I
nd
o
ne
s
ia
n J
o
urna
l o
f
E
lect
rica
l En
g
ineering
a
nd
Co
m
pu
t
er
Science
Vo
l.
23
,
No
.
2
,
A
u
g
u
s
t
20
21
,
p
p
.
8
1
1
~
8
2
0
I
SS
N:
2
5
0
2
-
4
7
5
2
,
DOI
: 1
0
.
1
1
5
9
1
/ijeecs.v
23
.i
2
.
pp
811
-
8
2
0
811
J
o
ur
na
l ho
m
ep
a
g
e
:
h
ttp
:
//ij
ee
cs.ia
esco
r
e.
co
m
Clo
ud
-
ba
sed a
rch
itect
u
re f
o
r
f
a
ce
i
dentif
i
ca
tion wi
th
deep
lea
rning
using
co
nv
o
lutiona
l neural network
Adity
a
H
er
la
m
ba
ng
,
P
utu
Wira
B
ua
na
,
I
Ny
o
m
a
n P
ia
r
s
a
Dep
ar
tm
en
t o
f
I
n
f
o
r
m
atio
n
T
e
ch
n
o
lo
g
y
,
Ud
a
y
an
a
Un
iv
e
r
s
ity
,
I
n
d
o
n
esia
Art
icle
I
nfo
AB
S
T
RAC
T
A
r
ticle
his
to
r
y:
R
ec
eiv
ed
Feb
1
2
,
2
0
2
1
R
ev
is
ed
Ma
y
2
0
,
2
0
2
1
Acc
ep
ted
J
u
n
1
,
2
0
2
1
Th
e
u
se
o
f
a
fa
c
e
a
s
a
b
io
m
e
tri
c
to
id
e
n
ti
fy
a
p
e
rso
n
i
n
o
rd
e
r
t
o
k
e
e
p
t
h
e
sy
ste
m
sa
fe
fro
m
a
n
u
n
a
u
th
o
r
ize
d
p
e
rso
n
h
a
s
a
d
v
a
n
tag
e
s
o
v
e
r
o
t
h
e
r
b
io
m
e
tri
c
c
h
a
ra
c
teristics
.
Th
e
fa
c
e
a
s
a
b
io
m
e
tri
c
h
a
s
m
o
re
stru
c
tu
re
a
n
d
a
wid
e
r
a
re
a
th
a
n
o
th
e
r
b
io
m
e
tri
c
s,
wh
il
e
c
a
n
b
e
re
tri
e
v
e
d
i
n
a
n
o
n
-
in
v
a
siv
e
m
a
n
n
e
r.
We
p
r
o
p
o
se
d
a
c
lo
u
d
-
b
a
se
d
a
rc
h
it
e
c
tu
re
fo
r
fa
c
e
id
e
n
ti
fic
a
ti
o
n
wit
h
d
e
e
p
lea
rn
i
n
g
u
sin
g
c
o
n
v
o
l
u
ti
o
n
a
l
n
e
u
ra
l
n
e
two
rk
.
F
a
c
e
id
e
n
ti
fica
t
io
n
i
n
t
h
is
stu
d
y
u
se
d
a
c
lo
u
d
-
b
a
se
d
e
n
g
i
n
e
with
fo
u
r
sta
g
e
s,
n
a
m
e
ly
fa
c
e
d
e
t
e
c
ti
o
n
wit
h
h
isto
g
ra
m
o
f
o
r
ien
ted
g
ra
d
ien
ts
(HO
G
),
ima
g
e
e
n
h
a
n
c
e
m
e
n
t,
fe
a
tu
re
e
x
trac
ti
o
n
u
si
n
g
c
o
n
v
o
lu
t
io
n
a
l
n
e
u
ra
l
n
e
two
rk
,
a
n
d
c
las
sifica
ti
o
n
u
si
n
g
k
-
n
e
a
re
st
n
e
ig
h
b
o
r
(KN
N),
S
VM
,
a
s
we
ll
a
s
ra
n
d
o
m
fo
re
st
a
lg
o
ri
th
m
.
Th
is
stu
d
y
c
o
n
d
u
c
ted
a
c
las
sifica
ti
o
n
e
x
p
e
rime
n
t
with
c
l
o
u
d
-
b
a
se
d
a
rc
h
it
e
c
tu
re
u
sin
g
th
re
e
d
iffere
n
t
d
a
tas
e
ts,
n
a
m
e
ly
F
a
c
e
s9
4
,
F
a
c
e
s9
6
a
n
d
Un
iv
e
rsity
o
f
M
a
n
c
h
e
ste
r
In
stit
u
te
o
f
S
c
ien
c
e
a
n
d
Tec
h
n
o
lo
g
y
(
U
M
IS
T
)
fa
c
e
d
a
tas
e
t.
Th
e
re
su
lt
s
fro
m
th
is
stu
d
y
a
re
with
th
e
p
ro
p
o
se
d
c
l
o
u
d
-
b
a
se
d
a
rc
h
it
e
c
tu
r
e
,
th
e
b
e
st
a
c
c
u
ra
c
y
is
o
b
tain
e
d
b
y
K
NN
a
lg
o
rit
h
m
with
a
n
a
c
c
u
ra
c
y
o
f
9
9
%
o
n
F
a
c
e
s9
4
d
a
tas
e
t,
9
9
%
a
c
c
u
ra
c
y
o
n
F
a
c
e
s9
6
d
a
tas
e
t,
9
7
%
o
n
U
M
IS
T
fa
c
e
d
a
tas
e
t
,
a
n
d
p
e
rfo
rm
a
n
c
e
o
f
t
h
e
th
re
e
a
lg
o
rit
h
m
s
d
e
c
re
a
se
d
in
UMIS
T
fa
c
e
d
a
tas
e
t
with
fa
c
ial
v
a
riati
o
n
s fr
o
m
v
a
rio
u
s a
n
g
les
fr
o
m
left
to
rig
h
t
p
ro
f
il
e
.
K
ey
w
o
r
d
s
:
Dee
p
lear
n
in
g
Face
id
en
tific
atio
n
K
-
n
ea
r
est n
eig
h
b
o
r
R
an
d
o
m
f
o
r
est
Su
p
p
o
r
t
v
ec
to
r
m
ac
h
in
e
T
h
is i
s
a
n
o
p
e
n
a
c
c
e
ss
a
rticle
u
n
d
e
r th
e
CC B
Y
-
SA
li
c
e
n
se
.
C
o
r
r
e
s
p
o
nd
ing
A
uth
o
r
:
Ad
ity
a
Her
lam
b
an
g
Dep
ar
tm
en
t o
f
I
n
f
o
r
m
atio
n
T
e
ch
n
o
lo
g
y
Ud
ay
an
a
Un
iv
e
r
s
ity
R
ay
a
Kam
p
u
s
Un
u
d
J
im
b
ar
a
n
St.,
Ku
ta
Selatan
,
Kab
u
p
aten
B
ad
u
n
g
,
B
ali,
I
n
d
o
n
esia
E
m
ail:
h
er
lam
b
an
g
ad
t@
g
m
ail.
co
m
1.
I
NT
RO
D
UCT
I
O
N
Facial
b
io
m
etr
ics
h
av
e
m
an
y
d
is
tin
ct
ad
v
an
ta
g
es
co
m
p
a
r
e
d
to
o
th
er
b
io
m
etr
ic
c
h
ar
ac
te
r
is
tics
[
1
]
.
T
h
e
ca
m
er
a
ca
n
b
e
u
s
ed
t
o
p
ef
o
r
m
non
-
in
v
asiv
e
way
to
ca
p
tu
r
e
f
ac
es.
Face
also
h
as
a
r
ich
er
tex
tu
r
e
an
d
a
wid
e
ar
ea
.
I
t
b
ec
am
e
a
n
im
p
o
r
tan
t
m
eth
o
d
o
f
b
io
lo
g
ical
au
th
en
ticatio
n
an
d
attr
ac
ted
i
n
ter
est
in
d
if
f
er
en
t
d
o
m
ain
s
[
2
]
.
T
h
er
e
ar
e
s
ev
e
r
al
m
eth
o
d
s
th
at
ca
n
b
e
ap
p
lied
b
ef
o
r
e
id
en
tif
y
in
g
f
ac
es,
o
n
e
o
f
wh
ich
is
d
etec
tin
g
f
ac
es
in
th
e
im
ag
e
th
at
ca
n
b
e
d
o
n
e
u
s
in
g
h
is
to
g
r
am
o
f
o
r
i
en
ted
g
r
a
d
ien
ts
(
HOG
)
.
B
ased
o
n
th
e
f
ir
s
t
s
tu
d
y
t
h
a
t
p
r
o
p
o
s
e
d
t
h
e
H
O
G
m
eth
o
d
,
a
f
t
e
r
e
v
a
l
u
a
t
i
n
g
c
u
r
r
e
n
t
d
e
s
c
r
i
p
t
o
r
s
b
as
e
d
o
n
e
d
g
e
a
n
d
g
r
a
d
i
e
n
t
,
i
t
w
a
s
e
x
p
e
r
i
m
e
n
t
a
l
l
y
s
h
o
w
n
t
h
a
t
t
h
e
g
r
i
d
s
o
f
t
h
e
HOG
d
e
s
c
r
i
p
t
o
r
s
o
u
t
p
e
r
f
o
r
m
t
h
e
a
r
r
a
y
o
f
f
e
a
t
u
r
e
s
u
s
a
b
l
e
f
o
r
h
u
m
a
n
d
e
t
e
c
t
i
o
n
b
y
a
w
i
d
e
m
a
r
g
i
n
[
3
]
.
F
o
r
p
e
r
s
o
n
-
d
e
p
e
n
d
e
n
t
a
n
d
p
e
r
s
o
n
-
i
n
d
e
p
e
n
d
e
n
t
v
e
r
s
i
o
n
s
,
T
a
m
b
i
e
t
a
l
.
[
4
]
cr
ea
ted
an
im
p
r
o
v
ed
f
ac
e
r
ec
o
g
n
itio
n
s
y
s
tem
b
y
d
etec
ti
n
g
f
ac
e
u
s
in
g
HOG
an
d
co
n
v
o
lu
tio
n
al
n
eu
r
al
n
etwo
r
k
(
C
NN
)
f
o
r
ex
tr
ac
tin
g
r
elev
a
n
t f
ac
ial
f
ea
tu
r
es.
T
h
eir
r
esear
ch
r
esu
lted
in
9
6
.
1
9
% a
cc
u
r
ate
r
esu
lt f
o
r
Yale
d
ataset.
Dee
p
lear
n
in
g
h
a
v
e
d
is
tin
g
u
is
h
ed
class
if
icatio
n
an
d
lear
n
in
g
p
er
f
o
r
m
a
n
ce
,
wh
er
e
it
ca
n
au
t
o
m
atica
lly
ex
tr
ac
ts
lo
w
an
d
h
ig
h
lev
el
f
ea
tu
r
es
f
o
r
class
if
i
ca
tio
n
[
5
]
.
