I
n
d
o
n
e
s
i
a
n
J
o
u
r
n
a
l
o
f
E
l
e
c
tr
i
c
a
l
E
n
g
i
n
e
e
r
i
n
g
a
n
d
C
o
m
p
u
te
r
S
c
i
e
n
c
e
V
ol
.
8,
N
o.
1,
O
c
t
ob
er
20
17
,
pp
.
154
~
158
D
O
I
:
10.
115
91/
i
j
eec
s
.
v
8
.i
1
.
pp
15
4
-
15
8
1
54
R
ec
ei
v
ed
J
une
1
0
,
20
1
7
;
R
ev
i
s
ed
A
ug
us
t
2
3
,
20
1
7
;
A
c
c
ept
ed
S
e
pt
e
mber
1
1
,
201
7
The
A
p
pl
i
c
a
ti
on
of M
o
di
fi
e
d
L
e
a
s
t Tr
i
mme
d S
q
u
a
r
e
s
w
i
th G
e
ne
ti
c
A
l
gor
i
t
hms
M
e
t
ho
d
in
F
ace Re
co
g
n
it
io
n
N
u
r
A
z
i
m
a
h
A
b
d
u
l
R
a
h
i
m
1
,
N
o
r
A
z
u
r
a
M
d
.
G
h
a
n
i
*
2
,
N
o
r
az
an
M
o
h
am
ed
3
,
H
i
s
h
a
m
u
d
d
i
n
H
ash
i
m
4
,
Is
m
a
il M
u
s
ir
in
5
1
,2
,3
C
ent
er
f
or
S
t
at
i
s
t
i
c
al
S
t
ud
i
e
s
an
d D
ec
i
s
i
on S
c
i
e
nc
e
s
,
F
ac
ul
t
y
of
C
o
m
put
er
an
d M
at
hem
at
i
c
al
S
c
i
en
c
e
s
,
U
ni
v
er
s
i
t
i
T
ek
no
l
og
i
M
A
R
A
,
S
el
ang
or
D
ar
u
l
E
h
s
an,
M
al
ay
s
i
a
5
C
ent
er
f
or
E
l
e
c
t
r
i
c
a
l
P
ow
er
E
ngi
ne
er
i
n
g S
t
u
di
e
s
,
F
a
c
u
l
t
y
of
E
l
ec
t
r
i
c
al
E
n
gi
n
ee
r
i
ng
,
U
ni
v
er
s
i
t
i
T
ek
no
l
og
i
M
A
R
A
,
S
el
ang
or
D
ar
u
l
E
h
s
an,
M
al
ay
s
i
a
4
P
i
l
i
h
s
e
pak
at
S
dn
.
B
hd.
,
Lo
t
5
431,
T
i
ng
k
at
1(
B
)
,
J
al
a
n J
9,
F
as
a
6,
T
am
an
M
el
aw
at
i
,
K
ual
a
Lum
pur
,
M
al
ay
s
i
a
*
C
or
r
es
po
ndi
ng a
ut
hor
,
e
-
m
a
i
l:
nur
az
i
m
ah1
809@
g
m
ai
l
.
c
om
1
,
az
ur
a@
t
m
s
k
.
ui
t
m
.
ed
u.
m
y
2
,
nor
az
an@
t
m
s
k
.
ui
t
m
.
e
du.
m
y
3
,
ps
s
bf
6@
g
m
ai
l
.
c
om
4
,
is
m
a
ilb
m
@
s
a
la
m
.
u
it
m
.
e
d
u
.
m
y
5
A
b
st
r
act
Se
v
e
re
l
y
o
c
c
l
ude
d f
ac
e i
m
ag
es
ar
e t
h
e m
ai
n pr
ob
l
em
i
n l
ow
per
f
or
m
anc
e of
f
a
c
e r
ec
o
gni
t
i
on
al
gor
i
t
hm
s
.
I
n
t
hi
s
p
aper
,
w
e
a
ppl
y
a
new
al
g
or
i
t
hm
,
a m
od
i
f
i
ed v
er
s
i
on
of
t
he
l
e
as
t
t
r
i
m
m
ed s
q
uar
e
s
(
LT
S
)
w
i
t
h a
gene
t
i
c
al
gor
i
t
hm
s
i
nt
r
oduc
e b
y
[1
]
.
W
e
f
oc
us
ed o
n
t
he
ap
p
li
c
at
i
on
o
f
m
o
di
f
ie
d
LT
S
w
i
t
h g
enet
i
c
al
gor
i
t
hm
m
et
hod f
or
f
ac
e i
m
a
ge r
ec
o
gni
t
i
on
.
