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
.
3
,
Sep
tem
b
er
2
0
2
1
,
p
p
.
1
6
9
9
~
1
7
0
7
I
SS
N:
2
5
0
2
-
4
7
5
2
,
DOI
: 1
0
.
1
1
5
9
1
/ijeecs.v
23
.i
3
.
pp
1
6
9
9
-
1
7
0
7
1699
J
o
ur
na
l ho
m
ep
a
g
e
:
h
ttp
:
//ij
ee
cs.ia
esco
r
e.
co
m
An ef
fec
tive face
recog
nition
met
h
o
d using
guided i
ma
g
e f
ilter
a
nd conv
o
lutiona
l neural ne
twork
Ya
lla
m
a
nd
a
ia
h S
.
,
P
urna
cha
nd
N
.
S
c
h
o
o
l
o
f
El
e
c
tro
n
ics
En
g
in
e
e
rin
g
,
VIT
-
AP
U
n
iv
e
rsit
y
,
Am
a
ra
v
a
ti
,
An
d
h
ra
P
ra
d
e
sh
,
In
d
ia
Art
icle
I
nfo
AB
S
T
RAC
T
A
r
ticle
his
to
r
y:
R
ec
eiv
ed
Ju
n
1
9
,
2
0
2
1
R
ev
is
ed
J
u
l
23
,
2
0
2
1
Acc
ep
ted
Au
g
1
0
,
2
0
2
1
In
t
h
e
a
re
a
o
f
c
o
m
p
u
ter
v
isio
n
,
fa
c
e
re
c
o
g
n
it
i
o
n
is
a
c
h
a
ll
e
n
g
i
n
g
tas
k
b
e
c
a
u
se
o
f
t
h
e
p
o
se
,
fa
c
ial
e
x
p
re
ss
io
n
,
a
n
d
il
l
u
m
in
a
ti
o
n
v
a
riati
o
n
s.
Th
e
p
e
rfo
rm
a
n
c
e
o
f
fa
c
e
re
c
o
g
n
it
io
n
sy
ste
m
s
re
d
u
c
e
s
in
a
n
u
n
c
o
n
stra
in
e
d
e
n
v
ir
o
n
m
e
n
t.
I
n
t
h
is
wo
rk
,
a
n
e
w
fa
c
e
re
c
o
g
n
i
ti
o
n
a
p
p
r
o
a
c
h
is
p
r
o
p
o
se
d
u
si
n
g
a
g
u
i
d
e
d
ima
g
e
fil
ter,
a
n
d
a
c
o
n
v
o
lu
t
io
n
a
l
n
e
u
ra
l
n
e
two
rk
(CNN
).
T
h
e
g
u
id
e
d
ima
g
e
fil
ter
is
a
sm
o
o
th
i
n
g
o
p
e
ra
to
r
a
n
d
p
e
rfo
r
m
s
we
ll
n
e
a
r
th
e
e
d
g
e
s.
In
it
ially
,
th
e
Vio
la
-
Jo
n
e
s
a
lg
o
rit
h
m
is
u
se
d
t
o
d
e
tec
t
th
e
fa
c
e
re
g
i
o
n
a
n
d
t
h
e
n
sm
o
o
th
e
n
e
d
b
y
a
g
u
i
d
e
d
ima
g
e
fil
ter.
Late
r
t
h
e
p
r
o
p
o
se
d
CNN
is
u
se
d
to
e
x
trac
t
t
h
e
fe
a
tu
re
s
a
n
d
re
c
o
g
n
ize
th
e
fa
c
e
s.
Th
e
e
x
p
e
rime
n
ts
we
re
p
e
rfo
rm
e
d
o
n
fa
c
e
d
a
tab
a
se
s
li
k
e
ORL,
JA
F
F
E,
a
n
d
YA
LE
a
n
d
a
tt
a
in
e
d
a
re
c
o
g
n
it
i
o
n
ra
te
o
f
9
8
.
3
3
%
,
9
9
.
5
3
%
,
a
n
d
9
8
.
6
5
%
re
sp
e
c
ti
v
e
l
y
.
T
h
e
e
x
p
e
rime
n
tal
re
su
lt
s
sh
o
w
th
a
t
t
h
e
su
g
g
e
ste
d
fa
c
e
re
c
o
g
n
it
i
o
n
m
e
th
o
d
a
tt
a
in
s
g
o
o
d
re
su
lt
s
th
a
n
so
m
e
o
f
th
e
sta
te
-
of
-
th
e
-
a
rt
tec
h
n
i
q
u
e
s.
K
ey
w
o
r
d
s
:
C
o
m
p
u
ter
v
is
io
n
C
o
n
v
o
lu
tio
n
al
n
eu
r
al
n
etwo
r
k
Face
r
ec
o
g
n
itio
n
Gu
id
ed
im
ag
e
f
ilter
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
:
Pu
r
n
ac
h
an
d
N
.
Sc
h
o
o
l o
f
E
lectr
o
n
ics E
n
g
i
n
ee
r
in
g
V
IT
-
AP U
n
iv
er
s
ity
Am
ar
av
ati
,
An
d
h
r
a
Pra
d
esh
5
2
2
2
3
7
,
I
n
d
ia
E
m
ail: c
h
an
d
u
in
ec
e@
g
m
ail.
co
m
1.
I
NT
RO
D
UCT
I
O
N
T
h
e
f
ac
e
is
an
im
p
o
r
tan
t
b
io
lo
g
ical
tr
ait
th
at
d
if
f
er
en
ti
ates
an
in
d
i
v
id
u
al
f
r
o
m
o
t
h
er
s
.
Face
r
ec
o
g
n
itio
n
is
a
p
r
o
ce
s
s
o
f
e
x
tr
ac
tin
g
f
ea
tu
r
es
an
d
id
e
n
tif
y
in
g
in
d
iv
i
d
u
als,
an
d
it
is
an
ef
f
icien
t
ap
p
r
o
ac
h
am
o
n
g
m
a
n
y
b
io
m
etr
ics
b
ec
au
s
e
o
f
its
f
u
ll
au
to
m
atio
n
an
d
u
n
iq
u
e
n
ess
[
1
]
.
Pre
p
r
o
ce
s
s
in
g
,
f
ea
tu
r
e
ex
tr
ac
tio
n
,
an
d
class
if
icatio
n
ar
e
th
e
th
r
ee
s
tep
s
in
th
e
f
ac
e
r
ec
o
g
n
itio
n
task
.
T
h
e
ex
tr
ac
tio
n
o
f
f
ea
tu
r
es
an
d
th
e
co
n
s
tr
u
ctio
n
o
f
th
e
p
r
o
p
er
class
if
ier
p
lay
a
m
ajo
r
r
o
le
in
t
h
e
f
ac
e
r
ec
o
g
n
itio
n
p
r
o
ce
s
s
.