I
n
d
ee
p
lear
n
in
g
,
f
a
cial
f
ea
tu
r
e
ca
n
b
e
ex
tr
ac
ted
b
y
s
in
g
le
C
NN
ar
ch
itectu
r
e
f
r
o
m
lar
g
e
am
o
u
n
t
o
f
im
ag
e
co
n
tain
i
n
g
f
ac
es.
C
NN
is
in
ten
d
ed
to
Evaluation Warning : The document was created with Spire.PDF for Python.
I
SS
N
:
2
5
0
2
-
4
7
5
2
I
n
d
o
n
esian
J
E
lec
E
n
g
&
C
o
m
p
Sci,
Vo
l.
23
,
No
.
2
,
Au
g
u
s
t
20
21
:
811
-
8
2
0
812
p
r
o
ce
s
s
d
ata
th
at
h
as
a
k
n
o
wn
n
etwo
r
k
s
u
ch
as
a
to
p
o
lo
g
y
.
C
NN
is
g
en
er
ally
u
s
ed
to
id
en
tify
im
ag
e
ch
ar
ac
ter
is
tics
a
n
d
tr
en
d
s
in
ti
m
e
s
er
ies
im
ag
es.
C
NN
in
v
o
lv
es
m
u
ltip
le
co
n
n
ec
tio
n
s
.
C
o
n
v
o
lu
tio
n
,
p
o
o
lin
g
,
an
d
f
u
lly
-
co
n
n
ec
ted
lay
er
s
ar
e
th
e
b
u
ild
in
g
b
lo
ck
s
o
r
la
y
er
s
o
f
C
NN
ar
ch
itectu
r
e
[
6
]
.
As
i
n
Din
g
a
n
d
T
ao
[
7
]
C
NN
s
et
s
ar
e
u
s
ed
to
ex
tr
ac
t
f
ac
ial
ch
ar
ac
ter
is
tics
f
r
o
m
m
u
ltimo
d
al
in
f
o
r
m
atio
n
.
A
v
er
if
icatio
n
r
ate
o
f
9
8
.
4
3
%
an
d
r
ec
o
g
n
itio
n
r
ate
o
f
9
9
.
0
%
ac
h
iev
ed
o
n
th
e
la
b
eled
f
ac
es
in
th
e
wild
(
L
F
W
)
d
atab
ase.
Oth
er
ex
am
p
le
o
f
f
ac
ial
f
ea
tu
r
e
ex
tr
ac
tio
n
is
in
W
id
iak
u
m
ar
a
et
a
l.
[
8
]
wh
er
e
th
is
s
tu
d
y
r
esu
lted
in
a
f
ac
e
id
en
tific
atio
n
ap
p
licatio
n
u
s
in
g
An
d
r
o
i
d
-
b
ased
E
i
g
en
f
ac
e
wi
th
a
tr
ial
s
u
cc
ess
o
f
6
8
%
an
d
a
f
alse
p
o
s
itiv
e
r
ate
o
f
3
2
%.
C
l
as
s
i
f
i
c
at
i
o
n
u
s
i
n
g
K
NN
c
a
n
b
e
d
o
n
e
a
f
t
e
r
f
e
at
u
r
e
e
x
t
r
a
ct
i
o
n
t
o
i
d
e
n
ti
f
y
f
a
c
e
.
A
s
i
n
W
i
r
d
i
a
n
i
e
t
a
l
.
[
9
]
ca
r
r
ied
o
u
t
th
r
ee
s
tag
es
to
id
en
tify
f
ac
e
in
clu
d
in
g
f
ac
e
d
et
ec
tio
n
,
f
ea
tu
r
e
ex
tr
ac
tio
n
an
d
class
if
icat
io
n
.
T
h
e
m
eth
o
d
u
s
ed
to
ex
tr
ac
t
f
ea
t
u
r
es
is
p
r
in
cip
al
co
m
p
o
n
en
t
an
aly
s
is
(
P
C
A
)
an
d
th
e
m
eth
o
d
to
p
er
f
o
r
m
class
if
icatio
n
is
k
-
n
ea
r
est
n
ei
g
h
b
o
r
(
KNN)
.
T
h
is
p
ap
e
r
p
r
o
d
u
ce
d
a
p
r
o
g
r
am
u
s
in
g
Py
t
h
o
n
p
r
o
g
r
am
m
in
g
lan
g
u
ag
e
to
id
en
tify
f
ac
es.
T
h
e
r
esu
lt
o
b
tain
ed
f
r
o
m
s
ev
er
al
test
o
f
k
v
alu
es
g
iv
es
th
e
b
es
t
ac
cu
r
ac
y
o
f
8
1
%
with
=
1
an
d
t
h
e
g
r
ea
ter
k
v
alu
e
g
iv
es
s
m
aller
ac
cu
r
ac
y
.
A
p
ar
t
f
r
o
m
th
e
KNN,
th
e
class
if
icatio
n
o
f
th
e
d
ata
f
r
o
m
f
ac
ial
f
ea
t
u
r
e
ex
tr
ac
tio
n
ca
n
also
b
e
d
o
n
e
u
s
in
g
t
h
e
s
u
p
p
o
r
t
v
ec
t
o
r
m
ac
h
i
n
e
(
SVM
)
alg
o
r
ith
m
as
in
Sen
th
ilk
u
m
ar
an
d
Gn
an
am
u
r
t
h
y
[
1
0
]
.
T
h
is
r
esear
ch
u
s
ed
th
e
SVM
class
if
ier
to
co
m
p
ar
e
p
er
f
o
r
m
a
n
ce
ch
an
g
es
in
th
e
r
ec
o
g
n
itio
n
r
ate
o
f
d
i
f
f
er
en
t
f
ac
ial
r
ec
o
g
n
itio
n
tec
h
n
iq
u
es.
SVM
r
esu
lted
in
b
etter
class
if
icatio
n
co
m
p
ar
ed
to
o
th
er
m
eth
o
d
s
.
An
o
th
er
class
if
icatio
n
alg
o
r
ith
m
th
at
ca
n
b
e
u
s
ed
f
o
r
class
if
icatio
n
f
r
o
m
f
ac
ia
l
f
ea
tu
r
e
r
esu
lt
is
r
an
d
o
m
f
o
r
e
s
t
alg
o
r
ith
m
.
As
in
Ma
d
y
a
n
d
Hilles
[
1
1
]
R
an
d
o
m
Fo
r
est
c
lass
if
ier
is
u
s
ed
to
class
if
y
f
ac
ial
f
ea
tu
r
e
ex
tr
ac
te
d
u
s
in
g
h
is
to
g
r
am
o
f
o
r
ien
ted
g
r
ad
ien
ts
(
HOG
)
an
d
lo
ca
l
b
in
ar
y
p
atter
n
(
L
B
P
)
.
T
h
is
s
tu
d
y
r
esu
lted
in
9
7
.
6
% r
ec
o
g
n
itio
n
ac
c
u
r
ac
y
o
n
Me
d
iu
s
taf
f
d
atab
ase.
Face
I
d
en
tific
atio
n
with
d
ee
p
l
ea
r
n
in
g
u
s
in
g
C
o
n
v
o
lu
tio
n
a
l
Neu
r
al
Netwo
r
k
in
th
is
s
tu
d
y
will
b
e
r
u
n
n
in
g
in
a
cl
o
u
d
-
b
ased
ar
c
h
itectu
r
e
u
s
in
g
Flas
k
Fra
m
ew
o
r
k
s
o
th
at
im
a
g
e
ca
n
b
e
p
r
o
ce
s
s
ed
im
m
ed
iately
af
ter
r
ec
eiv
ed
b
y
th
e
cl
o
u
d
s
er
v
er
,
an
d
t
h
en
p
e
r
f
o
r
m
H
OG
f
o
r
f
ac
e
d
etec
tio
n
b
ef
o
r
e
d
o
in
g
t
h
e
im
ag
e
en
h
an
ce
m
e
n
t
p
r
o
ce
s
s
,
f
ea
tu
r
e
ex
tr
ac
tio
n
with
th
e
C
NN
to
p
r
o
d
u
c
e
1
2
8
-
d
em
b
e
d
d
i
n
g
s
,
th
en
p
er
f
o
r
m
s
class
if
icatio
n
alg
o
r
ith
m
co
m
p
ar
is
o
n
s
b
etwe
en
KNN,
L
in
ea
r
SVM
an
d
R
an
d
o
m
Fo
r
e
s
t
to
f
in
d
th
e
b
est
alg
o
r
ith
m
in
te
r
m
s
o
f
ac
c
u
r
ac
y
to
class
if
y
th
e
1
2
8
-
d
em
b
ed
d
i
n
g
s
g
en
er
ate
d
b
y
d
ee
p
lear
n
in
g
u
s
in
g
C
NN
.
2.
RE
S
E
ARCH
M
E
T
H
O
D
C
lo
u
d
-
b
ased
ar
ch
itectu
r
e
f
o
r
f
ac
e
id
en
tific
atio
n
with
d
ee
p
lear
n
in
g
co
n
s
is
t
o
f
two
p
h
ases
,
n
am
ely
tr
ain
in
g
an
d
test
in
g
s
tag
e.
T
r
ain
in
g
s
tag
e
aim
s
to
p
r
o
d
u
ce
a
class
if
icatio
n
m
o
d
el
th
at
will
b
e
u
s
ed
in
th
e
test
in
g
p
h
ase
an
d
s
to
r
e
it
in
th
e
class
if
icatio
n
m
o
d
el
d
atab
ase,
an
d
s
av
e
th
e
r
esu
lts
o
f
f
ac
ia
l
f
ea
tu
r
e
ex
tr
ac
tio
n
to
th
e
f
ac
ial
f
ea
tu
r
e
d
atab
ase
in
th
e
clo
u
d
.
At
th
e
tr
ain
in
g
s
tag
e,
th
e
p
r
e
p
r
o
ce
s
s
in
g
s
tag
e
will
b
e
ca
r
r
ied
o
u
t
af
ter
th
e
d
ev
ice
s
en
d
s
tr
ain
i
m
ag
es,
th
en
f
ac
e
d
etec
tio
n
is
co
n
d
u
cted
with
HOG,
f
ea
tu
r
e
ex
tr
ac
tio
n
u
s
in
g
C
NN,
th
en
th
e
m
o
d
el
is
tr
ain
ed
with
KNN
.
Fig
u
r
e
1
s
h
o
ws
th
e
tr
ain
in
g
s
tag
e
with
clo
u
d
-
b
ased
ar
c
h
itectu
r
e
.
T
h
e
s
ec
o
n
d
s
tag
e
o
f
f
ac
ial
i
d
e
n
tific
atio
n
af
ter
th
e
tr
ain
i
n
g
s
tag
e
is
test
in
g
s
tag
e.
T
est
im
ag
e
will
b
e
r
ec
eiv
ed
b
y
th
e
Flas
k
Fra
m
ewo
r
k
i
n
th
e
clo
u
d
.
Af
ter
th
e
test
im
ag
e
is
r
ec
eiv
ed
,
th
e
test
in
g
p
r
o
ce
s
s
will
b
e
c
ar
r
ied
o
u
t
im
m
ed
iately
.
