
r
is
posit
ion
cl
assifi
cat
ion
us
i
ng
A
rtific
ia
l
Neural
Net
wor
k
(
A
NN).
Alth
ough
the
t
raini
ng
res
ult
of
A
NN
te
c
hniq
ue
reache
d
100%
bu
t
t
he
vali
dation
an
d
te
sti
ng
pa
rts
of
the
data
is
qu
it
e
low
w
hich
are
71.
9%
a
nd
68.8%
res
pect
ively
.
A
par
t
f
r
om
the
resu
lt
of
t
he
cl
assifi
cat
ion
process
us
in
g
ANN
is
not
sta
ble
an
d
co
ns
i
ste
nt,
SV
M
c
ontrib
utes
to
a
higher
per
ce
nt
age
in
te
rm
s o
f
it
s acc
ur
acy
a
nd this
value
a
dded
to
the r
eas
ons
on
us
in
g
S
VM i
nst
ead of A
NN f
or this
pro
j
ect
.
Evaluation Warning : The document was created with Spire.PDF for Python.
Ind
on
esi
a
n
J
E
le
c Eng &
Co
m
p
Sci
IS
S
N:
25
02
-
4752
Perso
nali
ty
Pred
ic
ti
on
Ba
sed
on Iris P
os
it
io
n
Cl
as
sif
ic
atio
n Usin
g S
uppo
rt
Vect
or
M
ac
hi
nes
(
Sofe
a Ra
mli
)
671
Figure
2
.
Trai
ni
ng
Mat
ri
x
Figure
3
.
Vali
da
ti
on
Mat
ri
x
Figure
4
.
Test
Ma
trix
4.
C
ONCLU
S
ION
So
m
e
extensiv
e
and
ext
ra
hard
w
ork
that
ha
s
been
put
toge
ther
in
the
ef
fort
of
the
com
pleti
on
of
this
pro
j
ect
has
co
m
e
to
an
end
.
The
overall
di
sco
ver
y
in
this
pr
oject
,
it
is
pro
ved
that
th
e
ste
ps
on
im
a
ge
pre
-
processi
ng
m
eth
ods
f
or
the
e
ye
s
i
m
a
ges
is
way
m
or
e
c
ompli
cat
ed
a
nd
diff
ic
ult
t
hat
defi
nitel
y
need
e
d
extra
effor
t
to
be
co
m
ple
te
d.
For
this
pa
rtic
ular
s
yst
e
m
,
so
m
e
of
these
rec
omm
end
at
ions
w
ould
be
able
to
help
f
or
the
sake
of
th
e
fu
tu
re
w
ork.
They
include
prov
i
ding
a
lot
m
or
e
of
im
ages
to
b
e
pro
cessed
s
o
as
the
iris
po
sit
io
ns
co
ul
d
be
bette
r
trai
ned
a
nd
cl
assif
ie
d
par
ti
cula
rly
in
the
Su
pp
ort
Vector
Ma
c
hin
e
(
SV
M
)
m
et
hod.
Ap
a
rt
f
r
om
that,
by
m
od
ify
ing
t
he
pr
e
-
pr
ocess
m
et
ho
ds
w
ou
l
d
be
m
uch
hel
p
the
r
efore
to
im
pr
ov
e
it
s
eff
ic
ie
ncy
of
the
syst
em
fo
r
instance
by
a
pply
ing
a
di
ff
e
r
ent
m
e
tho
d
for
ey
e
detect
ion
su
c
h
as
us
in
g
Gabo
r
Fil
te
r
as
pr
op
ose
d
by
K
.
Sud
hak
a
r
an
d
P.
N
it
hyanandam
(
2017)
.
This
is
fo
r
the
pur
pose
of
m
aking
bette
r
detect
ion
s
of
the
iris
po
sit
io
n
of
the
ey
e
im
ages
without
gr
antin
g
the
c
on
t
ro
l
to
the
us
ers
on
the
m
anu
al
set
ti
ng
s.