T
hi
s
al
gor
i
t
hm
us
es
ge
net
i
c
al
g
or
i
t
hm
s
t
o
c
ons
t
r
uc
t
a ba
s
i
c
s
ub
s
et
r
at
her
t
ha
n s
el
e
c
t
i
ng
t
he
b
as
i
c
s
ub
s
et
r
and
om
l
y
.
T
he
m
odi
f
i
c
at
i
on
i
n t
h
i
s
m
et
hod
le
s
s
e
ns
t
h
e
num
ber
of
t
r
i
al
s
t
o obt
ai
n t
h
e
m
i
ni
m
um
of
t
he LT
S
ob
j
e
c
t
i
v
e
f
unc
t
i
on.
T
hi
s
m
et
hod w
a
s
t
he
n appl
i
ed t
o t
w
o
benc
hm
ar
k
dat
as
e
t
s
w
i
th
c
l
e
an and
oc
c
l
ude
d quer
y
i
m
a
ges
.
T
he
per
f
or
m
anc
e
of
t
hi
s
m
et
hod
w
as
m
eas
ur
ed
b
y
r
ec
ogn
i
t
i
o
n
r
at
e
s
.
T
he
A
T
&
T
dat
a
s
e
t
a
nd
Y
al
e
D
at
as
e
t
w
i
t
h
di
f
f
er
ent
i
m
age
p
i
x
el
s
i
z
es
w
er
e
us
ed t
o as
s
es
s
t
h
e m
et
hod
i
n per
f
or
m
i
ng f
ac
e r
ec
o
gni
t
i
on
.
T
he quer
y
i
m
age
s
w
er
e c
ont
am
i
nat
e
d w
i
t
h s
al
t
and pe
pper
n
oi
s
e
. T
he
m
odi
f
i
ed LT
S
w
i
t
h G
A
s
m
et
hod i
s
a
ppl
i
ed i
n f
ac
e r
ec
o
gni
t
i
on f
r
a
m
ew
or
k
by
u
s
i
ng
t
he
c
ont
am
i
nat
e
d
i
m
ag
es
as
q
uer
y
i
m
age
i
n t
he
c
on
t
ex
t
of
l
i
near
r
e
gr
e
s
s
i
on.
B
y
t
he e
nd
o
f
t
hi
s
s
t
ud
y
,
w
e
c
an
det
er
m
i
n
e t
h
i
s
ei
t
h
er
t
h
i
s
m
et
hod c
an p
er
f
or
m
w
el
l
i
n
de
al
i
n
g
w
i
t
h oc
c
l
u
ded i
m
age
s
or
v
i
c
e
v
er
s
a
.
Ke
y
w
o
rd
:
f
ac
e r
e
c
ogn
i
t
i
on,
g
e
net
i
c
al
gor
i
t
hm
,
l
ea
s
t
t
r
i
m
m
ed
s
quar
es
,
LT
S
w
i
t
h G
A
s
,
noi
s
e
i
m
ages
C
o
p
y
r
i
g
h
t
©
2
01
7
I
n
s
t
i
t
u
t
e
o
f
A
d
v
a
n
c
e
d
E
n
g
i
n
e
e
r
i
n
g
a
n
d
S
c
i
e
n
c
e
.
A
l
l
r
i
g
h
t
s r
es
er
ved
.
1
.
I
n
tr
o
d
u
c
ti
o
n
F
ac
e
r
ec
ogn
i
t
i
on
m
a
y
b
e
a
s
t
an
dou
t
am
ongs
t
t
ho
s
e
t
he
gr
eat
er
par
t
m
om
ent
o
us
abi
l
i
t
i
es
t
hat
h
um
an
m
i
ght
do.
T
hi
s
i
s
l
i
k
ew
i
s
e
a
pr
i
nc
i
pl
e
f
i
e
l
d
ab
out
ex
am
i
nat
i
o
n
f
or
bi
om
et
r
i
c
i
nd
i
c
at
or
t
r
ans
f
or
m
i
ng.
I
n y
e
ar
s
,
i
t
add
i
t
i
ons
an
ev
e
r
-
i
nc
r
eas
i
ng
am
ount
c
ons
i
der
at
i
on f
r
o
m
c
la
im
in
g m
ar
k
et
er
s
.
A
f
ew
i
nv
es
t
i
ga
t
i
o
ns
w
er
e l
e
d f
or
f
ac
e
r
ec
ogni
t
i
o
n
[2
-
5]
.
T
her
e ex
i
s
t
di
f
f
er
ent
f
r
a
m
ew
or
k
s
and al
g
or
i
t
h
m
s
f
or
a f
ac
e r
ec
ogni
t
i
on
f
r
am
ew
or
k
.