Fig
u
r
e
2
s
h
o
ws th
e
test
in
g
p
h
ase
with
clo
u
d
-
b
a
s
ed
ar
ch
itectu
r
e.
Fig
u
r
e
1
.
T
r
ain
in
g
s
tag
e
o
f
f
ac
e
id
en
tific
atio
n
Evaluation Warning : The document was created with Spire.PDF for Python.
I
n
d
o
n
esian
J
E
lec
E
n
g
&
C
o
m
p
Sci
I
SS
N:
2502
-
4
7
5
2
C
lo
u
d
-
b
a
s
ed
a
r
ch
itectu
r
e
fo
r
fa
ce
id
en
tifi
ca
tio
n
w
ith
d
ee
p
le
a
r
n
in
g
u
s
in
g
…
(
A
d
itya
Herla
mb
a
n
g
)
813
Fig
u
r
e
2
.
T
esti
n
g
s
tag
e
o
f
f
ac
e
id
en
tific
atio
n
2
.
1
.
F
a
ce
ima
g
e
da
t
a
s
et
T
h
er
e
ar
e
3
s
ep
ar
ate
d
atasets
,
n
am
ely
th
e
Un
iv
er
s
ity
o
f
Ma
n
ch
ester
I
n
s
titu
te
o
f
Sci
en
ce
an
d
T
ec
h
n
o
lo
g
y
(
UM
I
ST
)
f
ac
e
d
a
taset,
Face
s
9
4
an
d
Face
s
9
6
,
u
s
ed
f
o
r
t
h
e
tr
ain
in
g
an
d
t
esti
n
g
p
h
ases
.
T
h
e
UM
I
ST
f
ac
e
d
ataset
co
n
tain
s
5
7
5
f
ac
es
f
r
o
m
1
5
2
s
u
b
jects
tak
en
f
r
o
m
lef
t
p
r
o
f
ile
to
r
ig
h
t
an
g
le
[
1
2
]
.
T
h
e
UM
I
ST
f
ac
e
d
ataset
e
x
am
p
le
ca
n
b
e
s
ee
n
i
n
Fig
u
r
e
3
.
Face
s
9
4
co
m
p
r
is
es
3
0
6
0
f
ac
es
f
r
o
m
1
5
3
s
u
b
jects
wh
o
ar
e
s
ea
ted
at
ar
o
u
n
d
t
h
e
s
am
e
d
is
tan
ce
f
r
o
m
th
e
ca
m
e
r
a
an
d
ask
ed
to
talk
as
twen
ty
co
n
s
ec
u
tiv
e
p
h
o
t
o
g
r
ap
h
s
ar
e
tak
en
.
T
h
e
s
p
ee
ch
is
u
s
ed
to
in
c
o
r
p
o
r
ate
ch
a
n
g
es
in
f
ac
ial
ex
p
r
ess
io
n
s
th
at
ar
e
m
il
d
an
d
n
o
r
m
al.
T
h
e
Face
s
9
4
Data
s
et
ex
am
p
le
ca
n
b
e
s
ee
n
i
n
Fig
u
r
e
4
.
Face
s
9
6
i
n
clu
d
es
3
0
4
0
f
ac
es f
r
o
m
1
5
2
s
u
b
jects
tak
en
u
s
in
g
a
f
ix
ed
ca
m
er
a,
wh
er
e
th
e
s
u
b
ject
tak
es
o
n
e
o
r
m
o
r
e
s
tep
s
f
o
r
war
d
to
wa
r
d
s
th
e
ca
m
er
a
wh
en
th
e
p
ictu
r
e
is
tak
en
to
i
n
tr
o
d
u
ce
m
ajo
r
h
ea
d
d
if
f
e
r
en
ce
s
b
etwe
en
im
ag
es
o
f
th
e
s
am
e
p
e
r
s
o
n
[
1
3
]
.
T
h
e
Face
s
9
6
Data
s
et
ex
am
p
le
ca
n
b
e
s
ee
n
in
Fig
u
r
e
5
.
Fig
u
r
e
3
.
UM
I
ST
f
ac
e
d
ataset
ex
am
p
le
Fig
u
r
e
4
.
Face
s
9
4
d
ataset
ex
am
p
le
Fig
u
r
e
5
.
Face
s
9
6
d
ataset
ex
am
p
le
Evaluation Warning : The document was created with Spire.PDF for Python.
I
SS
N
:
2
5
0
2
-
4
7
5
2
I
n
d
o
n
esian
J
E
lec
E
n
g
&
C
o
m
p
Sci,
Vo
l.
23
,
No
.
2
,
Au
g
u
s
t
20
21
:
811
-
8
2
0
814
2
.
2
.
F
a
ce
ima
g
e
da
t
a
s
et
I
m
ag
e
en
h
a
n
ce
m
en
t'
s
m
ain
g
o
al
is
to
im
p
r
o
v
e
g
r
ap
h
ical
clar
ity
an
d
ac
cu
r
ac
y
o
f
an
im
a
g
e,
o
r
to
g
iv
e
an
en
h
a
n
ce
d
p
ictu
r
e
o
f
tr
an
s
f
o
r
m
atio
n
f
o
r
u
p
co
m
i
n
g
im
ag
e
p
r
o
ce
s
s
in
g
[
1
4
]
.
T
h
e
im
a
g
e
en
h
an
ce
m
en
t
s
tag
e
i
n
th
is
s
tu
d
y
co
n
s
is
ts
o
f
co
n
v
er
s
io
n
to
g
r
a
y
s
ca
le,
im
ag
e
s
m
o
o
t
h
in
g
with
a
Gau
s
s
ian
Fil
te
r
,
an
d
d
iv
id
in
g
g
r
a
y
b
y
m
o
r
p
h
o
lo
g
y
im
ag
e.
No
is
e
in
th
e
im
ag
e
is
r
ed
u
ce
d
b
y
d
o
in
g
im
ag
e
s
m
o
o
th
i
n
g
with
g
au
s
s
ian
f
ilter
,
th
er
eb
y
im
p
r
o
v
in
g
th
e
q
u
ality
o
f
an
i
m
ag
e
[
1
5
]
.
W
e
u
s
e
im
ag
e
en
h
an
ce
m
en
t
f
ea
tu
r
e
f
r
o
m
Op
e
n
C
V
b
ec
au
s
e
it
h
as
th
e
b
en
ef
it
o
f
b
ein
g
a
m
u
lti
-
p
latf
o
r
m
lib
r
ar
y
[
1
6
]
s
o
o
u
r
clo
u
d
-
b
ased
ar
ch
itectu
r
e
is
n
o
t
lim
ited
to
a
s
p
ec
if
ic
p
latf
o
r
m
.
Fig
u
r
e
6
is
th
e
r
esu
lt o
f
th
e
im
a
g
e
en
h
a
n
ce
m
en
t c
a
r
r
ied
o
u
t in
th
is
s
tu
d
y
.
Fig
u
r
e
6
.
I
m
ag
e
e
n
h
an
ce
m
e
n
t
2
.
3
.
F
a
ce
det
ec
t
io
n
Face
d
etec
tio
n
is
th
e
p
r
o
ce
s
s
o
f
d
is
co
v
e
r
in
g
b
o
u
n
d
in
g
b
o
x
e
s
o
f
h
u
m
an
f
ac
e
in
an
im
ag
e
s
eq
u
en
ce
.
T
h
is
r
esear
ch
m
ak
es
u
s
e
o
f
t
h
e
HOG
m
eth
o
d
an
d
SVM
to
d
etec
t
f
ac
e
in
an
im
ag
e.
W
it
h
HOG,
th
e
p
ictu
r
e
p
ix
el'
s
h
o
r
izo
n
tal
g
r
ad
ien
t
an
d
th
e
p
ictu
r
e
p
ix
el'
s
v
er
tical
g
r
ad
ien
t
is
p
r
esen
ted
in
(
1
)
a
n
d
(
2
)
.
T
h
e
g
r
ad
ien
t
m
ag
n
itu
d
e
a
n
d
p
ix
el
d
ir
ec
tio
n
ca
n
b
e
s
ee
n
in
(
3
)
a
n
d
(
4
)
.
(
,
)
=
(
+
1
,
)
−
(
−
1
,
)
(
1
)
(
,
)
=
(
,
+
1
)
−
(
,
−
1
)
(
2
)
(
,
)
=
√
(
,
)
2
+
(
,
)
2
(
3
)
(
,
)
=
ta
n
−
1
(
(
,
)
(
,
)
)
(
4
)
T
h
e
g
r
a
d
ien
t
h
as
a
m
a
g
n
itu
d
e
an
d
a
d
i
r
ec
tio
n
at
ea
ch
p
ix
el
.
I
n
(
3
)
is
u
s
ed
to
m
ea
s
u
r
e
th
e
g
r
ad
ie
n
t
d
ir
ec
tio
n
,
wh
ile
(
4
)
is
u
s
ed
to
ca
lcu
late
th
e
g
r
ad
ien
t
m
ag
n
itu
d
e.
T
h
e
m
a
x
im
u
m
o
f
t
h
e
th
r
ee
ch
an
n
els'
g
r
ad
ien
ts
is
th
e
m
a
g
n
itu
d
e
o
f
th
e
g
r
ad
ie
n
t
at
a
p
ix
el,
an
d
th
e
an
g
le
is
th
e
a
n
g
le
e
q
u
iv
alen
t
to
th
e
m
a
x
im
u
m
g
r
ad
ien
t.
T
o
c
o
u
n
t
th
e
g
r
ad
ie
n
t
d
ir
ec
tio
n
u
s
in
g
(
3
)
an
d
g
r
a
d
ien
t
m
ag
n
itu
d
e
u
s
in
g
(
4
)
,
th
e
p
r
o
v
id
e
d
f
r
am
e
is
d
iv
id
ed
in
to
ce
lls
,
wh
ich
a
r
e
p
ix
el
-
s
ized
r
ec
tan
g
u
lar
o
r
ci
r
cu
lar
a
r
ea
s
.
Fo
r
ea
ch
ce
ll,
t
h
e
g
r
ad
ien
t
f
ea
tu
r
e
v
ec
to
r
s
ar
e
th
en
ca
lcu
lated
.
I
n
ea
ch
s
in
g
le
f
r
am
e,
th
e
f
ea
t
u
r
e
v
ec
to
r
is
th
en
co
n
s
tr
u
cted
u
s
in
g
th
is
g
r
ad
ien
t
f
ea
tu
r
e
v
ec
t
o
r
.
Fin
ally
,
th
e
H
OG
f
ea
tu
r
e
v
ec
t
o
r
is
p
r
o
d
u
ce
d
b
y
c
o
m
b
in
i
n
g
all
g
r
ad
ie
n
t
f
ea
tu
r
e
v
ec
to
r
s
d
er
iv
e
d
f
r
o
m
d
if
f
e
r
en
t
im
ag
es,
an
d
th
en
in
p
u
tted
to
th
e
SVM
to
ex
tr
ac
t
an
ar
r
ay
o
f
b
o
u
n
d
i
n
g
b
o
x
es
f
o
r
th
e
h
u
m
an
f
ac
e
[
1
7
]
.