Face
r
ec
o
g
n
itio
n
h
as
b
ec
o
m
e
p
o
p
u
lar
b
ec
au
s
e
o
f
its
u
tili
ty
in
d
if
f
er
e
n
t
ar
ea
s
s
u
ch
as
c
o
m
m
u
n
icatio
n
,
f
ile
m
an
ag
e
m
en
t,
h
u
m
an
-
c
o
m
p
u
ter
in
ter
ac
tio
n
,
s
u
r
v
eillan
ce
,
s
ec
u
r
ity
,
an
d
law
en
f
o
r
ce
m
en
t.
I
n
p
ar
ticu
lar
,
f
ac
e
r
ec
o
g
n
itio
n
is
wid
ely
u
s
e
d
to
id
e
n
tify
m
is
s
in
g
b
ab
ies,
d
etec
t
p
ass
p
o
r
t
f
r
a
u
d
s
,
u
n
lo
ck
in
g
ap
p
s
o
n
m
o
b
ile
p
h
o
n
es,
an
d
s
to
p
b
lack
lis
ted
p
er
s
o
n
s
in
r
estau
r
an
ts
.
T
h
e
ef
f
icien
cy
o
f
th
e
f
ac
e
r
ec
o
g
n
iti
o
n
s
y
s
tem
s
r
ed
u
ce
s
b
ec
au
s
e
o
f
v
ar
iatio
n
s
in
illu
m
in
atio
n
s
,
p
o
s
es,
ex
p
r
ess
io
n
s
o
f
th
e
p
er
s
o
n
.
I
n
a
n
u
n
co
n
tr
o
ll
ed
en
v
ir
o
n
m
en
t,
th
e
f
ac
e
im
ag
es
ar
e
af
f
ec
ted
b
y
d
if
f
er
en
t
illu
m
in
atio
n
s
an
d
n
o
is
es.
A
g
u
id
ed
im
ag
e
f
ilte
r
h
as
a
v
ar
iety
o
f
ap
p
licatio
n
s
s
u
ch
as
n
o
is
e
r
ed
u
ctio
n
,
h
az
e
r
e
m
o
v
al,
en
h
a
n
ce
m
en
t/s
m
o
o
th
in
g
[
2
]
.
R
ec
e
n
tly
co
n
v
o
l
u
tio
n
al
n
eu
r
al
n
etwo
r
k
s
p
r
o
d
u
cin
g
g
o
o
d
r
esu
lts
in
th
e
ca
s
e
o
f
im
ag
e
class
if
icatio
n
.
T
h
e
tr
ad
eo
f
f
wh
en
m
o
v
in
g
f
r
o
m
tr
ad
itio
n
al
ap
p
r
o
ac
h
es
t
o
d
ee
p
lear
n
in
g
ap
p
r
o
ac
h
es
is
tr
ain
in
g
tim
e
i.e
.
in
itially
it
tak
es
a
lo
n
g
tim
e
f
o
r
tr
ain
in
g
th
e
d
ata
to
th
e
C
NN,
b
u
t
th
e
class
if
icatio
n
ac
cu
r
ac
y
will
b
e
h
ig
h
co
m
p
ar
e
d
to
ea
r
lier
m
eth
o
d
s
.
T
o
d
ea
l
with
th
e
v
ar
io
u
s
illu
m
in
atio
n
s
,
p
o
s
es,
an
d
ex
p
r
ess
io
n
s
o
f
th
e
p
er
s
o
n
,
th
is
p
a
p
er
p
r
o
p
o
s
es
a
n
ew
f
ac
e
r
ec
o
g
n
itio
n
m
eth
o
d
u
s
in
g
a
g
u
id
ed
im
ag
e
f
ilter
a
n
d
c
o
n
v
o
lu
t
io
n
al
n
eu
r
al
n
etwo
r
k
(
C
NN)
.
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
.
3
,
Sep
tem
b
er
2
0
2
1
:
1
6
9
9
-
1
7
0
7
1700
T
h
e
s
tr
u
ctu
r
e
o
f
th
e
p
a
p
er
i
s
p
lan
n
ed
as
f
o
llo
ws:
R
elate
d
wo
r
k
is
d
is
cu
s
s
ed
,
in
Sectio
n
2
.
I
n
Sectio
n
3
,
th
e
s
u
g
g
ested
f
ac
e
r
ec
o
g
n
itio
n
m
et
h
o
d
is
ex
p
lain
ed
.
T
h
e
ex
p
er
im
e
n
tal
o
u
tco
m
es
ar
e
d
em
o
n
s
tr
ated
in
Sectio
n
4
.
T
h
e
co
n
cl
u
s
io
n
o
f
th
e
p
ap
e
r
is
p
r
esen
ted
in
Sectio
n
5.
2.
RE
L
AT
E
D
WO
RK
Featu
r
e
ex
tr
ac
tio
n
an
d
t
h
e
co
n
s
tr
u
ctio
n
o
f
th
e
class
if
ier
p
l
ay
a
cr
u
cial
r
o
le
in
th
e
f
ac
e
r
ec
o
g
n
itio
n
p
r
o
ce
s
s
.
T
h
e
p
r
in
ci
p
al
co
m
p
o
n
en
t
an
al
y
s
is
(
PC
A)
is
a
p
r
o
m
in
en
t
m
eth
o
d
f
o
r
f
ea
tu
r
e
e
x
tr
ac
tio
n
.
Kir
b
y
a
n
d
Siro
v
ich
u
s
ed
p
r
in
cip
al
co
m
p
o
n
en
ts
to
r
ep
r
esen
t
h
u
m
an
f
a
ce
s
[
3
]
.
T
u
r
k
et
a
l.
u
tili
ze
d
th
e
p
lan
o
f
R
ef
.
3
f
o
r
f
ac
e
d
etec
tio
n
an
d
r
ec
o
g
n
itio
n
[
4
]
.
PC
A
r
ed
u
ce
s
th
e
d
im
e
n
s
io
n
an
d
elim
in
ates
co
r
r
elatio
n
,
h
o
wev
er
,
it
is
n
o
t
ap
p
r
o
p
r
iate
f
o
r
class
if
icatio
n
[
5
]
,
[
6
]
.
Me
ed
en
i
y
a
an
d
R
atn
av
ee
r
a
p
r
o
p
o
s
ed
a
n
en
h
an
ce
d
f
ac
e
r
ec
o
g
n
itio
n
m
eth
o
d
t
h
r
o
u
g
h
t
h
e
v
ar
iatio
n
o
f
PC
A,
in
wh
ich
th
e
a
u
th
o
r
s
p
er
f
o
r
m
ed
t
h
e
ec
o
n
o
m
ic
s
ize
s
in
g
u
lar
v
al
u
e
d
ec
o
m
p
o
s
itio
n
to
g
en
e
r
ate
a
u
n
itar
y
m
atr
ix
[
7
]
.