D
es
pi
t
e t
he i
m
pr
es
s
i
v
e
ac
hi
e
v
em
ent
s
t
hat
ha
v
e
b
een m
ade,
i
t
i
s
s
t
i
l
l
op
en
under
u
nc
ons
t
r
ai
ned
en
v
i
r
onm
ent
s
,
w
i
t
h
v
ar
i
at
i
ons
of
i
l
l
um
i
nat
i
o
n,
e
x
pr
es
s
i
on,
he
ad p
os
e,
as
w
el
l
as
par
t
i
al
oc
c
l
us
i
ons
.
F
a
c
e r
ec
ogni
t
i
o
n
pr
oc
es
s
w
i
l
l
be
a
w
h
ol
e
l
ot
i
nf
l
uenc
ed
b
y
out
l
i
er
s
.
O
ut
l
i
er
s
i
n
f
ac
e
r
ec
ogni
t
i
o
n
i
s
al
s
o
k
now
n
as
noi
s
e or
oc
c
l
us
i
o
n.
N
o
i
s
e
i
s
a r
andom
v
ar
i
at
i
on
of
i
m
age i
n
t
ens
i
t
y
.
A
n
i
m
age s
houl
d
s
ho
w
t
he
t
r
ue pi
x
e
l
s
v
a
l
u
e.
B
ut
,
w
h
e
n an i
m
age do
es
not
,
t
hi
s
i
nd
i
c
at
es
t
h
at
t
h
e i
m
age h
av
e
noi
s
e
[
6]
.
T
hi
s
s
t
ud
y
w
as
don
e
t
o
ex
pl
or
e
t
he
app
l
i
c
at
i
on
of
m
o
di
f
i
ed
l
e
as
t
t
r
i
m
m
ed
s
quar
es
w
i
t
h
g
en
et
i
c
al
g
or
i
t
hm
s
(
LT
S
w
i
t
h
G
A
s
)
m
et
hod
i
n
f
a
c
e
r
ec
og
ni
t
i
o
n.
T
hi
s
pap
er
i
s
ar
r
an
ged
as
f
ol
l
o
w
s
.
T
he
f
ac
e r
ec
ogni
t
i
on
us
i
n
g LT
S
w
i
t
h G
A
s
m
et
hod
i
s
pr
es
ent
e
d i
n S
ec
t
i
o
n 2.
A
l
s
o,
ex
am
i
nat
i
on
r
es
ul
t
s
al
on
gs
i
de
ar
ea
3.
T
he p
aper
at
l
ong
l
as
t
f
i
n
i
s
he
s
up i
n
s
egm
ent
4.
2.
M
o
d
i
fi
e
d
L
T
S
w
i
th
G
e
n
e
ti
c
A
l
g
o
r
i
t
h
m
(L
T
S
-
G
As
)
T
hi
s
m
et
hod w
as
i
n
t
r
oduc
e b
y
[
1]
. T
he LT
S
-
G
A
m
et
hod
w
as
r
ef
i
ne
d
t
o enh
a
nc
e i
t
s
per
f
or
m
anc
e
w
he
n t
he
dat
a
s
et
c
ont
ai
ns
out
l
i
er
s
b
y
us
i
ng
t
he
bes
t
50%
of
obs
er
v
a
t
i
ons
t
o
obt
ai
n
i
ni
t
i
a
l
p
ar
am
et
er
es
t
i
m
at
es
f
or
get
t
i
ng
in
it
ia
l r
e
s
id
u
a
ls
.
Evaluation Warning : The document was created with Spire.PDF for Python.
I
J
E
EC
S
IS
S
N
:
2
502
-
4
752
T
he
A
p
p
lic
a
t
io
n
of
Mod
i
f
i
ed
L
eas
t
T
r
i
mm
ed
S
qu
ar
es
…
(
N
ur
A
z
i
m
ah
A
b
dul
R
ah
i
m
)
155
=
2
(1
)
w
her
e
n
i
s
t
h
e
num
ber
o
f
t
ot
al
o
bs
er
v
a
t
i
o
ns
i
n
da
t
a.
T
he
c
ont
am
i
nat
e
d
o
bs
er
v
at
i
ons
ar
e
det
ec
t
e
d
i
n
C
-
s
t
eps
.
A
f
t
er
w
ar
d
t
hos
e
r
ec
ogn
i
z
ed
5
0%
c
ont
am
i
nat
ed
o
bs
er
v
at
i
ons
ar
e
t
r
i
m
m
ed.