I
n
p
u
t im
a
g
e
an
d
f
ea
tu
r
e
v
ec
to
r
s
g
e
n
er
ated
f
r
o
m
th
e
HOG
m
eth
o
d
ca
n
b
e
s
ee
n
in
Fig
u
r
e
7
.
Fig
u
r
e
7
.
I
n
p
u
t
i
m
ag
e
an
d
HOG
f
ea
tu
r
e
v
ec
to
r
s
r
esu
lt
Evaluation Warning : The document was created with Spire.PDF for Python.
I
n
d
o
n
esian
J
E
lec
E
n
g
&
C
o
m
p
Sci
I
SS
N:
2502
-
4
7
5
2
C
lo
u
d
-
b
a
s
ed
a
r
ch
itectu
r
e
fo
r
fa
ce
id
en
tifi
ca
tio
n
w
ith
d
ee
p
le
a
r
n
in
g
u
s
in
g
…
(
A
d
itya
Herla
mb
a
n
g
)
815
2
.
4
.
F
e
a
t
ure
ex
t
r
a
ct
io
n us
ing
CNN
Featu
r
e
ex
tr
ac
tio
n
is
a
p
h
ase
f
o
r
e
x
tr
ac
tin
g
f
ac
ial
c
h
ar
ac
ter
i
s
tic
f
ea
tu
r
es
f
r
o
m
a
n
im
ag
e.
C
NN
f
r
o
m
Dlib
an
d
R
esNet
ar
ch
itectu
r
e
with
2
9
C
o
n
v
lay
er
s
,
wh
ich
is
a
v
ar
iety
o
f
R
esNet
-
3
4
u
s
in
g
f
ewe
r
lay
er
s
an
d
h
alf
th
e
n
u
m
b
e
r
o
f
f
il
ter
s
in
e
ac
h
lay
er
[
1
8
]
ar
e
u
s
ed
in
th
is
s
tu
d
y
to
e
x
tr
ac
t
f
ea
tu
r
es.
R
esNet
allo
ws
d
ee
p
er
ar
ch
itectu
r
al
tr
ain
i
n
g
,
b
ec
au
s
e
th
e
la
y
er
lea
r
n
s
r
esid
u
al
f
u
n
ctio
n
s
b
y
r
ef
e
r
r
in
g
to
lay
er
i
n
p
u
ts
an
d
d
o
es
n
o
t
lear
n
f
u
n
ctio
n
s
th
at
ar
e
n
o
t
r
e
f
er
en
ce
d
.
T
h
is
allo
ws
th
e
n
et
wo
r
k
to
b
e
r
esil
ien
t
to
t
h
e
g
r
ad
ien
t
d
is
ap
p
ea
r
in
g
p
r
o
b
lem
an
d
h
an
d
le
th
e
d
r
o
p
in
ac
cu
r
ac
y
th
at
o
cc
u
r
s
in
co
n
v
en
tio
n
al
d
ee
p
g
r
id
s
[
1
9
]
.
T
h
i
s
s
tu
d
y
u
s
ed
C
NN
ar
ch
itectu
r
e
f
r
o
m
Dlib
[
2
0
]
to
ex
tr
ac
t
f
ac
ial
f
ea
t
u
r
es.
T
h
e
C
NN
ar
ch
itectu
r
e
wh
ich
co
m
b
i
n
es
lo
ca
l
r
ec
e
p
tiv
e
f
ield
s
,
s
h
ar
ed
weig
h
ts
an
d
p
o
o
lin
g
[
2
1
]
u
s
ed
in
th
is
s
tu
d
y
ca
n
b
e
s
ee
n
i
n
Fig
u
r
e
8
.
T
h
is
m
eth
o
d
g
en
er
ates
128
-
d
f
ea
tu
r
e
v
ec
to
r
s
f
r
o
m
f
a
cial
im
ag
es
th
at
h
av
e
b
ee
n
d
e
tecte
d
in
th
e
f
ac
e
d
etec
tio
n
s
tag
e.
Fig
u
r
e
9
is
an
ex
am
p
le
o
f
th
e
f
ea
tu
r
e
ex
tr
ac
ti
o
n
r
esu
lts
f
r
o
m
o
n
e
o
f
th
e
f
ac
e
s
in
th
e
d
ataset.
Fig
u
r
e
8
.
C
NN
ar
ch
itectu
r
e
Fig
u
r
e
9
.
Featu
r
e
ex
tr
ac
tio
n
u
s
in
g
C
NN
2
.
5
.
Cla
s
s
if
ica
t
io
n wit
h K
N
N
a
lg
o
rit
hm
On
e
o
f
th
e
s
im
p
le
alg
o
r
ith
m
s
th
at
ca
n
b
e
u
s
ed
in
th
e
class
if
icatio
n
p
r
o
ce
s
s
to
m
atc
h
d
at
a
b
etwe
en
test
in
g
an
d
tr
ain
in
g
d
ata
f
r
o
m
f
ac
e
d
atasets
is
K
NN
[
9
]
.
KNN
was
u
s
ed
in
th
e
ea
r
ly
1
9
7
0
s
f
o
r
s
tatis
tical
esti
m
atio
n
an
d
p
atter
n
r
ec
o
g
n
itio
n
[
2
2
]
.
KNN
p
er
f
o
r
m
s
well
o
n
m
an
y
s
am
p
les.
W
h
e
n
a
n
ew
test
s
am
p
le
ap
p
ea
r
s
,
th
e
d
is
tan
ce
b
etwe
en
it
an
d
o
th
er
tr
ain
e
d
s
am
p
les
will
b
e
d
eter
m
in
ed
u
s
in
g
th
e
k
v
alu
e,
an
d
th
e
test
s
am
p
le
will
b
e
ca
lcu
late
d
b
y
th
e
class
m
em
b
er
wh
o
s
e
s
am
p
le
is
n
ea
r
est
to
th
e
test
s
am
p
le
[
2
3
]
.
T
h
is
s
tu
d
y
u
s
ed
k
=1
to
p
r
o
d
u
ce
p
r
ed
ictio
n
r
esu
lt
b
ased
o
n
th
e
n
ea
r
es
t
n
eig
h
b
o
r
.
T
h
is
s
tu
d
y
u
s
e
class
if
icatio
n
f
u
n
ctio
n
f
r
o
m
Scik
it
-
lear
n
[
2
4
]
.
T
o
f
in
d
th
e
n
ea
r
est
n
eig
h
b
o
r
,
th
is
s
tu
d
y
u
s
ed
E
u
clid
ea
n
d
is
tan
ce
f
o
r
m
u
la
th
at
ca
n
b
e
s
ee
n
in
(
5
)
.
(
,
)
=
√
∑
(
−
)
2
=
1
(
5
)
T
h
e
(
5
)
is
th
e
eu
clid
ea
n
d
is
tan
ce
f
o
r
m
u
la
wh
er
e
x
is
th
e
p
ar
a
m
eter
o
f
test
in
g
d
ata,
x
i is th
e
p
ar
am
eter
o
f
tr
ain
in
g
d
ata.
T
h
e
p
ar
am
et
er
n
is
th
e
d
im
en
s
io
n
o
f
th
e
f
ea
tu
r
e
v
ec
t
o
r
.
E
u
clid
ea
n
Dis
tan
ce
is
u
tili
ze
d
to
in
cr
ea
s
e
ac
cu
r
ac
y
b
y
m
ea
s
u
r
in
g
th
e
d
is
tan
ce
b
etw
ee
n
p
o
in
ts
alo
n
g
a
s
tr
aig
h
t
lin
e,
a
n
d
is
teac
h
in
g
an
d
Evaluation Warning : The document was created with Spire.PDF for Python.
I
SS
N
:
2
5
0
2
-
4
7
5
2
I
n
d
o
n
esian
J
E
lec
E
n
g
&
C
o
m
p
Sci,
Vo
l.
23
,
No
.
2
,
Au
g
u
s
t
20
21
:
811
-
8
2
0
816
r
esear
ch
r
esu
lts
.
T
h
is
d
is
tan
c
e
m
eth
o
d
u
s
es
th
e
p
y
th
ag
o
r
e
an
th
eo
r
em
.
C
alcu
latio
n
with
eu
clid
ea
n
d
is
tan
ce
aim
s
to
co
m
p
ar
e
t
h
e
m
in
im
u
m
d
is
tan
ce
o
f
th
e
tr
ai
n
in
g
im
a
g
e
an
d
th
e
test
im
ag
e.
2
.
6
.
Cla
s
s
if
ica
t
io
n wit
h SVM
a
lg
o
rit
hm
Su
p
p
o
r
t
Vec
to
r
Ma
ch
in
e
(
SVM)
is
a
clas
s
if
icatio
n
alg
o
r
ith
m
th
at
s
tu
d
y
h
o
w
to
lab
el
o
b
jects
b
y
u
s
in
g
ex
am
p
les.
SVM
is
a
m
a
th
em
atica
l
o
b
ject
th
at
o
p
tim
iz
es
a
m
ath
em
atica
l
f
u
n
ctio
n
in
r
elatio
n
to
a
g
iv
en
d
ata
s
et
[
25]
.
B
etwe
en
two
g
r
o
u
p
s
o
f
r
esu
lts
,
th
e
s
ep
ar
atin
g
h
y
p
er
p
lan
e
with
th
e
lar
g
est
m
ar
g
in
is
f
o
u
n
d
b
y
SVM
[
2
6
]
.
T
h
e
k
er
n
el
u
s
ed
in
th
is
s
tu
d
y
is
L
in
ea
r
wh
ich
th
e
f
u
n
ctio
n
ca
n
b
e
s
ee
n
in
(
6
)
.
(
,
)
=
(
6
)
2
.
7
.
Cla
s
s
if
ica
t
io
n wit
h
ra
nd
o
m
f
o
re
s
t
a
lg
o
ri
t
hm
R
an
d
o
m
f
o
r
est
is
co
llectio
n
o
f
tr
ee
p
r
e
d
icto
r
s
p
u
t
to
g
eth
er
i
n
a
r
an
d
o
m
o
r
d
er
[
2
7
]
.
E
ac
h
t
r
ee
in
th
e
f
o
r
est
is
b
ased
o
n
a
r
an
d
o
m
v
ec
to
r
v
alu
e
t
h
at
it
s
am
p
les
in
d
ep
en
d
en
tly
an
d
with
th
e
s
am
e
d
is
tr
ib
u
tio
n
.
Fo
r
an
en
s
em
b
le
o
f
class
if
ier
s
ℎ
1
(
)
,
ℎ
2
(
)
,
…
,
ℎ
(
)
,
th
e
m
ar
g
in
f
u
n
ctio
n
ca
n
b
e
u
s
ed
in
an
eq
u
atio
n
with
a
tr
ain
in
g
r
an
g
e
r
a
n
d
o
m
ly
s
elec
ted
f
r
o
m
a
d
is
tr
ib
u
tio
n
o
f
r
an
d
o
m
v
ec
to
r
s
,
.
T
h
e
m
ar
g
in
f
u
n
ctio
n
ca
n
b
e
s
ee
n
in
(
7
)
.