L
in
ea
r
d
is
cr
im
in
an
t a
n
al
y
s
is
(
L
DA)
is
an
em
in
en
t d
im
en
s
io
n
ality
r
ed
u
ctio
n
m
eth
o
d
,
b
u
t it
f
ails
wh
en
th
e
co
u
n
t
o
f
tr
ain
in
g
s
am
p
le
s
is
less
co
m
p
ar
ed
to
th
e
co
u
n
t
o
f
d
im
en
s
io
n
s
o
f
th
e
f
ea
tu
r
e
s
p
ac
e
[
8
]
.
T
o
o
v
er
co
m
e
th
is
p
r
o
b
lem
B
elh
u
m
er
et
a
l.
p
r
o
p
o
s
ed
th
e
f
is
h
e
r
f
ac
es
m
et
h
o
d
[
9
]
.
Yu
n
a
n
d
R
u
an
p
r
o
p
o
u
n
d
ed
en
h
an
ce
d
f
is
h
er
’
s
lin
ea
r
d
is
cr
im
in
an
t
(
E
FLD)
m
eth
o
d
an
d
it
o
u
tp
er
f
o
r
m
s
th
e
ea
r
lier
alg
o
r
i
th
m
s
[
1
0
]
.
Z
h
o
u
et
a
l.
s
u
g
g
ested
a
f
ac
e
r
ec
o
g
n
iti
o
n
m
eth
o
d
b
y
co
m
b
in
i
n
g
a
No
n
-
Neg
ativ
e
Ma
tr
ix
Facto
r
iza
tio
n
with
a
R
ad
ia
l
B
asi
s
Fu
n
ctio
n
class
if
ier
[
1
1
]
.
Ab
u
s
h
am
et
a
l.
d
em
o
n
s
tr
a
ted
an
ap
p
r
o
ac
h
f
o
r
f
ac
e
r
ec
o
g
n
itio
n
b
y
f
u
s
in
g
L
o
ca
lly
L
in
ea
r
E
m
b
ed
d
in
g
a
n
d
PC
A
[
1
2
]
.
L
o
ca
l
b
in
ar
y
p
atter
n
s
(
L
B
P)
[
1
3
]
an
d
lo
ca
l
p
h
ase
q
u
an
tizatio
n
(
L
PQ)
[
1
4
]
,
[
1
5
]
h
av
e
attain
ed
g
o
o
d
f
ac
e
r
ec
o
g
n
itio
n
im
p
lem
en
tatio
n
in
co
n
s
tr
ain
ed
en
v
ir
o
n
m
en
ts
.
T
h
e
p
er
f
o
r
m
an
ce
o
f
th
ese
h
an
d
cr
af
ted
f
ea
tu
r
es
r
ed
u
ce
s
co
n
s
id
er
a
b
ly
in
an
u
n
c
o
n
s
tr
ain
ed
en
v
ir
o
n
m
en
t.
Z
h
o
u
et
a
l.
in
tr
o
d
u
ce
d
a
f
ac
e
r
ec
o
g
n
itio
n
m
eth
o
d
u
s
in
g
PC
A
an
d
L
D
A
[
1
6
]
.
Dai
et
a
l.
m
an
i
f
ested
a
d
ec
o
r
r
elate
d
2
D
-
f
ee
d
-
f
o
r
war
d
n
eu
r
al
n
etwo
r
k
(
2
D
-
FNN)
en
s
em
b
le
with
r
a
n
d
o
m
weig
h
ts
[
1
7
]
.
T
h
e
f
ea
t
u
r
e
-
lev
el
f
u
s
io
n
o
f
lo
ca
l
b
in
ar
y
p
atter
n
s
an
d
c
o
ef
f
icien
t
e
n
h
an
ce
m
en
t
alg
o
r
ith
m
s
o
n
co
n
to
u
r
let
-
su
b
b
an
d
s
m
ad
e
a
r
o
b
u
s
t
ex
p
r
ess
io
n
in
v
ar
ia
n
t
f
ac
e
r
ec
o
g
n
itio
n
s
y
s
tem
[
1
8
]
.
Kh
a
n
e
t
a
l.
p
r
o
p
o
s
ed
a
s
y
s
tem
th
at
ca
n
r
ec
o
g
n
ize
f
ac
es
u
n
d
er
v
a
r
y
in
g
e
x
p
r
ess
io
n
s
an
d
illu
m
in
atio
n
u
s
in
g
p
ar
ticle
s
war
m
o
p
tim
izatio
n
(
PS
O)
[
1
9
]
.
T
ai
et
a
l.
p
r
o
p
o
s
ed
th
e
o
r
th
o
g
o
n
al
p
r
o
cr
u
s
tes
p
r
o
b
lem
(
OOP)
as
a
f
r
am
ewo
r
k
to
p
o
s
e
ch
an
g
es
in
f
ac
e
im
a
g
es
[
2
0
]
.
L
i
et
a
l.
in
tr
o
d
u
ce
d
a
p
r
o
jectiv
e
lo
w
-
r
an
k
d
escr
ip
tio
n
m
eth
o
d
f
o
r
f
a
ce
r
ec
o
g
n
itio
n
[
2
1
]
.
C
h
e
n
Y
et
a
l.
a
d
d
r
ess
ed
th
e
p
r
o
b
lem
o
f
m
u
lti
-
p
o
s
e
class
if
icatio
n
u
s
in
g
2
D
-
Gab
o
r
f
ea
t
u
r
es
an
d
th
e
Dee
p
B
elief
Ne
ts
[
2
2
]
.
Yin
et
a
l.
s
u
g
g
ested
m
u
lti
-
task
lear
n
in
g
f
o
r
r
ec
o
g
n
izin
g
f
ac
es
with
th
e
p
o
s
e
an
d
ex
p
r
ess
io
n
esti
m
atio
n
as
th
e
s
id
e
task
s
[
2
3
]
.
Din
g
et
a
l.
in
tr
o
d
u
ce
d
an
im
p
r
o
v
ed
h
u
m
an
ac
tiv
it
y
r
ec
o
g
n
itio
n
s
y
s
tem
b
ased
o
n
a
r
an
d
o
m
f
o
r
est
class
if
ier
[
2
4
]
.
L
iao
et
a
l.
s
u
g
g
ested
a
n
o
v
el
clu
s
ter
m
u
ltip
le
k
er
n
el
lear
n
i
n
g
alg
o
r
ith
m
f
o
r
r
ec
o
g
n
izin
g
th
e
o
il
p
ain
ter
s
[
2
5
]
.