E
ac
h
o
bs
er
v
a
t
i
o
n
w
i
l
l
be
al
l
oc
at
ed
w
it
h
a
w
e
i
gh
t
bas
e
d
on
a
c
ut
-
of
f
v
al
u
e
w
h
i
c
h
w
i
l
l
gi
v
e
a
z
er
o
w
ei
g
ht
f
or
an
y
obs
er
v
at
i
on
w
i
t
h r
es
i
dua
l
er
r
or
gr
ea
t
er
t
han t
he c
ut
-
of
f
v
al
ue
and
a w
ei
ght
“
1”
ot
her
w
i
s
e.
T
hes
e w
e
i
ght
s
w
i
l
l
ens
ur
e t
he r
egr
es
s
i
on c
oef
f
i
c
i
ent
es
t
i
m
at
es
ar
e not
af
f
ec
t
ed b
y
t
he
pr
es
enc
e
of
out
l
i
er
s
as
t
he
obs
er
v
a
t
i
ons
w
i
t
h
a h
i
g
her
l
ev
e
l
of
noi
s
e
w
i
l
l
r
ec
ei
v
e
w
ei
g
ht
s
c
l
os
e
t
o
z
er
o.
A
r
o
bus
t
s
t
an
dar
d
de
v
i
at
i
on
w
as
us
ed
and c
om
p
ut
ed
as
f
ol
l
o
w
s
.
=
me
dia
n
−
0
.
6
4
7
5
⁄
(2
)
T
he w
e
i
g
ht
f
or
eac
h r
es
i
du
al
i
s
f
or
m
ul
at
ed
as
f
ol
l
o
w
s
:
=
1
,
if
a
bs
(
)
≤
2
×
0
,
oth
e
r
w
i
s
e
(
3)
T
he r
ef
i
nem
ent
i
n t
h
e ex
i
s
t
i
ng LT
S
-
G
A
i
s
des
c
r
i
be
d as
f
ol
l
o
w
s
:
1
I
nput
:
M
at
r
i
x
es
o
f
t
r
ai
n
i
n
g
s
a
m
p
l
e s
et
=
(
1
)
,
(
2
)
,
…
,
(
)
f
o
r
p
c
l
a
sse
s
a
n
d
∈
ℝ
×
M
at
r
i
x
es
f
o
r
te
s
t s
a
m
p
le
s
e
t
=
(
1
)
,
(
2
)
,
…
,
(
−
)
∈
ℝ
×
(
−
)
2
fo
r
each
s
u
b
j
ect
i
d
o
3
f
i
t
ne
s
s
f
u
nc
t
i
o
n (
)
4
f
u
n
c
tio
n
f
i
tn
e
s
s
(
c
a
n
d
id
a
te
s
)
{
5
n
e
w
b
et
as
:
= cs
t
ep
(
can
d
i
d
at
es
)
6
r
es
i
d
u
al
s
:
= cal
c
u
l
at
e r
es
i
d
u
al
s
u
s
i
n
g
n
e
w
b
et
as
7
appl
y
e
q
uat
i
on
(
2)
and (
3
)
t
o c
om
put
e
r
obus
t
s
t
and
ar
d e
r
r
or
and r
e
s
i
dual
s
8
f
i
t
n
es
s
:
= cal
cu
l
a
t
e L
T
S
cr
i
t
er
i
o
n
u
s
i
n
g
r
es
i
d
u
al
s
.
9
ret
u
rn
(
f
itn
e
s
s
)
;}
10
e
nd
11
O
ut
put
:
id
e
n
tif
y
(
y
)
=
a
r
g
m
i
n
d
(
y
,
i)
.
T
he ps
eudoc
o
de f
or
t
he c
s
t
eps
ar
e
as
be
l
o
w
:
c
-
s
te
p
()
1
f
u
n
ct
i
o
n
cs
t
ep
(
can
d
i
d
at
es
)
{
2
r
es
:
= cal
cu
l
at
e r
es
i
d
u
al
s
u
s
i
n
g
can
d
i
d
at
es
3
s
e
t
i
ndi
c
e
s
:
=
I
ndi
c
e
s
of
p
obs
e
r
v
at
i
ons
f
r
om
e
quat
i
on
(1
)
4
f
o
r
i
=
1 t
o n
um
be
r
of
c
-
st
e
p
s
{
5
o
ls
:
= C
al
cu
l
at
e O
L
S
u
s
i
n
g
t
h
e
su
b
s
e
t
(
i
n
d
i
c
e
s)
;
6
r
e
s
id
u
a
ls
: =
G
e
t r
e
s
id
u
a
ls
f
r
o
m
(o
l
s
);
7
o
rd
er
i
n
g
o
f
r
es
i
d
u
al
s
:
= O
r
d
er
(a
b
s
(re
s
i
d
u
a
l
s
));
8
i
nd
ic
e
s
: =
F
ir
s
t
h
e
le
m
e
n
ts
o
f
or
de
r
i
n
g
of
r
e
s
i
du
a
l
s
;
9
}
e
nd
10
s
et
b
et
as
:
= C
o
ef
f
i
c
i
en
t
s
o
f
O
L
S
;
11
re
t
u
rn
(b
e
t
a
s
);
}
12
e
va
l
ua
t
e
f
i
t
ne
s
s
v
al
u
e
s
an
d
p
er
f
o
r
m
s
el
ec
t
i
o
n
,
c
r
o
s
s
i
ng o
ve
r
,
a
nd
m
ut
a
t
i
o
n o
p
e
r
a
t
i
o
ns
o
n t
h
e
c
h
r
o
mo
s
o
me
s
u
n
t
i
l
ma
x
i
m
u
m
n
um
be
r
of
i
t
e
r
a
t
i
ons
r
each
ed
.