(
,
)
=
(
ℎ
(
)
=
)
−
≠
(
ℎ
(
)
=
)
(
7
)
T
h
e
m
ar
g
i
n
in
d
icate
s
h
o
w
m
u
ch
th
e
r
ig
h
t
class
's
av
er
ag
e
n
u
m
b
er
o
f
v
o
tes
in
X,
Y
e
x
ce
ed
s
th
e
o
th
er
class
es
'
av
er
ag
e
n
u
m
b
er
o
f
v
o
tes.
T
h
e
s
m
aller
th
e
m
ar
g
in
o
f
er
r
o
r
,
th
e
m
o
r
e
co
n
f
id
en
t
th
e
class
if
icatio
n
[
2
7
]
.
R
an
d
o
m
Fo
r
est b
lo
ck
d
iag
r
a
m
o
f
r
an
d
o
m
v
ec
to
r
a
n
d
th
e
r
esu
lts
ar
e
s
h
o
wn
in
Fig
u
r
e
1
0
.
T
h
e
f
a
c
i
a
l
f
ea
t
u
r
e
d
a
t
a
s
e
t
i
s
u
s
e
d
a
s
tr
a
i
n
in
g
d
a
ta
f
o
r
d
if
f
er
e
n
t
d
e
c
i
s
i
o
n
tr
e
e
s
.
T
h
i
s
d
a
t
a
s
e
t
i
n
c
lu
d
e
s
o
b
s
e
r
v
a
t
io
n
s
an
d
a
tt
r
i
b
u
t
e
s
th
a
t
w
i
l
l
b
e
p
ic
k
ed
a
t
r
an
d
o
m
a
s
n
o
d
e
s
ar
e
s
ep
a
r
a
te
d
.
E
a
c
h
t
r
e
e
's
l
e
a
f
n
o
d
e
r
e
p
r
e
s
e
n
t
s
th
e
f
in
a
l
o
u
tp
u
t
g
e
n
e
r
a
te
d
b
y
th
a
t
p
ar
t
i
c
u
la
r
d
e
c
i
s
i
o
n
t
r
e
e.
T
h
e
en
d
p
r
o
d
u
c
t
i
s
ch
o
s
en
b
y
a
m
a
j
o
r
i
ty
-
v
o
t
in
g
m
e
t
h
o
d
.
I
n
th
i
s
c
a
s
e
,
th
e
f
in
a
l
o
u
t
p
u
t
o
f
th
i
s
m
e
t
h
o
d
i
s
th
e
o
u
tp
u
t
s
e
l
ec
t
e
d
b
y
th
e
p
lu
r
a
l
i
ty
o
f
d
e
c
i
s
i
o
n
tr
e
e
s
.
Fig
u
r
e
1
0
.
R
an
d
o
m
f
o
r
est cla
s
s
if
ier
b
lo
ck
d
iag
r
am
3.
RE
SU
L
T
S
A
ND
D
IS
CU
SS
I
O
N
I
n
t
h
i
s
s
t
u
d
y
,
t
h
r
e
e
d
i
f
f
e
r
e
n
t
d
at
a
s
et
s
,
n
a
m
e
l
y
F
a
c
es
9
4
,
F
a
c
es
9
6
,
a
n
d
U
M
I
S
T
f
a
ce
d
a
t
as
et
,
wer
e
u
s
e
d
t
o
c
a
l
c
u
l
a
te
t
h
e
a
c
c
u
r
a
c
y
o
f
t
h
e
i
d
e
n
t
i
f
i
c
at
i
o
n
r
e
s
u
l
ts
.
T
h
e
p
u
r
p
o
s
e
o
f
u
s
i
n
g
t
h
r
e
e
d
i
f
f
e
r
e
n
t
d
a
t
a
s
e
t
s
w
as
t
o
c
a
l
c
u
l
at
e
t
h
e
a
c
c
u
r
a
c
y
o
f
f
a
c
e
i
d
e
n
ti
f
i
c
a
tio
n
f
r
o
m
d
i
f
f
e
r
e
n
t
c
o
n
d
i
t
i
o
n
s
.
T
a
b
l
e
1
s
h
o
ws
t
h
e
d
a
t
as
e
t
u
s
e
d
in
t
h
i
s
a
n
a
l
y
s
is
.
Evaluation Warning : The document was created with Spire.PDF for Python.
I
n
d
o
n
esian
J
E
lec
E
n
g
&
C
o
m
p
Sci
I
SS
N:
2502
-
4
7
5
2
C
lo
u
d
-
b
a
s
ed
a
r
ch
itectu
r
e
fo
r
fa
ce
id
en
tifi
ca
tio
n
w
ith
d
ee
p
le
a
r
n
in
g
u
s
in
g
…
(
A
d
itya
Herla
mb
a
n
g
)
817
Face
s
9
4
is
u
s
ed
to
ca
lcu
late
ac
cu
r
ac
y
f
o
r
m
o
d
er
ate
an
d
n
at
u
r
al
v
ar
iatio
n
s
in
f
ac
ial
e
x
p
r
e
s
s
io
n
s
with
th
e
s
p
ea
k
in
g
s
u
b
ject,
Face
s
9
6
is
u
s
ed
to
ca
lcu
late
ac
cu
r
a
cy
f
o
r
v
ar
iatio
n
s
in
h
ea
d
m
o
v
em
en
t
at
d
if
f
er
en
t
d
is
tan
ce
s
f
r
o
m
th
e
ca
m
e
r
a,
a
n
d
th
e
UM
I
ST
f
ac
e
d
ataset
i
s
u
s
ed
to
ca
lcu
late
ac
cu
r
ac
y
f
o
r
f
ac
ial
v
ar
iatio
n
s
f
r
o
m
v
ar
i
o
u
s
an
g
les
f
r
o
m
lef
t
to
r
ig
h
t
p
r
o
f
ile.
B
o
th
Face
s
9
4
an
d
Face
s
9
6
ar
e
all
f
r
o
n
tal
v
iew
wh
ile
UM
I
ST
f
ac
e
d
ataset
ca
p
tu
r
ed
f
r
o
m
d
if
f
er
en
t a
n
g
le
f
r
o
m
le
f
t to
r
ig
h
t p
r
o
f
ile.
T
ab
le
1
.
T
r
ai
n
in
g
a
n
d
test
in
g
d
ata
D
a
t
a
s
e
t
Tr
a
i
n
i
n
g
d
a
t
a
Te
st
i
n
g
d
a
t
a
To
t
a
l
c
l
a
ss
F
a
c
e
s
9
4
2
1
4
2
9
1
8
1
5
3
F
a
c
e
s
9
6
2
1
8
9
9
0
5
1
5
2
U
M
I
S
T
F
a
c
e
D
a
t
a
set
4
0
2
1
7
3
20
Fo
r
k
-
n
ea
r
est
n
eig
h
b
o
r
,
lin
e
ar
SVM,
an
d
r
a
n
d
o
m
f
o
r
est
(
RF
)
class
if
icatio
n
,
th
is
s
tu
d
y
u
s
ed
Scik
it
-
lear
n
lib
r
ar
y
.
T
h
e
k
v
alu
e
p
ar
am
eter
u
s
ed
in
th
is
s
tu
d
y
is
=
1
to
p
r
o
d
u
ce
p
r
e
d
ictio
n
r
esu
lt
b
ased
o
n
th
e
n
ea
r
est
n
eig
h
b
o
r
.
T
h
e
r
esu
lt
o
f
p
r
ec
is
io
n
,
r
ec
all,
F1
-
s
co
r
e,
s
u
p
p
o
r
t
a
n
d
p
r
o
ce
s
s
in
g
ti
m
e
ca
n
b
e
s
ee
n
i
n
T
ab
le
s
2
-
4.
T
ab
le
2
.
Pre
cisi
o
n
,
r
ec
all
,
F1
-
s
co
r
e,
s
u
p
p
o
r
t a
n
d
p
r
o
ce
s
s
in
g
tim
e
f
o
r
Face
s
9
4
d
ataset
A
l
g
o
r
i
t
h
m
P
r
e
c
i
s
i
o
n
R
e
c
a
l
l
F1
-
S
c
o
r
e
S
u
p
p
o
r
t
P
r
o
c
e
ss
i
n
g
t
i
me
(
s)
K
N
N
9
9
%
9
9
%
9
9
%
9
1
8
0
.
7
6
Li
n
e
a
r
S
V
M
9
8
%
9
9
%
9
9
%
9
1
8
1
.
4
2
R
a
n
d
o
m F
o
r
e
s
t
9
8
%
9
8
%
9
8
%
9
1
8
0
.
5
9
T
ab
le
3
.
Pre
cisi
o
n
,
r
ec
all
,
F1
-
s
co
r
e,
s
u
p
p
o
r
t a
n
d
p
r
o
ce
s
s
in
g
tim
e
f
o
r
Face
s
9
6
d
ataset
A
l
g
o
r
i
t
h
m
P
r
e
c
i
s
i
o
n
R
e
c
a
l
l
F1
-
S
c
o
r
e
S
u
p
p
o
r
t
P
r
o
c
e
ss
i
n
g
t
i
me
(
s)
K
N
N
9
9
%
9
9
%
9
9
%
9
0
5
0
.
9
2
Li
n
e
a
r
S
V
M
9
8
%
9
8
%
9
8
%
9
0
5
1
.
4
0
R
a
n
d
o
m F
o
r
e
s
t
9
7
%
9
8
%
9
7
%
9
0
5
0
.
9
0
T
ab
le
4
.
Pre
cisi
o
n
,
r
ec
all
,
F1
-
s
co
r
e,
s
u
p
p
o
r
t a
n
d
p
r
o
ce
s
s
in
g
tim
e
f
o
r
UM
I
ST
f
ac
e
d
ataset
A
l
g
o
r
i
t
h
m
P
r
e
c
i
s
i
o
n
R
e
c
a
l
l
F1
-
S
c
o
r
e
S
u
p
p
o
r
t
P
r
o
c
e
ss
i
n
g
t
i
me
(
s)
K
N
N
9
7
%
9
8
%
9
7
%
1
7
3
0
.
0
6
Li
n
e
a
r
S
V
M
8
5
%
7
4
%
7
6
%
1
7
3
0
.
0
7
R
a
n
d
o
m F
o
r
e
s
t
9
3
%
9
3
%
9
2
%
1
7
3
0
.
0
5
Fro
m
T
ab
le
2
to
T
ab
le
4
,
it is
s
h
o
wn
th
at
th
e
th
r
ee
al
g
o
r
ith
m
s
h
av
e
r
elativ
ely
g
o
o
d
p
er
f
o
r
m
an
ce
with
F1
-
s
co
r
e
>
9
5
%
o
n
th
e
Face
s
9
4
an
d
Face
s
9
6
d
atasets
.
Ho
wev
er
,
th
e
t
h
r
ee
alg
o
r
ith
m
s
ex
p
er
ien
ce
d
a
d
ec
r
ea
s
e
in
F1
-
s
co
r
e
o
n
th
e
UM
I
ST
f
ac
e
d
ataset
.