Last
ly
,
the
GUI
of
th
e
syst
e
m
wo
ul
d
be
way
m
uch
bette
r
with
the
bette
rm
ent
of
s
om
e
go
od
gr
a
phic
us
a
ge
an
d
us
e
r
f
rien
dly.
H
oweve
r,
al
l
the
chall
en
ges
ha
ve
bee
n
acce
pt
ed
al
so
with
the
accu
racy
f
act
or
pro
vid
e
d
us
i
ng
Suppor
t Vecto
r
Ma
chine
(SV
M)
is
acce
ptab
le
fo
r
t
he
w
hole
pro
j
ect
com
pleti
on
.
Be
sides
,
wit
h
the Go
d’
s
w
il
li
ng, th
is
r
esea
rc
h pro
j
ect
h
as
c
om
plete
d
al
l of the
processes
in
the
r
a
nge
of
t
i
m
e g
iven.
ACKN
OWLE
DGE
MENT
S
The
a
uthor
s
gr
at
efu
ll
y
ack
nowled
ge
t
he
hel
p
of
t
he
Mi
nist
ry
of
Scie
nce,
Tech
no
l
og
y
a
nd
I
nnovat
io
n
(MOST
I)
of
M
al
ay
sia
in
pro
vi
din
g
the
Scie
nc
e
Fun
d
resear
ch
gr
a
nt
a
nd
U
niv
e
rsiti
Tek
nolog
y
Ma
ra
(
Ui
TM)
for
m
aking
ava
il
able
it
s
Fel
lows
hi
p
schem
e.
The
auth
or
s
a
re
al
so
than
kfu
l
to
the
kn
owle
dg
e
p
r
ov
i
ders
of
the
trai
nees
of
ce
rtifie
d
c
ouns
el
lors
t
hat
s
pecifica
ll
y
in
Ne
uro
-
Lin
guist
ic
Pr
og
ram
m
ing
(N
L
P)
from
K
&
J
Con
s
ultancy
in
Ban
gi,
Ma
la
ysi
a.
Evaluation Warning : The document was created with Spire.PDF for Python.
IS
S
N
:
2502
-
4752
Ind
on
esi
a
n
J
E
le
c Eng &
Co
m
p
Sci,
Vol
.
9
,
No.
3
,
Ma
rc
h
201
8
:
667
–
672
672
REFERE
NCE
S
[1]
Brindha
J.
,
Jijayakum
ar
V.
(201
7).
Underwater
I
m
age
Enha
n
ce
m
ent
using
Histog
ra
m
Method.
In
donesian
Journal
of
Elec
t
rical
Engi
ne
erin
g
and
Computer
Sci
en
ce
,
8(3)
,
34
1
-
348.
IAES.
[2]
Djamal,
E
.
C.
,
Darm
awa
ti
,
R.
,
&
Ramdlan,
S.
N.
(2013).
Applic
a
ti
on
ima
ge
proc
essing
t
o
pre
dic
t
per
sonalit
y
bas
e
d
on
struct
ure
of
handwri
ti
ng
an
d
signat
ure
.
201
3
In
te
rnational
Confe
renc
e
on
Computer
,
Control,
In
formatic
s and
Its
App
li
cations
(IC3IN
A)
.
IE
EE
.
[3]
Fati
m
a
Z.
S.
,
Abdellah
M.,
Mo
hamed
K.
(2017).
Emotion
Re
co
gnit
ion
from
Faci
a
l
Expre
ss
ion
base
d
on
Fiduci
al
Points
Dete
c
ti
on
and
u
sing
Neura
l
Net
work.
Indone
s
ia
n
Journal
of
E
l
ec
tri
cal
Eng
inee
ring
and
Computer
Scien
ce
,
8(1), IAES.
[4]
Florea
,
L
.
,
Flore
a,
C
.
,
Vranc
ea
n
u,
R.