G
et
t
h
e b
es
t
c
h
r
o
mo
s
o
me
a
n
d
p
e
rf
o
rm
C
-
st
e
p
s
u
s
i
n
g
b
e
s
t
c
h
r
o
mo
s
o
me
.
3.
R
e
su
l
t
s
a
n
d
A
n
a
l
y
s
i
s
F
or
t
hi
s
s
t
ud
y
,
t
he
A
T
&
T
da
t
as
et
an
d
Y
a
l
e d
at
as
et
w
er
e us
ed t
o a
ppl
y
t
h
e m
odi
f
i
ed LT
S
w
i
t
h
G
A
s
m
et
hod
i
n
f
ac
e
r
ec
ogn
i
t
i
on
an
al
y
s
i
s
.
I
m
ages
f
r
o
m
bot
h
dat
a
s
et
s
ha
v
e
di
f
f
er
ent
pi
x
e
l
s
i
z
e.
T
w
o
di
f
f
er
ent
l
e
v
e
l
s
of
i
m
age pi
x
e
l
s
;
s
i
z
e 7
8x
64
pi
x
el
s
and
s
i
z
e 1
12x
92
pi
x
e
l
s
w
er
e c
r
eat
ed
f
or
t
he
A
T
&
T
dat
ab
as
e.
A
s
f
or
t
he
i
m
ages
f
r
om
t
he
Y
al
e
dat
abas
e
w
er
e
do
w
n
s
am
pl
ed
t
o
t
he
Evaluation Warning : The document was created with Spire.PDF for Python.
I
SSN
:
25
02
-
4
752
I
J
E
EC
S
V
o
l.
8
,
N
o.
1,
O
c
t
o
ber
20
17
:
1
54
–
1
58
156
s
i
z
e
of
32x
32 p
i
x
el
s
.
I
m
ages
f
r
o
m
al
l
dat
as
et
s
w
er
e
di
v
i
d
ed
i
nt
o
t
w
o p
ar
t
s
eac
h.
H
a
l
f
i
m
ages
f
r
o
m
eac
h s
ubj
ec
t
of
eac
h dat
as
et
s
w
er
e us
ed as
q
uer
y
i
m
ages
w
h
i
l
e t
he l
at
t
er
hal
f
w
er
e f
or
t
es
t
i
n
g
i
m
ages
[
7]
.
S
al
t
an
d
P
e
pp
er
no
i
s
e
w
er
e
ad
de
d
ar
t
i
f
i
c
i
al
l
y
t
o
t
he
quer
y
i
m
ages
w
i
t
h
f
i
v
e
di
f
f
er
ent
l
ev
el
s
;
10%
,
20%
,
30%
,
40%
a
nd 50
%.
I
m
ages
i
n quer
y
i
m
age s
et
c
anno
t
be t
he s
am
e
i
m
ages
i
n
t
he
t
r
ai
n
i
ng
i
m
ages
s
et
.
A
n
i
m
age
f
r
o
m
quer
y
i
m
age
s
et
i
s
r
ul
ed
t
o
be
m
at
c
h
w
i
t
h
t
h
e
t
r
ai
n
i
ng
i
m
ages
s
et
w
he
n
t
he
di
f
f
er
enc
es
bet
w
een
t
h
e
qu
er
y
i
m
age
f
r
o
m
quer
y
i
m
ages
s
et
has
t
he m
i
ni
m
u
m
di
s
t
anc
e w
he
n c
om
par
ed w
i
t
h al
l
i
m
ages
i
n t
he t
r
ai
ni
n
g i
m
ages
s
et
.