T
h
e
ca
u
s
e
o
f
th
e
d
e
cr
ea
s
e
in
t
h
e
F1
-
s
co
r
e
o
n
th
e
UM
I
ST
f
ac
e
d
ataset
is
th
e
v
ar
i
atio
n
in
ta
k
in
g
f
ac
ial
im
ag
es
f
r
o
m
t
h
e
lef
t
s
id
e
to
t
h
e
r
ig
h
t
s
id
e
o
n
th
e
UM
I
ST
f
a
ce
d
ataset
,
wh
ich
is
n
o
t
av
ailab
le
in
th
e
Face
s
9
4
an
d
Face
s
9
6
d
atasets
.
I
n
ter
m
s
o
f
r
ec
o
g
n
itio
n
tim
e
,
r
a
n
d
o
m
f
o
r
est
y
ield
s
t
h
e
f
astes
t id
en
tific
atio
n
tim
e
am
o
n
g
th
e
o
th
er
2
a
lg
o
r
ith
m
s
.
Am
o
n
g
th
e
t
h
r
ee
al
g
o
r
ith
m
s
,
KNN
h
as
a
s
tab
le
p
er
f
o
r
m
an
c
e
co
m
p
a
r
ed
to
lin
ea
r
SVM
a
n
d
r
a
n
d
o
m
f
o
r
est.
On
th
e
UM
I
ST
f
ac
e
d
a
taset
,
th
e
KNN
F1
-
s
co
r
e
was
o
n
ly
r
e
d
u
ce
d
b
y
2
%
f
r
o
m
t
h
e
h
ig
h
est
F1
-
s
co
r
e
o
n
th
e
Face
s
9
6
d
ataset,
n
am
ely
9
9
%,
th
e
SVM
lin
ea
r
F1
-
s
co
r
e
was
2
3
%
less
th
an
th
e
h
i
g
h
est
F1
-
s
co
r
e
o
n
Face
s
9
4
,
an
d
th
e
r
an
d
o
m
f
o
r
e
s
t
F1
-
s
co
r
e
was
r
ed
u
ce
d
b
y
6
%
o
f
th
e
h
ig
h
est
F1
-
s
co
r
e
o
n
th
e
Face
s
9
4
d
ataset.
Fro
m
th
e
th
r
ee
test
s
co
n
d
u
cte
d
,
KNN
o
b
tain
e
d
th
e
h
ig
h
est
F1
-
s
co
r
e
th
at
is
9
9
%
o
n
th
e
Face
s
9
4
d
ataset,
9
9
%
o
n
th
e
Face
s
9
6
d
ataset,
an
d
9
7
% o
n
th
e
UM
I
ST
f
ac
e
d
ataset
.
Ap
ar
t
f
r
o
m
t
h
e
F1
-
s
co
r
e,
e
v
alu
atio
n
o
f
t
h
e
th
r
ee
alg
o
r
ith
m
s
ca
n
also
b
e
d
o
n
e
b
y
l
o
o
k
in
g
a
t
th
e
f
alse
ac
ce
p
tan
ce
r
ate
(
FAR
)
an
d
f
al
s
e
r
ejec
tio
n
r
ate
(
FR
R
)
.
FAR
is
th
e
n
u
m
b
er
o
f
tim
es
th
at
u
n
wan
ted
in
d
iv
id
u
als
o
r
im
p
o
s
to
r
s
ar
e
wr
o
n
g
l
y
ac
ce
p
ted
d
u
r
in
g
r
ec
o
g
n
itio
n
,
o
r
th
e
p
er
ce
n
tag
e
o
f
tim
es
th
a
t
f
alse
ac
ce
p
tan
ce
h
ap
p
en
s
.
FR
R
is
th
e
n
u
m
b
er
o
f
id
en
tific
atio
n
in
s
tan
ce
s
in
wh
ich
au
th
o
r
ized
in
d
iv
id
u
al
s
o
r
tr
u
e
lab
el
was
wr
o
n
g
ly
r
ejec
ted
.
T
h
e
co
m
p
a
r
is
o
n
b
etwe
en
ac
c
u
r
ac
y
,
FAR
,
an
d
FR
R
f
o
r
ea
c
h
alg
o
r
ith
m
ca
n
b
e
s
ee
n
in
t
h
e
g
r
ap
h
i
n
Fig
u
r
e
s
11
-
1
3
.
Evaluation Warning : The document was created with Spire.PDF for Python.
I
SS
N
:
2
5
0
2
-
4
7
5
2
I
n
d
o
n
esian
J
E
lec
E
n
g
&
C
o
m
p
Sci,
Vo
l.
23
,
No
.
2
,
Au
g
u
s
t
20
21
:
811
-
8
2
0
818
Fig
u
r
e
1
1
.
Acc
u
r
ac
y
,
FAR
,
an
d
FR
R
o
f
KNN
alg
o
r
ith
m
Fig
u
r
e
1
2
.
Acc
u
r
ac
y
,
FAR
,
an
d
FR
R
o
f
lin
ea
r
SVM
alg
o
r
ith
m
Fig
u
r
e
1
3
.
Acc
u
r
ac
y
,
FAR
,
an
d
FR
R
o
f
r
an
d
o
m
f
o
r
est
alg
o
r
ith
m
Evaluation Warning : The document was created with Spire.PDF for Python.
I
n
d
o
n
esian
J
E
lec
E
n
g
&
C
o
m
p
Sci
I
SS
N:
2502
-
4
7
5
2
C
lo
u
d
-
b
a
s
ed
a
r
ch
itectu
r
e
fo
r
fa
ce
id
en
tifi
ca
tio
n
w
ith
d
ee
p
le
a
r
n
in
g
u
s
in
g
…
(
A
d
itya
Herla
mb
a
n
g
)
819
Fro
m
Fig
u
r
e
1
1
th
r
o
u
g
h
Fig
u
r
e
1
3
,
it
ca
n
b
e
s
ee
n
t
h
at
th
er
e
is
an
in
c
r
ea
s
e
in
FAR
an
d
FR
R
an
d
a
d
ec
r
ea
s
e
in
ac
cu
r
ac
y
o
n
t
h
e
U
MI
ST
f
ac
e
d
ataset
.
T
h
e
in
cr
e
ase
in
FAR
an
d
FR
R
o
n
th
e
UM
I
ST
f
ac
e
d
atas
et
in
d
icate
s
th
at
th
e
th
r
ee
class
if
icatio
n
alg
o
r
ith
m
s
ar
e
m
o
r
e
v
u
ln
er
ab
le
to
f
alse
ac
ce
p
tan
ce
an
d
f
alse
r
ejec
tio
n
o
n
th
e
f
ac
e
d
ataset
tak
en
f
r
o
m
th
e
lef
t
p
r
o
f
ile
t
o
th
e
r
ig
h
t
p
r
o
f
ile
th
an
th
e
d
ataset
tak
e
n
f
r
o
m
th
e
f
r
o
n
t
s
id
e
o
f
th
e
f
ac
e.
T
h
e
s
m
aller
th
e
F
AR
an
d
FR
R
v
alu
es,
th
e
h
ig
h
er
th
e
r
eliab
ilit
y
o
f
th
e
class
if
icatio
n
alg
o
r
ith
m
b
ec
au
s
e
it
in
d
icate
s
th
e
f
ewe
r
f
alse
ac
ce
p
tan
ce
an
d
f
alse
r
eje
ctio
n
p
er
ce
n
tag
es.
Fo
r
ea
c
h
al
g
o
r
ith
m
,
th
e
lo
west
FAR
ar
e
0
.
0
0
6
%
f
o
r
KNN
an
d
L
in
ea
r
SVM,
an
d
0
.
0
1
%
f
o
r
r
an
d
o
m
f
o
r
est
.
Fo
r
ea
c
h
alg
o
r
ith
m
,
th
e
lo
west
FR
R
ar
e
1
% f
o
r
KNN,
0
.
9
% f
o
r
L
in
ea
r
SVM,
an
d
2
% f
o
r
r
a
n
d
o
m
f
o
r
est
.
4.
CO
NCLU
SI
O
N
I
n
th
is
s
tu
d
y
,
we
c
o
n
d
u
ct
an
ex
p
er
im
en
t
in
clo
u
d
-
b
ased
ar
ch
itectu
r
e
u
s
in
g
class
if
icatio
n
alg
o
r
ith
m
s
o
n
f
ac
e
id
en
tific
atio
n
with
C
NN
to
ex
tr
ac
t
f
ac
ial
f
ea
tu
r
e
f
r
o
m
im
a
g
e
an
d
p
r
o
d
u
ce
s
1
2
8
-
d
em
b
e
d
d
in
g
s
.
T
h
e
d
ata
s
o
u
r
ce
s
ca
m
e
f
r
o
m
th
r
e
e
d
if
f
er
en
t
d
atasets
,
n
am
ely
Face
s
9
4
,
Face
s
9
6
an
d
UM
I
S
T
f
ac
e
da
taset
with
d
if
f
er
en
t
ch
ar
ac
ter
is
tics
.
T
h
is
s
tu
d
y
u
s
es
f
o
u
r
s
tag
es
to
i
d
en
tify
f
ac
es,
n
am
ely
im
ag
e
en
h
a
n
c
em
en
t,
f
ac
e
d
etec
tio
n
,
f
ea
tu
r
e
ex
tr
ac
tio
n
,
an
d
class
if
icatio
n
u
s
in
g
th
e
KNN
alg
o
r
ith
m
,
lin
ea
r
SVM,
an
d
r
an
d
o
m
f
o
r
est
.
Fro
m
th
e
class
if
icatio
n
r
esu
lt
s
,
th
e
th
r
ee
alg
o
r
ith
m
s
h
av
e
d
ec
r
ea
s
ed
ac
cu
r
ac
y
o
n
th
e
UM
I
ST
f
ac
e
d
ataset
wh
ich
h
as
th
e
ch
ar
ac
ter
is
tics
o
f
i
m
ag
e
ca
p
tu
r
e
d
f
r
o
m
t
h
e
l
ef
t
s
id
e
to
th
e
r
ig
h
t
s
id
e.
Fro
m
th
e
r
esu
lt
o
f
th
is
s
tu
d
y
,
it
is
co
n
clu
d
ed
th
at
wit
h
th
e
p
r
o
p
o
s
ed
clo
u
d
-
b
ased
ar
ch
itectu
r
e,
th
e
b
est
ac
cu
r
ac
y
i
s
o
b
tain
ed
b
y
KNN
alg
o
r
ith
m
with
an
ac
cu
r
ac
y
o
f
9
9
%
f
o
r
th
e
Face
s
9
4
,
9
9
%
ac
cu
r
ac
y
f
o
r
Fac
es9
6
,
an
d
9
7
%
ac
cu
r
ac
y
f
o
r
UM
I
ST
f
ac
e
d
ataset
.
ACK
NO
WL
E
DG
E
M
E
NT
S
W
e
wo
u
ld
lik
e
to
e
x
p
r
ess
o
u
r
g
r
atitu
d
e
to
Ud
ay
an
a
Un
iv
er
s
ity
Dep
ar
tm
e
n
t
o
f
I
n
f
o
r
m
atio
n
T
ec
h
n
o
lo
g
y
f
o
r
p
r
o
v
i
d
in
g
s
u
f
f
icien
t f
ac
ilit
ies f
o
r
th
e
co
m
p
let
io
n
o
f
t
h
is
s
tu
d
y
.