Mu
q
ee
t
et
a
l.
p
r
esen
ted
a
f
ac
e
r
ec
o
g
n
itio
n
m
eth
o
d
b
y
u
tili
zin
g
d
ir
ec
tio
n
al
wav
elet
tr
an
s
f
o
r
m
an
d
lo
ca
l b
i
n
ar
y
p
atter
n
s
[
2
6
]
.
I
n
r
ec
en
t
y
ea
r
s
C
NN
m
eth
o
d
s
h
av
e
g
r
a
b
b
ed
s
u
b
s
tan
tial
atten
tiv
en
ess
in
f
ac
e
r
ec
o
g
n
itio
n
.
T
h
e
C
NNs
co
n
s
id
er
ab
ly
en
h
a
n
ce
s
th
e
m
o
d
el
g
en
e
r
atio
n
a
b
ilit
y
b
y
estab
lis
h
in
g
ef
f
ec
tiv
e
r
eg
u
lar
izatio
n
s
tr
ateg
ies
s
u
ch
as
d
r
o
p
o
u
t
[
2
7
]
.
T
h
e
r
esear
ch
g
r
o
u
p
at
Face
b
o
o
k
d
e
v
elo
p
ed
a
d
e
ep
lear
n
in
g
f
ac
ial
r
ec
o
g
n
it
io
n
s
y
s
tem
n
am
e
d
Dee
p
Face
[
2
8
]
.
W
u
et
a
l.
d
is
co
v
er
ed
th
e
co
r
r
elatio
n
s
am
o
n
g
t
h
e
s
u
s
tain
ab
le
d
ev
el
o
p
m
en
t
g
o
als
an
d
co
m
m
u
n
icatio
n
tech
n
o
lo
g
ies
[
2
9
]
.
Var
io
u
s
p
atter
n
r
ec
o
g
n
i
tio
n
alg
o
r
ith
m
s
f
o
r
h
u
m
a
n
a
ctiv
ity
r
ec
o
g
n
itio
n
wer
e
r
ev
iewe
d
an
d
d
is
cu
s
s
ed
in
[
3
0
]
.
L
in
et
a
l.
p
r
o
p
o
u
n
d
e
d
a
n
ew
r
o
b
u
s
t
d
ictio
n
a
r
y
lea
r
n
in
g
ap
p
r
o
ac
h
f
o
r
f
ac
e
r
ec
o
g
n
itio
n
[
3
1
]
.
Geo
r
g
e
l
et
a
l.
u
s
ed
d
ee
p
s
tack
ed
d
e
-
n
o
is
in
g
an
d
s
p
ar
s
e
au
to
-
e
n
co
d
er
s
(
DSDSA)
f
o
r
f
ac
e
r
ec
o
g
n
itio
n
[
3
2
]
.
I
n
th
is
wo
r
k
,
a
n
ew
f
ac
e
r
ec
o
g
n
itio
n
tech
n
iq
u
e
is
p
r
o
p
o
s
ed
u
s
in
g
a
g
u
id
ed
im
ag
e
f
ilter
an
d
a
co
n
v
o
l
u
tio
n
al
n
e
u
r
al
n
et
wo
r
k
.
3.
P
RO
P
O
SE
D
WO
RK
T
h
is
wo
r
k
p
r
o
p
o
s
es a
n
ew
ap
p
r
o
ac
h
f
o
r
f
ac
e
r
ec
o
g
n
itio
n
u
s
in
g
a
g
u
id
ed
im
a
g
e
f
ilter
an
d
a
C
NN.
T
h
e
p
r
o
p
o
s
ed
m
eth
o
d
co
n
s
is
ts
o
f
t
h
e
f
o
llo
win
g
s
tep
s
:
f
ac
e
d
etec
tio
n
,
im
ag
e
r
esizin
g
,
ap
p
l
y
in
g
g
u
id
ed
im
a
g
e
f
ilter
o
n
th
e
r
esized
im
ag
e
,
ex
tr
ac
ti
n
g
f
ea
tu
r
es,
an
d
r
ec
o
g
n
izin
g
f
ac
es
with
th
e
h
elp
o
f
th
e
p
r
o
p
o
s
ed
C
NN.
I
n
itially
,
th
e
f
ac
e
r
eg
i
o
n
is
ex
tr
ac
ted
u
s
in
g
th
e
Vio
la
-
J
o
n
es
alg
o
r
ith
m
an
d
r
esized
to
1
2
8
x
1
2
8
.
T
h
e
en
tire
m
eth
o
d
o
lo
g
y
is
d
ep
icted
in
Fig
u
r
e
1
.
Fig
u
r
e
1
.
B
lo
ck
d
iag
r
am
o
f
th
e
p
r
o
p
o
s
ed
f
ac
e
r
ec
o
g
n
itio
n
m
eth
o
d
G
u
i
d
ed
i
m
a
g
e
fi
l
te
r
P
ro
p
o
s
e
d
CNN
R
ec
o
g
n
i
z
ed
fa
ce
w
i
th
cl
a
s
s
l
a
b
el
F
a
ce
d
ete
cti
o
n
a
n
d
re
s
i
z
i
n
g
I
n
p
u
t
fa
ce
i
m
a
g
e
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
A
n
effec
tive
fa
ce
r
ec
o
g
n
itio
n
meth
o
d
u
s
in
g
g
u
id
e
d
ima
g
e
fil
ter a
n
d
co
n
v
o
lu
tio
n
a
l…
(
Ya
lla
ma
n
d
a
ia
h
S
)
1701
3
.
1
.
G
uid
ed
im
a
g
e
f
ilte
r
T
o
o
b
tain
in
f
o
r
m
atio
n
in
im
ag
es,
th
e
m
ajo
r
ity
o
f
ap
p
licatio
n
s
in
p
atter
n
r
ec
o
g
n
itio
n
u
s
es
im
ag
e
f
ilter
in
g
.
T
h
e
m
ea
n
,
L
a
p
lacia
n
,
So
b
el,
a
n
d
Gau
s
s
ian
f
ilter
s
h
av
e
b
ee
n
ex
ten
s
iv
ely
u
tili
ze
d
in
im
ag
e
f
ea
tu
r
e
ex
tr
ac
tio
n
,
s
h
ar
p
en
in
g
/b
lu
r
r
i
n
g
,
r
esto
r
atio
n
an
d
e
d
g
e
d
etec
tio
n
.
T
h
e
b
ilater
al
f
ilter
is
th
e
m
o
s
t
in
tu
itiv
e
an
d
s
im
p
lest
o
n
e
am
o
n
g
weig
h
te
d
av
e
r
ag
e
f
ilter
s
[
3
3
]
.