T
he
r
ec
ogni
t
i
o
n r
at
e
us
ed
i
n f
ac
e r
ec
ogn
i
t
i
on s
t
u
d
y
her
e r
epr
es
ent
s
t
he p
er
c
ent
a
ge of
t
he t
ot
a
l
num
ber
of
c
or
r
ec
t
l
y
m
at
c
he
d i
m
ages
bet
w
een
t
he
t
w
o
s
et
s
.
3.
1.
A
T
&
T
D
at
ab
a
se
T
abl
e 1 an
d T
abl
e 2 g
i
v
es
f
ac
e r
ec
ogn
i
t
i
on r
at
es
of
A
T
&
T
dat
a
s
et
wi
t
h
s
i
z
e 78x
6
4
pi
x
el
s
and s
i
z
e
11
2x
92
pi
x
e
l
s
.
F
i
g
ur
e
1
is
t
h
e
ex
am
pl
es
of
i
m
ages
f
r
om
A
T
&
T
D
at
a
s
et
w
hi
l
e F
i
gur
e
2 i
s
t
he
ex
am
pl
e
of
i
m
ages
f
r
om
A
T
&
T
dat
as
et
w
it
h
v
ar
i
o
us
l
e
v
e
l
s
of
S
al
t
an
d
P
ep
p
er
N
o
i
s
e.
F
ro
m
bot
h
T
ab
l
e
1
an
d
T
abl
e
2,
w
hen
t
he
per
c
en
t
ag
e
of
no
i
s
e
ge
t
t
i
ng
hi
g
her
,
i
t
c
a
n
b
e
s
een
c
l
ear
l
y
t
hat
t
h
e
per
c
ent
age
of
r
ec
ogni
t
i
o
n
r
at
e
dr
o
ps
.
W
hen
w
e
c
om
par
e
t
he
per
c
ent
ag
e
of
r
ec
ogni
t
i
on
r
at
e
bet
w
ee
n
T
abl
e
1
an
d
T
abl
e
2,
t
h
e
i
m
ages
w
i
t
h
s
i
z
e
7
8x
64
p
i
x
el
s
f
r
om
T
abl
e
1
ga
v
e
h
i
g
her
r
ec
ogni
t
i
o
n r
at
e
f
or
oc
c
l
ude
d i
m
ages
.
F
i
gur
e
1
.
E
x
am
pl
e
s
of
i
m
age
s
f
r
om
A
T
&
T
D
at
as
et
(a
)
10%
(b
)
20%
(c
)
30%
(d
)
40%
(e
)
50%
F
i
gur
e
2
.
E
x
am
pl
e
s
of
i
m
age
s
f
r
om
A
T
&
T
D
at
a
s
et
c
on
t
am
i
na
t
ed w
i
t
h di
f
f
er
en
t
l
ev
e
l
s
o
f
S
al
t
and P
e
pper
No
i
s
e
T
abl
e 1
.
F
ac
e R
ec
o
gni
t
i
on
R
at
es
f
or
A
T
&
T
D
at
abas
e I
m
ages
w
i
t
h
s
i
z
e 7
8x
64
pi
x
e
l
s
No
i
s
e
(%
)
R
ec
ogni
t
i
on R
a
t
e
U
s
i
ng M
et
hod
M
O
D
I
FI
E
D
L
TS
W
i
t
h G
as
(
%
)
0
86.
50
10
84.
50
20
77.
50
30
71.
00
40
54.
50
50
45.
00
T
abl
e 2
.
F
ac
e R
ec
o
gni
t
i
on
R
at
es
f
or
A
T
&
T
D
at
abas
e I
m
ages
w
i
t
h
s
i
ze
1
12x
92
p
ix
e
ls
No
i
s
e
(%
)
R
ec
ogni
t
i
on R
a
t
e
U
s
i
ng M
et
hod
M
O
D
I
FI
E
D
L
TS
W
i
t
h G
as
(
%
)
0
88.
00
10
81.
50
20
77.
50
30
69.
00
40
50.
50
50
34.
50
Evaluation Warning : The document was created with Spire.PDF for Python.
I
J
E
EC
S
IS
S
N
:
2
502
-
4
752
T
he
A
p
p
lic
a
t
io
n
of
Mod
i
f
i
ed
L
eas
t
T
r
i
mm
ed
S
qu
ar
es
…
(
N
ur
A
z
i
m
ah
A
b
dul
R
ah
i
m
)
157
3.
2.
Y
al
e D
at
ab
ase
T
abl
e 3
d
i
sp
l
a
y
s
r
ec
o
gn
i
t
i
on r
at
es
f
or
Y
a
l
e
dat
a s
e
t
i
m
ages
.
W
e c
an s
ee t
h
at
t
he
r
ec
ogni
t
i
o
n
r
at
e
i
s
hi
gh
w
hen t
he t
h
e d
at
a
i
s
c
l
ea
n.