RE
F
E
R
E
NC
E
S
[1
]
R.
Blan
c
o
-
G
o
n
z
a
lo
,
N.
P
o
h
,
R.
Wo
n
g
,
a
n
d
R.
S
a
n
c
h
e
z
-
Re
il
lo
,
“
Ti
m
e
e
v
o
lu
ti
o
n
o
f
fa
c
e
re
c
o
g
n
it
i
o
n
i
n
a
c
c
e
ss
ib
le
sc
e
n
a
rio
s,”
Hu
ma
n
-
c
e
n
tric
Co
mp
u
ti
n
g
a
n
d
I
n
fo
rm
a
ti
o
n
S
c
ien
c
e
s
,
v
o
l.
5
,
n
o
.
1
,
p
p
.
0
–
1
1
,
2
0
1
5
,
d
o
i:
1
0
.
1
1
8
6
/s1
3
6
7
3
-
0
1
5
-
0
0
4
3
-
0.
[2
]
C.
Li
,
W.
Wei
,
J.
Li
,
a
n
d
W
.
S
o
n
g
,
“
A
c
lo
u
d
-
b
a
se
d
m
o
n
it
o
ri
n
g
s
y
ste
m
v
ia
fa
c
e
re
c
o
g
n
i
ti
o
n
u
sin
g
G
a
b
o
r
a
n
d
CS
-
LBP
fe
a
tu
re
s,”
J
.
S
u
p
e
rc
o
mp
u
t.
,
v
o
l.
7
3
,
n
o
.
4
,
p
p
.
1
5
3
2
–
1
5
4
6
,
2
0
1
7
,
d
o
i:
1
0
.
1
0
0
7
/s
1
1
2
2
7
-
0
1
6
-
1
8
4
0
-
6.
[3
]
N.
Da
lal
a
n
d
B.
Tri
g
g
s,
“
Histo
g
ra
m
s
o
f
o
rien
ted
g
ra
d
ien
ts
fo
r
h
u
m
a
n
d
e
tec
ti
o
n
,
”
in
Pro
c
e
e
d
i
n
g
s
-
2
0
0
5
IEE
E
Co
mp
u
ter
S
o
c
iety
Co
n
fer
e
n
c
e
o
n
Co
mp
u
ter
Vi
sio
n
a
n
d
Pa
t
t
e
rn
Rec
o
g
n
it
i
o
n
,
CV
PR
2
0
0
5
,
v
o
l.
I,
2
0
0
5
,
p
p
.
8
8
6
–
8
9
3
,
d
o
i:
1
0
.
1
1
0
9
/CV
P
R.
2
0
0
5
.
1
7
7
.
[4
]
P
.
Tam
b
i,
S
.
Ja
in
,
a
n
d
D
.
K.
M
is
h
ra
,
Per
so
n
-
De
p
e
n
d
e
n
t
F
a
c
e
Rec
o
g
n
it
io
n
Us
i
n
g
Hist
o
g
r
a
m
o
f
Or
i
e
n
ted
Gr
a
d
ien
ts
(HO
G) a
n
d
Co
n
v
o
lu
ti
o
n
Ne
u
ra
l
Ne
two
rk
(CNN)
,
v
o
l.
8
7
0
,
S
p
ri
n
g
e
r
S
in
g
a
p
o
re
,
2
0
1
9
.
[5
]
L.
Arn
o
l
d
,
S
.
Re
b
e
c
c
h
i,
S
.
Ch
e
v
a
ll
ier,
a
n
d
H.
P
a
u
g
a
m
-
M
o
isy
,
“
A
n
in
tr
o
d
u
c
ti
o
n
to
d
e
e
p
lea
rn
i
n
g
,
”
ES
ANN
2
0
1
1
-
1
9
t
h
E
u
r.
S
y
mp
.
A
rtif
.
Ne
u
ra
l
Ne
t
wo
rk
s
,
p
p
.
4
7
7
–
4
8
8
,
2
0
1
1
,
d
o
i
:
1
0
.
1
2
0
1
/9
7
8
0
4
2
9
0
9
6
2
8
0
-
1
4
.
[6
]
I.
Na
m
a
tēv
s,
“
De
e
p
Co
n
v
o
lu
ti
o
n
a
l
Ne
u
ra
l
Ne
two
rk
s:
S
tru
c
tu
re
,
F
e
a
tu
re
Ex
trac
ti
o
n
a
n
d
Trai
n
in
g
,
”
In
f.
T
e
c
h
n
o
l.
M
a
n
a
g
.
S
c
i.
,
v
o
l
.
2
0
,
n
o
.
1
,
p
p
.
4
0
–
4
7
,
2
0
1
8
,
d
o
i:
1
0
.
1
5
1
5
/
it
m
s
-
2
0
1
7
-
0
0
0
7
.
[7
]
C.
Din
g
a
n
d
D.
Tao
,
“
R
o
b
u
st
F
a
c
e
Re
c
o
g
n
it
i
o
n
v
ia
M
u
lt
im
o
d
a
l
De
e
p
F
a
c
e
Re
p
re
se
n
tati
o
n
,
”
IEE
E
T
r
a
n
s.
M
u
lt
ime
d
.
,
v
o
l
.
1
7
,
n
o
.
1
1
,
p
p
.
2
0
4
9
–
2
0
5
8
,
2
0
1
5
,
d
o
i:
1
0
.
1
1
0
9
/
TM
M
.
2
0
1
5
.
2
4
7
7
0
4
2
.
[8
]
N.
H.
Ba
r
n
o
u
ti
,
S
.
S
.
M
.
Al
-
Da
b
b
a
g
h
,
a
n
d
M
.
H.
J.
Al
-
Ba
m
a
rn
i,
“
R
e
a
l
-
Ti
m
e
F
a
c
e
De
tec
ti
o
n
a
n
d
Re
c
o
g
n
it
io
n
Us
in
g
P
rin
c
ip
a
l
C
o
m
p
o
n
e
n
t
A
n
a
ly
sis
(
P
CA
)
–
Ba
c
k
P
r
o
p
a
g
a
ti
o
n
Ne
u
ra
l
Ne
two
rk
(
BP
NN
)
a
n
d
Ra
d
ial
Ba
sis
F
u
n
c
ti
o
n
(
RBF
)
,
”
J
o
u
r
n
a
l
o
f
T
h
e
o
re
ti
c
a
l
a
n
d
A
p
p
li
e
d
I
n
fo
rm
a
ti
o
n
T
e
c
h
n
o
l
o
g
y
,
v
o
l.
91
,
n
o
.
1
,
p
p
.
28
–
34
,
2
0
1
6
.
[9
]
N.
K.
A.
Wi
rd
ian
i,
P
.
Hrid
a
y
a
m
i
,
N.
P
.
A.
Wi
d
iari,
K.
D.
R
ism
a
wa
n
,
P
.
B.
Ca
n
d
ra
d
i
n
a
ta,
a
n
d
I
.
P
.
D.
Ja
y
a
n
th
a
,
“
F
a
c
e
Id
e
n
ti
fica
ti
o
n
Ba
se
d
o
n
K
-
Ne
a
re
st
Ne
ig
h
b
o
r,
”
S
c
i.
J
.
I
n
fo
rm
a
t
ics
,
v
o
l.
6
,
n
o
.
2
,
p
p
.
1
5
0
–
1
5
9
,
2
0
1
9
,
d
o
i:
1
0
.
1
5
2
9
4
/sj
i.
v
6
i2
.
1
9
5
0
3
.
[1
0
]
R.
S
e
n
th
i
l
k
u
m
a
r
a
n
d
R.
K.
G
n
a
n
a
m
u
rth
y
,
"
P
e
rfo
rm
a
n
c
e
imp
ro
v
e
m
e
n
t
in
c
las
sifica
ti
o
n
ra
te
o
f
a
p
p
e
a
ra
n
c
e
b
a
se
d
sta
ti
stica
l
fa
c
e
re
c
o
g
n
it
io
n
m
e
th
o
d
s
u
si
n
g
S
VM
c
las
sifier,"
2
0
1
7
4
t
h
I
n
ter
n
a
t
io
n
a
l
C
o
n
fer
e
n
c
e
o
n
Ad
v
a
n
c
e
d
Co
mp
u
t
in
g
a
n
d
Co
mm
u
n
ica
ti
o
n
S
y
ste
ms
(ICACCS
)
,
2
0
1
7
,
p
p
.
1
-
7
,
d
o
i:
1
0
.
1
1
0
9
/ICACCS
.
2
0
1
7
.
8
0
1
4
5
8
4
.
[1
1
]
H.
M
a
d
y
a
n
d
S
.
M
.
S
.
Hill
e
s,
"
F
a
c
e
re
c
o
g
n
it
io
n
a
n
d
d
e
tec
ti
o
n
u
si
n
g
Ra
n
d
o
m
f
o
re
st
a
n
d
c
o
m
b
in
a
ti
o
n
o
f
LBP
a
n
d
HO
G
fe
a
tu
re
s,"
2
0
1
8
In
ter
n
a
ti
o
n
a
l
Co
n
fer
e
n
c
e
o
n
S
ma
rt
Co
m
p
u
t
in
g
a
n
d
El
e
c
tro
n
ic
En
ter
p
rise
(I
CS
CEE
)
,
2
0
1
8
,
p
p
.
1
-
7
,
d
o
i:
1
0
.
1
1
0
9
/ICS
CE
E.
2
0
1
8
.
8
5
3
8
3
7
7
.
[1
2
]
D.
B.
G
r
a
h
a
m
a
n
d
N.
M
.
Alli
n
so
n
,
“
Ch
a
ra
c
terisin
g
Virtu
a
l
Ei
g
e
n
si
g
n
a
tu
re
s fo
r
G
e
n
e
ra
l
P
u
rp
o
se
F
a
c
e
Re
c
o
g
n
it
io
n
,
”
in
F
a
c
e
Rec
o
g
n
it
i
o
n
:
Fro
m
T
h
e
o
ry
to
A
p
p
li
c
a
ti
o
n
s
,
H.
Wec
h
sle
r,
P
.
J.
P
h
il
li
p
s,
V.
Br
u
c
e
,
F
.
F
.
S
o
u
li
é
,
a
n
d
T
.
S
.
Hu
a
n
g
,
Ed
s.
Be
rli
n
,
He
id
e
lb
e
rg
:
S
p
rin
g
e
r
Be
rli
n
He
id
e
lb
e
r
g
,
p
p
.
4
4
6
–
4
5
6
,
1
9
9
8
.
[1
3
]
D.
Ho
n
d
a
n
d
L.
S
p
a
c
e
k
,
“
Distin
c
ti
v
e
d
e
sc
rip
ti
o
n
s
f
o
r
fa
c
e
p
ro
c
e
ss
in
g
,
”
i
n
Pr
o
c
e
e
d
in
g
s
o
f
t
h
e
8
t
h
B
riti
sh
M
a
c
h
i
n
e
Vi
sio
n
Co
n
fer
e
n
c
e
BM
VC9
7
,
C
o
l
c
h
e
ste
r,
En
g
l
a
n
d
,
1
9
9
7
,
p
p
.