E
v
en
th
o
u
g
h
th
e
b
ilater
al
f
ilter
is
s
u
cc
ess
f
u
l
in
m
an
y
cir
cu
m
s
tan
ce
s
,
g
r
a
d
ien
t
r
ev
er
s
al
ar
tifa
cts
d
im
in
is
h
its
p
e
r
f
o
r
m
an
ce
[
3
4
]
-
[
3
6
]
.
T
h
is
p
r
o
b
lem
is
o
v
e
r
co
m
e
b
y
th
e
g
u
id
e
d
f
ilter
.
T
h
e
g
u
i
d
ed
f
ilter
is
an
ex
p
licit
im
ag
e
f
ilter
o
b
tain
ed
f
r
o
m
a
lin
ea
r
m
o
d
el
a
n
d
d
eter
m
i
n
es
th
e
f
ilter
in
g
o
u
t
p
u
t b
ased
u
p
o
n
th
e
co
n
ten
t o
f
th
e
g
u
id
an
ce
i
m
ag
e
[
3
7
]
.
L
et
th
e
is
a
g
u
id
an
ce
im
ag
e,
an
d
is
a
f
ilter
in
g
in
p
u
t im
ag
e,
t
h
e
g
en
er
al
lin
ea
r
tr
an
s
latio
n
-
i
n
v
ar
ian
t
f
ilter
in
g
o
u
tc
o
m
e
at
a
p
i
x
el
is
g
iv
en
as
,
=
∑
(
)
(
1
)
wh
er
e
,
an
d
ar
e
p
ix
el
in
d
ices,
is
th
e
k
er
n
el.
T
h
e
g
u
id
e
d
f
ilter
is
a
lin
ea
r
m
o
d
el
b
etwe
en
g
u
id
an
ce
an
d
th
e
f
ilter
in
g
o
u
tp
u
t
an
d
is
g
iv
en
b
y
:
=
+
,
∀
∈
(
2
)
wh
er
e
(
,
)
ar
e
lin
ea
r
co
ef
f
icie
n
ts
in
.
T
o
d
eter
m
in
e
th
e
co
ef
f
icien
t
s
(
,
)
,
th
e
o
u
tp
u
t
is
m
o
d
eled
a
s
th
e
in
p
u
t
s
u
b
tr
ac
tin
g
f
ew
u
n
d
esira
b
le
co
m
p
o
n
en
ts
:
=
−
(
3
)
Min
im
ize
th
e
f
o
llo
win
g
co
s
t f
u
n
ctio
n
in
th
e
win
d
o
w
t
o
f
in
d
th
e
s
o
lu
t
io
n
f
o
r
th
e
(
2
)
(
,
)
=
∑
(
(
+
−
)
2
+
2
)
∈
(
4
)
wh
er
e
is
a
r
eg
u
lar
izatio
n
p
ar
a
m
eter
.
In
(
4
)
is
th
e
lin
ea
r
r
eg
r
e
s
s
iv
e
m
o
d
el
an
d
its
s
o
lu
tio
n
is
g
iv
en
b
y
,
=
1
|
|
∑
−
̅
∈
2
+
(
5
)
=
̅
−
(
6
)
Her
e
an
d
2
ar
e
th
e
m
ea
n
a
n
d
v
ar
ian
ce
o
f
in
,
|
|
is
th
e
to
tal
n
u
m
b
er
o
f
p
i
x
els
in
an
d
̅
=
1
|
|
∑
∈
is
th
e
m
ea
n
o
f
in
.
Af
ter
o
b
tain
in
g
(
,
)
,
we
ca
n
f
in
d
th
e
f
ilter
in
g
o
u
tp
u
t
b
y
(
2
)
.
T
h
e
in
p
u
t
f
ac
e
im
ag
e
a
n
d
th
e
f
ilter
ed
o
u
tp
u
t
f
r
o
m
th
e
g
u
id
ed
im
ag
e
f
ilter
ar
e
s
h
o
w
n
in
Fig
u
r
e
2
.
(
a)
(
b
)
(
c)
Fig
u
r
e
2
.
(
a
)
I
n
p
u
t f
ac
e
im
ag
e
(
b
)
Face
d
etec
tio
n
(
c)
Gu
i
d
ed
im
ag
e
f
ilter
o
u
t
p
u
t
3
.
2
.
P
ro
po
s
ed
co
nv
o
lutio
na
l
neura
l net
wo
rk
I
n
r
ec
en
t
y
ea
r
s
co
n
v
o
lu
tio
n
al
n
eu
r
al
n
etwo
r
k
s
h
av
e
r
e
m
ar
k
ab
ly
b
o
o
s
ted
t
h
e
s
tate
-
of
-
th
e
-
a
r
t
p
er
f
o
r
m
an
ce
f
o
r
v
ar
io
u
s
v
is
u
al
task
s
.
Fo
r
ex
am
p
le,
im
ag
e
r
etr
iev
al,
s
em
an
tic
s
eg
m
en
tatio
n
,
m
u
ltit
ask
lear
n
in
g
,
im
ag
e
class
if
icat
io
n
,
an
d
p
e
r
s
o
n
r
e
-
id
e
n
tific
atio
n
.
C
o
n
v
o
lu
tio
n
al
n
eu
r
al
n
etwo
r
k
s
in
teg
r
ate
b
o
th
f
ea
tu
r
e
ex
tr
ac
tio
n
an
d
class
if
icatio
n
.
T
h
e
f
ea
tu
r
e
ex
tr
ac
tio
n
is
d
o
n
e
b
y
th
e
c
o
n
v
o
lu
tio
n
al
lay
er
an
d
p
o
o
lin
g
lay
er
.
T
h
e
f
u
lly
co
n
n
ec
ted
lay
er
is
u
s
ed
f
o
r
class
if
icatio
n
p
u
r
p
o
s
es.
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
.
3
,
Sep
tem
b
er
2
0
2
1
:
1
6
9
9
-
1
7
0
7
1702
T
h
e
ar
ch
itectu
r
e
o
f
th
e
p
r
o
p
o
s
ed
C
NN
is
d
ep
icted
in
F
ig
u
r
e
3
.
T
h
e
p
r
o
p
o
s
ed
C
NN
co
n
s
i
s
ts
o
f
f
o
u
r
co
n
v
o
l
u
tio
n
al
lay
e
r
s
with
8
,
1
6
,
3
2
,
an
d
6
4
f
ilter
s
.
R
eL
U
n
o
n
lin
ea
r
ac
tiv
atio
n
f
u
n
ctio
n
is
u
tili
ze
d
in
ea
ch
co
n
v
o
l
u
tio
n
al
lay
e
r
.
I
n
ea
c
h
c
o
n
v
o
lu
ti
o
n
al
lay
e
r
,
th
e
s
tr
id
e
i
s
s
et
to
o
n
e.