E
v
ent
u
al
l
y
,
t
h
e r
ec
ogn
i
t
i
o
n r
at
e
dr
o
pp
ed
w
hen
t
h
e
l
e
v
el
of
no
i
s
e
i
nc
r
eas
es
t
i
l
l
t
h
e
i
m
age
w
as
c
o
nt
am
i
nat
ed
at
30
%
l
e
v
el
of
noi
s
e.
S
t
ar
t
i
ng
f
r
om
t
he l
ev
el
of
40%
of
noi
s
e i
n t
es
t
i
m
ages
,
w
e
c
an s
ee t
hat
t
h
e
r
ec
ogni
t
i
on r
at
e
of
t
h
is
m
et
hod w
as
i
m
pr
ov
ed.
T
he r
ec
og
n
i
t
i
on r
at
e
w
h
en t
he
noi
s
e
i
s
at
50%
i
s
h
i
gh
er
t
h
an ot
her
n
o
is
e
per
c
ent
a
ge l
e
v
e
l
.
F
i
gur
e 3
.
E
x
am
pl
e of
i
m
ages
f
r
o
m
Y
al
e
dat
as
et
F
i
gur
e 4
.
E
x
am
pl
e of
i
m
ages
f
r
o
m
Y
al
e
dat
as
et
w
i
t
h
di
f
f
er
ent
l
e
v
e
l
of
S
a
l
t
&
P
ep
pe
r
N
oi
s
e
T
abl
e 3
.
F
ac
e R
ec
o
gni
t
i
on
R
at
es
f
or
Y
a
l
e D
at
a
bas
e I
m
ages
No
i
s
e
(%
)
R
ec
ogni
t
i
on R
a
t
e
U
s
i
ng M
et
hod
M
odi
f
i
ed Lt
s
W
i
t
h G
a
s
(
%
)
0
66.
67
10
20.
00
20
24.
44
30
15.
56
40
31.
11
50
55.
56
4
.
C
o
n
c
l
u
s
i
o
n
T
h
is
s
t
ud
y
i
s
abo
ut
t
h
e ap
pl
i
c
a
t
i
o
n
of
m
odi
f
i
ed LT
S
w
it
h
G
A
s
i
n f
ac
e r
ec
ogni
t
i
on.
T
he
A
T
&
T
and Y
a
l
e d
at
as
et
s
w
i
t
h t
hr
ee di
f
f
er
ent
i
m
age pi
x
el
s
s
i
z
es
w
er
e us
ed t
o
m
eas
ur
e t
he
r
ec
ogni
t
i
o
n
r
at
e
w
he
n
m
odi
f
i
ed
LT
S
w
i
t
h
G
A
s
m
et
hod
w
as
us
e
i
n
f
ac
e
r
ec
ogn
i
t
i
o
n
an
al
y
s
i
s
.
A
ll
i
m
ages
w
er
e
add
ed w
i
t
h
a
r
t
i
f
i
c
i
al
n
oi
s
e
w
i
t
h
di
f
f
er
ent
l
e
v
el
of
S
al
t
and
P
e
pper
n
oi
s
e.
It
c
a
n
be
c
onc
l
ud
ed t
ha
t
w
hen t
h
e i
m
age dat
a s
et
i
s
c
l
ean,
t
he dat
as
e
t
f
r
o
m
A
T
&
T
i
m
ages
w
i
t
h s
i
z
e
112x
92
pi
x
e
l
s
gi
v
es
bet
t
er
r
ec
ogni
t
i
on r
at
e.
H
o
w
ev
er
,
w
h
en t
h
er
e i
s
no
i
s
e i
n t
h
e i
m
ages
,
t
he
dat
as
et
f
r
om
Y
a
l
e i
m
ages
w
i
t
h s
i
z
e 32x
3
2 pi
x
el
s
ar
e
bet
t
er
.
I
n t
h
e near
f
ut
ur
e,
a s
t
ud
y
ab
out
t
h
e
s
t
r
engt
h
of
t
hi
s
m
odi
f
i
ed
LT
S
w
i
t
h G
A
s
m
et
hod
i
n c
om
par
i
s
on
w
i
t
h t
he
ex
i
s
t
i
ng al
gor
i
t
hm
s
in
t
he
app
l
i
c
at
i
on
of
i
m
ages
r
ec
og
ni
t
i
on
s
h
ou
l
d b
e d
one
.