3
2
0
–
3
2
9
.
Evaluation Warning : The document was created with Spire.PDF for Python.
I
SS
N
:
2
5
0
2
-
4
7
5
2
I
n
d
o
n
esian
J
E
lec
E
n
g
&
C
o
m
p
Sci,
Vo
l.
23
,
No
.
2
,
Au
g
u
s
t
20
21
:
811
-
8
2
0
820
[1
4
]
P
.
Ja
n
a
n
i
,
J.
P
re
m
a
lad
h
a
,
a
n
d
K.
S
.
Ra
v
ic
h
a
n
d
ra
n
,
“
Im
a
g
e
E
n
h
a
n
c
e
m
e
n
t
Tec
h
n
i
q
u
e
s:
A
S
t
u
d
y
,
”
I
n
d
ia
n
J
o
u
rn
a
l
o
f
S
c
ien
c
e
a
n
d
T
e
c
h
n
o
l
o
g
y
,
v
o
l.
8
,
n
o
.
2
2
,
p
p
.
8
3
–
8
9
,
S
e
p
.
2
0
1
5
,
d
o
i:
1
0
.
1
7
4
8
5
/i
jst/
2
0
1
5
/
v
8
i2
2
/7
9
3
1
8
.
[1
5
]
I.
Ag
u
sti
n
a
,
F
.
Na
sir,
a
n
d
A.
S
e
ti
a
wa
n
,
“
Th
e
Im
p
lem
e
n
tatio
n
o
f
I
m
a
g
e
S
m
o
o
th
i
n
g
to
Re
d
u
c
e
No
is
e
u
sin
g
G
a
u
ss
ian
F
il
ter
,
”
In
ter
n
a
ti
o
n
a
l
J
o
u
r
n
a
l
o
f
Co
mp
u
ter
Ap
p
li
c
a
t
io
n
s
,
v
o
l.
1
7
7
,
n
o
.
5
,
p
p
.
15
–
19
,
2
0
1
7
,
d
o
i:
1
0
.
5
1
2
0
/i
jca
2
0
1
7
9
1
5
7
5
5
.
[1
6
]
S
.
Ema
m
i
a
n
d
V.
P
.
S
u
c
iu
,
“
F
a
c
ial
Re
c
o
g
n
it
i
o
n
u
si
n
g
Op
e
n
CV,”
J
.
M
o
b
il
e
,
Emb
e
d
.
Distrib
.
S
y
st.
,
v
o
l.
4
,
n
o
.
1
,
p
p
.
3
8
–
4
3
,
2
0
1
2
.
[1
7
]
A.
Ad
o
u
a
n
i
,
W.
M
.
Be
n
He
n
ia
a
n
d
Z.
Lac
h
iri
,
"
C
o
m
p
a
riso
n
o
f
Ha
a
r
-
li
k
e
,
HO
G
a
n
d
LB
P
a
p
p
r
o
a
c
h
e
s
fo
r
fa
c
e
d
e
tec
ti
o
n
in
v
id
e
o
se
q
u
e
n
c
e
s,"
2
0
1
9
1
6
th
In
ter
n
a
ti
o
n
a
l
M
u
lt
i
-
Co
n
fer
e
n
c
e
o
n
S
y
ste
ms
,
S
i
g
n
a
ls
&
De
v
ice
s
(S
S
D)
,
2
0
1
9
,
p
p
.
2
6
6
-
2
7
1
,
d
o
i:
1
0
.
1
1
0
9
/
S
S
D.2
0
1
9
.
8
8
9
3
2
1
4
.
[1
8
]
D.
E.
Ki
n
g
,
“
Dlib
-
m
l:
A
M
a
c
h
in
e
Lea
rn
in
g
To
o
lk
it
,
”
J
.
M
a
c
h
.
L
e
a
rn
.
Res
.
,
v
o
l.
1
0
,
n
o
.
1
0
,
p
p
.
1
7
5
5
–
1
7
5
8
,
2
0
0
9
,
d
o
i:
1
0
.
1
1
4
5
/1
5
7
7
0
6
9
.
1
7
5
5
8
4
3
.
[1
9
]
P
.
Ko
rfiatis,
T.
L.
Kli
n
e
,
D.
H.
Lac
h
a
n
c
e
,
I.
F
.
P
a
rn
e
y
,
J.
C.
Bu
c
k
n
e
r,
a
n
d
B.
J.
Eri
c
k
so
n
,
“
Re
sid
u
a
l
De
e
p
Co
n
v
o
l
u
ti
o
n
a
l
Ne
u
ra
l
Ne
two
rk
P
r
e
d
icts M
G
M
T
M
e
th
y
lati
o
n
S
tat
u
s
,
”
J
.
Dig
it
.
Ima
g
in
g
,
v
o
l
.
3
0
,
n
o
.
5
,
p
p
.
6
2
2
–
6
2
8
,
2
0
1
7
,
d
o
i:
1
0
.
1
0
0
7
/s
1
0
2
7
8
-
0
1
7
-
0
0
0
9
-
z.
[2
0
]
F
.
S
c
h
ro
ff,
D.
Ka
len
ich
e
n
k
o
,
a
n
d
J.
P
h
il
b
in
,
“
F
a
c
e
Ne
t:
A
u
n
ifi
e
d
e
m
b
e
d
d
in
g
f
o
r
fa
c
e
re
c
o
g
n
it
i
o
n
a
n
d
c
lu
ste
ri
n
g
,
”
2
0
1
5
I
EE
E
C
o
n
fer
e
n
c
e
o
n
C
o
mp
u
ter
Vi
sio
n
a
n
d
P
a
tt
e
rn
Rec
o
g
n
it
i
o
n
(C
VP
R)
,
v
o
l.
0
7
-
12
-
J
u
n
e
,
2
0
1
5
,
p
p
.
8
1
5
–
8
2
3
,
d
o
i:
1
0
.
1
1
0
9
/CVP
R.
2
0
1
5
.
7
2
9
8
6
8
2
.
[2
1
]
S
.
S
h
a
rm
a
,
K.
S
h
a
n
m
u
g
a
su
n
d
a
r
a
m
a
n
d
S
.
K.
Ra
m
a
sa
m
y
,
"
F
AREC
—
CNN
b
a
se
d
e
fficie
n
t
f
a
c
e
re
c
o
g
n
it
i
o
n
tec
h
n
iq
u
e
u
si
n
g
Dli
b
,
"
2
0
1
6
In
ter
n
a
ti
o
n
a
l
Co
n
fer
e
n
c
e
o
n
Ad
v
a
n
c
e
d
C
o
mm
u
n
ic
a
ti
o
n
C
o
n
tro
l
a
n
d
Co
mp
u
ti
n
g
T
e
c
h
n
o
l
o
g
ies
(ICACCCT
),
2
0
1
6
,
p
p
.
1
9
2
-
1
9
5
,
d
o
i:
1
0
.
1
1
0
9
/ICAC
CCT.
2
0
1
6
.
7
8
3
1
6
2
8
.
[2
2
]
R.
Du
d
a
,
P
.
Ha
rt,
a
n
d
D
.
S
t
o
rk
,
P
a
tt
e
rn
Cl
a
ss
if
ica
ti
o
n
,
Jo
h
n
Wi
ll
e
y
&
S
o
n
s,
2
0
1
2
.
[2
3
]
F
.
M
a
h
m
u
d
,
B
.
Isla
m
,
A.
Ho
ss
a
in
a
n
d
P
.
B.
G
o
a
la,
"
F
a
c
ial
Re
g
io
n
S
e
g
m
e
n
tatio
n
Ba
se
d
Emo
ti
o
n
Re
c
o
g
n
i
ti
o
n
Us
in
g
K
-
Ne
a
re
st
N
e
ig
h
b
o
rs,
"
2
0
1
8
In
t
e
rn
a
ti
o
n
a
l
C
o
n
fer
e
n
c
e
o
n
I
n
n
o
v
a
ti
o
n
in
E
n
g
in
e
e
rin
g
a
n
d
T
e
c
h
n
o
lo
g
y
(ICIE
T
)
,
2
0
1
8
,
p
p
.
1
-
5
,
d
o
i:
1
0
.
1
1
0
9
/CI
ET
.
2
0
1
8
.
8
6
6
0
9
0
0
.
[2
4
]
F
.
P
e
d
re
g
o
sa
e
t
a
l.
,
“
S
c
i
k
it
-
lea
rn
:
M
a
c
h
in
e
Lea
rn
i
n
g
in
P
y
th
o
n
,
”
J
o
u
rn
a
l
o
f
M
a
c
h
in
e
L
e
a
rn
i
n
g
Res
e
a
rc
h
,
v
o
l.
1
2
,
p
p
.
2
8
2
5
–
2
8
3
0
,
2
0
1
1
.
[2
5
]
W.
S
.
No
b
le,
“
Wh
a
t
is
a
su
p
p
o
rt
v
e
c
to
r
m
a
c
h
in
e
?
,
”
Na
t.
Bi
o
tec
h
n
o
l
.
,
v
o
l.
2
4
,
n
o
.
1
2
,
p
p
.
1
5
6
5
–
1
5
6
7
,
2
0
0
6
,
d
o
i:
1
0
.
1
0
3
8
/n
b
t
1
2
0
6
-
1
5
6
5
.
[2
6
]
Y.
Ch
a
n
g
a
n
d
C.
Li
n
,
“
F
e
a
tu
re
R
a
n
k
in
g
Us
i
n
g
Li
n
e
a
r
S
VM,
”
Fea
t
u
r.
R
a
n
k
.
Us
in
g
L
in
e
a
r
S
V
M
,
v
o
l
.
3
,
p
p
.
5
3
–
6
4
,
2
0
0
8
.
[2
7
]
R.
A.
Nu
g
ra
h
a
e
n
i
a
n
d
K.
M
u
ti
jars
a
,
"
Co
m
p
a
ra
ti
v
e
a
n
a
ly
sis
o
f
m
a
c
h
in
e
lea
rn
i
n
g
KN
N,
S
VM,
a
n
d
ra
n
d
o
m
f
o
re
sts
a
lg
o
rit
h
m
fo
r
fa
c
ial
e
x
p
re
ss
io
n
c
las
sifica
ti
o
n
,
"
2
0
1
6
In
ter
n
a
ti
o
n
a
l
S
e
min
a
r
o
n
Ap
p
li
c
a
t
io
n
fo
r
T
e
c
h
n
o
l
o
g
y
o
f
In
fo
rm
a
t
io
n
a
n
d
Co
mm
u
n
ica
ti
o
n
(IS
e
ma
n
ti
c
),
2
0
1
6
,
p
p
.
1
6
3
-
1
6
8
,
d
o
i:
1
0
.
1
1
0
9
/I
S
EM
AN
TIC.
2
0
1
6
.
7
8
7
3
8
3
1
.
Evaluation Warning : The document was created with Spire.PDF for Python.