T
h
e
in
p
u
t
to
th
e
p
r
o
p
o
s
ed
C
NN
is
a
1
2
8
x
1
2
8
x
1
im
ag
e
.
T
h
e
f
ir
s
t
co
n
v
o
lu
ti
o
n
al
lay
er
c
o
n
s
is
ts
o
f
5
x
5
k
e
r
n
els
with
eig
h
t
f
ilter
s
.
Hen
ce
th
e
o
u
tco
m
e
o
f
th
e
co
n
v
1
is
eig
h
t
f
ea
tu
r
e
m
ap
s
with
s
ize
1
2
4
x
1
2
4
.
Ma
x
p
o
o
lin
g
lay
er
s
f
o
llo
w
ea
ch
co
n
v
o
l
u
tio
n
al
lay
er
with
a
2
x
2
win
d
o
w
a
n
d
s
tr
id
e
two
.
Ma
x
p
o
o
lin
g
1
p
r
o
d
u
ce
s
f
ea
tu
r
e
m
ap
s
with
a
s
ize
o
f
6
2
x
6
2
.
I
n
ea
ch
m
ax
-
p
o
o
lin
g
lay
er
,
th
e
s
tr
id
e
is
s
et
to
two
.
T
h
e
d
im
en
s
io
n
s
o
f
th
e
f
ea
tu
r
e
m
ap
s
g
en
er
ated
b
y
co
n
v
2
,
co
n
v
3
,
an
d
co
n
v
4
ar
e
5
8
x
5
8
x
1
6
,
2
5
x
2
5
x
3
2
,
a
n
d
9
x
9
x
6
4
r
esp
ec
tiv
ely
.
Ma
x
p
o
o
lin
g
2
,
m
ax
p
o
o
lin
g
3
,
an
d
m
ax
p
o
o
lin
g
4
lay
er
s
p
r
o
d
u
ce
an
o
u
tp
u
t
with
d
im
en
s
io
n
s
2
9
x
2
9
x
1
6
,
1
3
x
1
3
x
3
2
,
an
d
4
x
4
x
6
4
r
esp
ec
tiv
ely
.
T
h
e
f
in
al
m
ax
-
p
o
o
lin
g
lay
er
is
f
o
llo
wed
b
y
two
f
u
lly
co
n
n
ec
ted
lay
er
s
with
1
0
2
4
a
n
d
5
1
2
u
n
its
.
Fi
n
ally
,
th
e
s
o
f
tm
ax
f
u
n
ctio
n
is
u
s
ed
f
o
r
class
if
icatio
n
p
u
r
p
o
s
es.
Sto
ch
asti
c
G
r
ad
ien
t
Descen
t
is
u
tili
ze
d
a
s
an
o
p
tim
izer
f
o
r
tr
ain
in
g
th
e
d
ata
to
th
e
p
r
o
p
o
s
ed
C
NN,
with
a
b
ase
lear
n
in
g
r
ate
o
f
0
.
0
0
1
.
we
u
s
ed
a
b
atch
s
ize
o
f
f
o
u
r
wh
ile
tr
ain
in
g
th
e
n
etwo
r
k
.
Fig
u
r
e
3
.
Ar
c
h
itectu
r
e
o
f
th
e
p
r
o
p
o
s
ed
C
NN
3
.
3
.
T
he
no
v
elt
y
o
f
t
he
pro
po
s
ed
m
et
ho
d
T
h
e
n
o
v
elty
o
f
th
e
p
r
o
p
o
s
ed
m
eth
o
d
o
l
o
g
y
is
r
em
o
v
in
g
th
e
n
o
is
e
p
r
esen
t
in
th
e
f
ac
e
im
ag
es
b
ef
o
r
e
ap
p
ly
in
g
th
em
t
o
th
e
p
r
o
p
o
s
e
d
C
NN
an
d
d
esig
n
i
n
g
th
e
C
NN
with
a
lim
ited
n
u
m
b
er
o
f
lay
er
s
.
R
em
o
v
in
g
th
e
n
o
is
es
ex
is
tin
g
in
th
e
f
ac
e
i
m
ag
es
im
p
r
o
v
es
th
e
f
ac
e
r
ec
o
g
n
itio
n
ac
cu
r
ac
y
an
d
d
esig
n
in
g
th
e
C
NN
with
a
f
ewe
r
n
u
m
b
e
r
o
f
lay
e
r
s
d
ec
r
e
ases
th
e
m
o
d
el
co
m
p
lex
ity
.
Ap
p
en
d
in
g
ex
tr
a
lay
e
r
s
h
elp
to
ex
tr
ac
t
th
e
m
o
r
e
d
etailed
f
ea
tu
r
es,
b
u
t
we
ca
n
ad
d
lay
er
s
u
p
to
a
ce
r
tain
lim
it.
Af
ter
th
at,
th
e
m
o
d
el
o
v
er
f
its
th
e
d
ata
wh
ich
lead
s
to
er
r
o
r
s
lik
e
f
alse
p
o
s
itiv
es.
I
n
a
d
d
itio
n
t
o
th
is
,
if
w
e
ad
d
m
o
r
e
la
y
er
s
th
e
n
u
m
b
e
r
o
f
weig
h
ts
in
th
e
n
etwo
r
k
in
cr
ea
s
es
an
d
lead
s
t
o
in
cr
ea
s
e
in
th
e
m
o
d
el
c
o
m
p
lex
ity
.
T
o
r
ed
u
ce
th
is
co
m
p
lex
i
ty
,
we
d
esig
n
ed
th
e
C
NN
with
th
e
o
p
tim
u
m
n
u
m
b
er
o
f
lay
e
r
s
.
4.
E
XP
E
R
I
M
E
N
T
A
L
RE
SUL
T
S AN
D
D
I
SC
USS
I
O
NS
W
e
s
h
o
w
th
e
ef
f
icien
cy
o
f
o
u
r
f
ac
e
r
ec
o
g
n
itio
n
s
y
s
tem
a
cr
o
s
s
d
if
f
er
en
t
p
o
s
es,
ex
p
r
es
s
io
n
s
,
an
d
illu
m
in
atio
n
s
u
s
in
g
th
e
OR
L
[
3
8
]
,
J
AFFE
[
3
9
]
,
an
d
YAL
E
[
4
0
]
f
ac
e
d
atasets
.
W
e
test
ed
o
u
r
m
et
h
o
d
f
o
r
a
d
if
f
er
en
t
r
ad
iu
s
(
r
)
o
f
th
e
s
q
u
ar
e
win
d
o
w
an
d
r
eg
u
lar
izatio
n
p
a
r
am
eter
s
(
ε
)
o
f
a
g
u
id
e
d
im
ag
e
f
ilter
.