A
c
k
n
o
w
l
e
d
g
e
m
e
n
t
W
e
w
oul
d
l
i
k
e
t
o
ex
pr
es
s
our
appr
ec
i
at
i
on
t
o
t
he
Mi
ni
s
t
r
y
of
H
i
gher
E
duc
a
t
i
on
(
MO
H
E
)
and
U
n
i
v
er
s
i
t
y
T
ek
nol
og
i
MA
R
A
f
or
f
i
nanc
i
al
s
u
ppor
t
under
t
h
e
G
r
ant
S
c
h
em
e
600
-
R
M
I/F
R
G
S
5/
3 (
57
/
20
14)
a
nd 6
00
-
R
MI
/
D
A
N
A
5/
3/
P
S
I
(
1
97/
201
3)
.
Evaluation Warning : The document was created with Spire.PDF for Python.
I
SSN
:
25
02
-
4
752
I
J
E
EC
S
V
o
l.
8
,
N
o.
1,
O
c
t
o
ber
20
17
:
1
54
–
1
58
158
R
ef
er
en
ces
[1
]
N
A
R
ahi
m
A
b
dul
,
N
M
R
am
l
i
,
N
A
M
d G
hani
.
T
he P
e
r
f
or
m
a
nc
e
of
M
odi
f
i
ed
Lea
s
t
T
r
i
m
m
e
d S
qu
ar
es
-
B
as
ed
M
et
hod
s
f
or
Lar
g
e
D
at
a
S
et
s
B
as
e
d
on
M
ont
e
C
ar
l
o
S
i
m
u
l
at
i
ons
.
A
dv
anc
ed
S
c
i
en
c
e
Le
t
t
er
s
.
2016
:
22(
1
2)
:
4
359
-
4
363
.
[2
]
N
Z
hen
g,
Q
Y
ou,
G
M
eng,
J
Z
hu,
S
Du
,
J
Li
u
.
50 Y
ear
s
of
I
m
age
P
r
o
c
es
s
i
n
g a
nd
P
at
t
er
n
R
ec
ogn
i
t
i
on
i
n
Ch
i
na.
I
E
E
E
I
nt
e
l
l
i
gent
S
y
s
t
em
s
.
200
8
;
2
3
(6
):
33
-
41
.
[3
]
L
Z
hang,
M
Y
ang,
X
F
eng,
Y
M
a,
D
Z
hang
.
C
ol
l
a
bor
at
i
v
e
r
e
pr
es
e
nt
at
i
on
ba
s
e
d
c
l
a
s
s
i
f
i
c
at
i
on
f
or
f
ac
e
r
ec
og
ni
t
i
on
.
ar
X
i
v
pr
e
pr
i
nt
ar
X
i
v
:
12
04.
2
358
.
201
2.
[4
]
M
Y
ang,
L
Z
ha
ng,
S
C
S
hi
u,
D
Z
hang
.
G
abor
f
ea
t
ur
e
ba
s
ed
r
obus
t
r
epr
es
ent
at
i
on
and
c
l
as
s
i
f
i
c
at
i
on
f
o
r
f
ac
e r
ec
o
gni
t
i
on
w
i
t
h G
abor
o
c
c
l
u
s
i
o
n di
c
t
i
onar
y
.
P
at
t
er
n R
e
c
ogni
t
i
on
.
201
3
;
46
(
7
):
1865
-
18
78.
[5
]
N
M
R
N
ur
A
z
i
m
ah A
bdu
l
R
ahi
m
,
N
or
A
z
ur
a M
d.
G
hani
.
T
h
e P
er
f
or
m
anc
e of
LT
S
-
bas
ed R
egr
es
s
i
o
n
M
et
hods
i
n
F
ac
e
R
ec
ogn
i
t
i
o
n
w
i
t
h
O
c
c
l
ud
ed
I
m
ag
es
.
I
nd
i
an J
our
n
al
o
f
S
c
i
en
c
e
a
nd
T
ec
h
n
ol
og
y
,
[
S
.
l
.
]
.
2016
;
9(
28)
.
[6
]
K
K
V
T
o
h
, N
A
M
Is
a
.
N
oi
s
e
A
dapt
i
v
e F
uz
z
y
S
w
i
t
c
hi
ng M
edi
an
F
i
l
t
er
f
or
S
al
t
-
an
d
-
P
e
p
per
N
oi
s
e
R
educ
t
i
on
.
S
i
gn
al
P
r
oc
e
s
s
i
ng
Let
t
er
s
,
I
E
E
E
.
201
0
;
17
(
3
):
281
-
284.
[7
]
D
C
ai
,
X
H
e,
Y
H
u,
J
H
an,
T
H
uang
.
Lear
ni
n
g a
s
pat
i
al
l
y
s
m
o
ot
h
s
ubs
pa
c
e f
or
f
a
c
e r
e
c
og
ni
t
i
on.
1
-
7.
Evaluation Warning : The document was created with Spire.PDF for Python.