,
&
Ver
ta
n
,
C.
(2013)
.
C
an
Your
E
y
es
T
ell
Me
How
You
T
hink
?
A
Gaz
e
Dire
cted
Esti
m
at
ion
of
th
e
Mental
Acti
v
ity
.
Proce
dings
of
the
Britis
h
Mac
hi
ne
Vision
Co
nfe
re
nc
e
2013
.
[5]
Fu,
Y.
J.
,
L
i,
J.
W
.
,
&
Xiang
,
R.
X.
(2010).
Robust
e
y
e
loca
li
z
at
ion
on
m
ulti
-
vie
w
f
ace
in
complex
bac
kground
b
ase
d
on
SV
M
al
gor
it
hm
.
2010
2nd
I
nte
rnational
Sy
mpos
ium
on
Info
rm
ati
on
Engi
n
eer
ing
and
El
e
ct
ronic
Comm
erc
e
,
IEE
C
20
10
,
39
–
43
.
IE
EE.
[6]
Huang,
Y.,
Dong,
X.,
&
Hao,
M.
(2011).
E
y
e
ga
ze
calibr
a
ti
on
ba
sed
on
support
v
ec
tor
re
gr
ession
m
ac
hine.
2011
9th
World
Congress
on
Int
el
li
g
ent Cont
rol
and
Aut
omat
ion
.
[7]
Hui
y
an
,
X.
,
&
Ji
ngfa
ng
,
W
.
(201
1).
Real
-
ti
m
e
iris
loc
al
i
zation
m
et
hod.
2011
In
ter
nati
onal
Confe
r
enc
e
on
Image
Analysis
and
Signal P
roc
essing
.
134
–
155
.
IEEE.
[8]
Jung,
Y.,
Kim
,
D.,
&
Kim
,
J.
(2
014).
E
y
e
detec
t
ion
for
e
y
e
gla
ss
wea
re
rs
in
iri
s
re
cogni
ti
on
.
Proc
e
edi
ngs
of
the
In
te
rn
at
ion
al S
y
m
posium
on
Consum
er
El
ectr
onic
s,
ISCE
,
(1)
,
3
–
4.
IEEE.
[9]
Orave
c
,
M.
(20
14).
Fea
ture
ex
tra
c
ti
on
and
c
las
sific
at
ion
b
y
m
ac
hine
le
arn
in
g
m
et
hods
for
biometri
c
re
cognition
of
fa
ce
and iri
s.
Proc
ee
dings
EL
MA
R
-
2014
.
79
–
83
.
I
E
EE
.
[10]
Rao,
P.
S.
(
2012
).
E
y
e
De
tecti
on
Us
ing
Gabor
Filt
er
and
SV
M,
(2), 1
–
4.
IEEE.
[11]
Ren,
X.,
Peng,
Z.
,
Z
eng,
Q.,
Pe
ng,
C.
,
Zha
ng
,
J.,
W
u,
S.,
&
Zeng,
Y.
(2008).
An
improved
me
thod
for
Daugm
an'
s ir
is
l
oca
l
iz
a
ti
on
al
gor
it
hm
.
Com
pute
rs
in
B
iol
og
y
and Medic
in
e,
38
(1), 111
-
115.
[12]
Sala
m
,
H.,
Stoib
er,
N.
,
&
Seguier,
R.
(2012)
.
A
m
ult
i
-
te
xtur
e
ap
proa
ch
for
esti
m
at
ing
iri
s
positi
o
ns
in
the
e
y
e
using
2.
5D
Acti
ve
Appea
ra
nce
Models.
2012
19th
IEE
E
Inte
rnational
Confe
renc
e
o
n
Image
Proce
ss
ing
.
[13]
Sarka
r,
A.
R.
,
San
y
a
l,
G.
,
&
Majumder,
S.
(
2015).