W
e
ch
o
o
s
e
7
0
% o
f
im
ag
es in
ea
c
h
class
f
o
r
tr
ain
in
g
a
n
d
th
e
r
est o
f
th
e
im
ag
es we
r
e
u
tili
ze
d
f
o
r
test
in
g
.
T
h
e
O
R
L
f
a
c
e
d
a
ta
b
a
s
e
c
o
m
p
r
i
s
e
s
4
0
0
f
a
c
e
i
m
a
g
es
c
o
ll
ec
t
e
d
f
r
o
m
4
0
d
i
f
f
e
r
e
n
t
p
e
r
s
o
n
s
w
it
h
t
e
n
d
i
s
t
i
n
ct
i
m
a
g
e
s
f
o
r
e
a
c
h
p
e
r
s
o
n
.
T
h
e
s
e
i
m
a
g
e
s
m
a
n
i
f
es
t
v
a
r
i
a
t
i
o
n
s
i
n
t
h
e
p
o
s
e
,
i
l
l
u
m
i
n
a
t
i
o
n
,
a
n
d
f
a
c
i
a
l
e
x
p
r
e
s
s
i
o
n
s
l
i
k
e
s
m
i
li
n
g
o
r
n
o
t
s
m
i
l
i
n
g
,
e
y
es
c
l
o
s
e
d
o
r
o
p
e
n
e
d
.
T
h
e
J
A
FF
E
f
a
c
e
d
a
t
a
b
as
e
is
a
c
o
l
l
e
c
t
i
o
n
o
f
2
1
3
g
r
a
y
s
c
a
l
e
i
m
a
g
e
s
o
f
t
e
n
J
a
p
a
n
e
s
e
f
e
m
al
e
m
o
d
e
ls
.
T
h
e
d
a
ta
b
a
s
e
c
o
m
p
r
is
e
s
f
a
c
i
al
e
x
p
r
e
s
s
i
o
n
s
l
i
k
e
t
h
e
s
u
r
p
r
is
e,
h
a
p
p
i
n
e
s
s
,
s
a
d
n
e
s
s
,
a
n
g
e
r
,
f
e
a
r
,
n
e
u
t
r
a
l
,
a
n
d
d
i
s
g
u
s
t
o
f
e
a
c
h
s
u
b
j
e
c
t
.
T
h
e
Y
A
L
E
f
a
c
e
d
a
ta
b
a
s
e
i
n
c
l
u
d
es
1
6
5
g
r
a
y
s
c
a
l
e
i
m
a
g
es
o
f
1
5
s
u
b
j
e
ct
s
w
it
h
1
1
i
m
a
g
e
s
p
e
r
s
u
b
j
e
ct
.
E
a
c
h
s
u
b
j
e
c
t
c
o
n
t
a
i
n
s
i
m
a
g
es
w
i
t
h
t
h
e
f
o
l
l
o
w
i
n
g
c
o
n
f
i
g
u
r
a
t
i
o
n
s
:
c
e
n
t
e
r
-
li
g
h
t
,
h
a
p
p
y
,
w
i
t
h
g
l
a
s
s
es
,
s
l
ee
p
y
,
l
e
f
t
-
l
i
g
h
t
,
n
o
r
m
a
l
,
s
a
d
,
w
i
t
h
n
o
g
l
a
s
s
e
s
,
wi
n
k
,
r
i
g
h
t
-
l
i
g
h
t
,
a
n
d
s
u
r
p
r
i
s
e
d
.
T
h
e
i
m
a
g
es
t
h
a
t
b
el
o
n
g
t
o
a
s
i
n
g
l
e
s
u
b
j
e
ct
o
f
t
h
e
s
e
d
a
t
a
b
as
es
a
r
e
s
h
o
w
n
in
F
i
g
u
r
e
4
.
Fig
u
r
e
5
s
h
o
ws
th
e
o
u
tp
u
t
im
ag
es
o
f
th
e
g
u
id
ed
im
ag
e
f
ilte
r
f
o
r
OR
L
,
J
AFFE,
an
d
YAL
E
d
atab
ases
with
d
if
f
er
en
t
s
ets
o
f
p
ar
a
m
et
er
s
.
Fo
r
co
m
p
ar
is
o
n
p
u
r
p
o
s
es,
we
ex
tr
ac
ted
th
e
f
ea
tu
r
es
f
r
o
m
th
e
s
ec
o
n
d
f
u
lly
co
n
n
ec
ted
la
y
er
o
f
th
e
p
r
o
p
o
s
ed
C
NN
an
d
class
if
ied
th
em
u
s
in
g
a
d
ec
is
io
n
tr
ee
a
n
d
r
a
n
d
o
m
f
o
r
est
class
if
ier
.
T
h
e
ef
f
icien
cy
o
f
th
e
class
if
ier
is
ev
alu
ated
u
s
in
g
r
ec
o
g
n
itio
n
r
ate
(
R
R
)
,
ar
ea
u
n
d
er
cu
r
v
e
(
AUC),
tr
u
e
p
o
s
itiv
e
r
a
te
(
T
PR
)
,
an
d
f
alse
p
o
s
itiv
e
r
ate
(
FP
R
)
.
T
h
e
R
R
o
f
th
e
p
r
o
p
o
s
ed
m
eth
o
d
f
o
r
th
e
d
if
f
er
en
t
r
a
d
iu
s
(
r
)
o
f
th
e
s
q
u
ar
e
win
d
o
w
an
d
r
eg
u
lar
izatio
n
p
a
r
am
eter
(
ε
)
is
g
i
v
en
in
T
ab
le
1
.
T
h
e
s
u
g
g
ested
m
eth
o
d
g
iv
es
th
e
b
est
R
R
o
f
9
8
.
3
3
%,
9
9
.
5
3
%,
an
d
9
8
.
6
5
%
o
n
OR
L
,
J
AFFE,
an
d
YAL
E
d
atab
ases
r
esp
ec
ti
v
ely
f
o
r
=
2
an
d
=
0
.
4
2
.
C
o
m
p
ar
ed
t
o
th
e
r
a
n
d
o
m
f
o
r
est an
d
d
ec
is
io
n
tr
ee
class
if
ier
s
s
o
f
tm
ax
h
as g
iv
en
t
h
e
b
est R
R
.
Con
v2
58x58x16
Co
nv1
124x124x8
Con
v3
25x25x32
Con
v4
9x9x64
P
oo
l
i
ng
1
62x62x8
P
ool
i
ng
2
29x29x16
P
ool
i
ng
3
13x13x32
P
ool
i
ng
4
4x4x64
f
=
5x
5
s
=1
f=
2x2
s
=
2
f
=
5x5
s
=1
f=
2x2
s
=
2
f
=
5x
5
s
=1
f=
2x2
s
=
2
f
=
5x
5
s
=1
f=
2x
2
s
=
2
F
C1
1024
F
C2
512
Inp
ut
128x128x1
S
oft
m
a
x
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
A
n
effec
tive
fa
ce
r
ec
o
g
n
itio
n
meth
o
d
u
s
in
g
g
u
id
e
d
ima
g
e
fil
ter a
n
d
co
n
v
o
lu
tio
n
a
l…
(
Ya
lla
ma
n
d
a
ia
h
S
)
1703
T
a
b
l
e
2
g
i
v
e
s
t
h
e
T
P
R
f
o
r
al
l
th
e
c
h
o
s
e
n
f
a
c
e
d
at
a
b
a
s
es
.