A
h
y
bri
d
appr
oac
h
for
e
y
e
-
c
ent
r
e
local
iz
a
ti
on
for
esti
m
at
ion
of
e
ye
-
gazes
using
low
-
cost
web
ca
m
.
2015
IEE
E
Inte
rnational
Co
nfe
renc
e
on
Re
search
in
Computati
onal I
nte
lligen
ce and Com
municat
ion Net
works
(ICRC
ICN)
.
[14]
Scott
i,
F.,
&
Pi
uri,
V.
(2010).
Adapti
ve
Reflec
ti
on
Det
ection
a
nd
Locat
ion
in
Iris
Biom
et
ri
c
I
m
age
s
b
y
Us
ing
Com
puta
ti
onal
In
te
l
li
g
en
ce
T
ec
hniqu
es.
IEE
E
Tr
ans.
Instrum
.
Mea
s.
IEE
E
Tr
ansacti
ons
on
Instrum
ent
ati
on
and
Me
asur
eme
nt
,
59
(7)
,
1825
-
1
833.
[15]
Shi,
J.
X.,
Gu,
X
.
F.,
L
i,
J.
P.
,
L
i
n,
J.,
L
iu,
L.,
&
Huang,
Y.
Y.
(2
009).
A
new
m
et
hod
of
iri
s
image
locati
on
re
sea
rc
h
.
2009
Inte
rnational
Co
nfe
renc
e
on
Appe
rce
i
vi
ng
Computing
and
Inte
lli
genc
e
Ana
ly
sis
,
ICACI
A
2009
,
329
–
332
.
[16]
Sudhakar
K.,
Nith
y
an
anda
m
P.,
(2017).
An
Ac
cur
ate
Fac
ia
l
C
om
ponent
Detec
ti
on
Us
ing
Gab
or
Filt
er
.
Indone
sian J
our
nal
of
Elec
tric
al
Engi
ne
ering
and
Computer
Sc
ie
n
ce
,
6(3), 287
-
29
4.
IAES.
[17]
V.,
&
Rao,
P.
S.
(2012).
E
y
e
dete
ct
ion
using
Gabor
Filt
er
and
SVM.
2012
12
th
Int
ernati
onal
Confer
enc
e
on
Inte
lligen
t
Syst
e
ms
Design
and
A
ppli
cations
(ISD
A)
.
[18]
Vrânc
ea
nu
,
R.
,
Vert
an,
C
.
,
Con
dorovic
i
,
R.
,
Florea
,
L.,
&
Flore
a,
C.
(2011)
.
A
fa
st
m
et
hod
for
det
e
ct
ing
e
y
e
acce
ss
ing
c
ues
used
in
Neu
ro
-
Li
nguisti
c
Pr
ogra
m
m
ing.
2011
IEEE
7th
Int
e
rn
ati
onal
Conf
e
renc
e
on
Inte
lligen
t
Comp
ute
r Comm
unic
a
ti
on
and
Proc
essing
.
[19]
Vrânc
ea
nu
,
R.
,
Florea
,
C
.
,
Flor
e
a,
L
.
,
&
Vert
an,
C.
(2013).
Auto
m
at
ic
de
te
c
ti
on
of
gaz
e
d
irect
ion
for
NLP
appl
i
ca
t
ions.
In
t
ernati
onal
Symp
osium on
Signals
,
Circuits and
S
yste
ms
ISS
CS
20
13
.
[20]
Vrânc
ea
nu
,
R.
,
Florea
,
C
.
,
Flor
ea
,
L
.
,
&
Vert
a
n,
C.
(2015).
Gaz
e
dir
ection
es
ti
m
at
ion
b
y
co
m
ponent
sepa
ra
t
ion
for
r
e
cogni
ti
on
of
E
ye
Acc
essing
Cues.
Mac
h
ine
Vis
ion
and
Appli
cations
,
26
(2
–
3)
,
267
–
278.
IEE
E
.
Evaluation Warning : The document was created with Spire.PDF for Python.