T
h
e
s
o
f
t
m
a
x
c
l
ass
i
f
i
e
r
h
as
g
i
v
e
n
a
m
a
x
i
m
u
m
T
P
R
o
f
9
8
.
5
7
%
w
h
e
r
e
as
t
h
e
d
e
c
i
s
i
o
n
t
r
ee
a
n
d
r
a
n
d
o
m
f
o
r
e
s
t
c
la
s
s
i
f
i
e
r
s
h
a
v
e
p
r
o
d
u
c
e
d
T
P
R
o
f
9
9
.
6
1
%
a
n
d
9
8
.
6
5
%
o
n
O
R
L
,
J
AF
F
E
,
a
n
d
Y
A
L
E
d
a
t
a
b
as
e
s
r
es
p
e
c
ti
v
e
l
y
f
o
r
=
2
a
n
d
=
0
.
4
2
.
F
r
o
m
T
a
b
l
e
2
,
it
is
n
o
ticed
th
at
th
e
s
o
f
tm
ax
class
if
ier
h
as
g
iv
en
h
i
g
h
T
PR
co
m
p
a
r
ed
to
t
h
e
d
ec
is
io
n
tr
ee
a
n
d
r
a
n
d
o
m
f
o
r
est.
T
h
e
FP
R
f
o
r
th
e
p
r
o
p
o
s
ed
m
eth
o
d
with
d
if
f
er
en
t
cl
ass
if
ier
s
is
g
iv
en
in
T
ab
le
3
.
L
o
w
FP
R
is
attain
ed
with
a
s
o
f
tm
ax
class
if
ier
co
m
p
ar
ed
t
o
th
e
d
e
cisi
o
n
tr
ee
an
d
r
an
d
o
m
f
o
r
est
class
if
ier
s
.
T
h
e
p
r
o
p
o
s
ed
m
e
th
o
d
with
s
o
f
tm
ax
class
if
ier
h
as
g
iv
en
an
FP
R
o
f
0
.
0
2
,
0
.
0
1
,
a
n
d
0
.
0
2
o
n
OR
L
,
J
AFFE,
an
d
YAL
E
d
atab
ases
f
o
r
=
2
an
d
=
0
.
4
2
.
T
h
e
AUC
is
d
eter
m
in
ed
an
d
g
iv
en
i
n
T
ab
le
4
.
T
h
e
m
ax
im
u
m
v
alu
e
o
f
th
e
AUC
with
s
o
f
t
m
ax
class
if
ier
o
n
OR
L
,
J
AF
FE,
an
d
YAL
E
d
atab
ases
is
9
9
.
5
2
%,
9
9
.
7
8
%,
an
d
9
9
.
6
3
%.
Fro
m
th
e
v
alu
e
s
o
f
T
ab
le
4
,
it
is
o
b
s
er
v
ed
th
at
th
e
p
r
o
p
o
s
ed
m
e
th
o
d
with
th
e
s
o
f
t
m
ax
class
if
ier
h
as m
o
r
e
AUC th
an
th
e
o
th
e
r
class
if
ier
s
.
(
a)
(
b
)
(
c)
Fig
u
r
e
4
.
I
m
ag
es b
elo
n
g
to
a
s
in
g
le
s
u
b
ject
o
f
(
a)
OR
L
,
(
b
)
J
AFFE
an
d
(
c)
YAL
E
d
atab
ases
r
=
2
I
n
p
u
t
f
a
c
e
i
ma
g
e
r
=
4
r
=
8
ε=0
.
1
2
ε=0
.
2
2
ε=0
.
4
2
(
a)
r
=
2
I
n
p
u
t
f
a
c
e
i
ma
g
e
r
=
4
r
=
8
ε=0
.
1
2
ε=0
.
2
2
ε=0
.
4
2
(
b
)
r
=
2
I
n
p
u
t
f
a
c
e
i
ma
g
e
r
=
4
r
=
8
ε=0
.
1
2
ε=0
.
2
2
ε=0
.
4
2
(
c)
Fig
u
r
e
5
.
T
h
e
o
u
t
p
u
t im
ag
es o
f
a
g
u
id
e
d
im
ag
e
f
ilter
f
o
r
th
e
g
r
ay
-
s
ca
le
in
p
u
t f
r
o
m
(
a)
OR
L
(
b
)
J
AFFE
an
d
(
c)
YAL
E
d
atab
ase
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
.
3
,
Sep
tem
b
er
2
0
2
1
:
1
6
9
9
-
1
7
0
7
1704
T
ab
le
1
.
R
ec
o
g
n
itio
n
r
ate
u
s
in
g
d
ec
is
io
n
tr
ee
,
r
an
d
o
m
f
o
r
est,
an
d
s
o
f
tm
a
x
class
if
ier
R
a
d
i
u
s
(r)
R
e
g
u
l
a
r
i
z
a
t
i
o
n
p
a
r
a
m
e
t
e
r
(
ε)
O
R
L
JA
F
F
E
Y
A
LE
D
e
c
i
s
i
o
n
Tr
e
e
R
a
n
d
o
m
F
o
r
e
st
S
o
f
t
ma
x
D
e
c
i
s
i
o
n
Tr
e
e
R
a
n
d
o
m
F
o
r
e
st
S
o
f
t
ma
x
D
e
c
i
s
i
o
n
Tr
e
e
R
a
n
d
o
m
F
o
r
e
st
S
o
f
t
ma
x
2
0
.
1
2
9
5
.
2
5
9
5
.
8
7
9
6
.
6
7
9
4
.
2
8
9
5
.
5
2
9
8
.
3
3
9
5
.
1
7
9
5
.
2
8
9
7
.
7
8
0
.
2
2
9
4
.
3
6
9
4
.
6
4
9
5
.
8
3
9
3
.
4
6
9
6
.
3
1
9
8
.
3
3
9
5
.
4
5
9
6
.
4
3
9
7
.
7
8
0
.
4
2
9
5
.
4
3
9
6
.
5
8
9
8
.
3
3
9
5
.
8
2
9
6
.
7
4
9
9
.
5
3
9
5
.
6
3
9
6
.
6
2
9
8
.
6
5
4
0
.
1
2
8
6
.
2
3
8
8
.
1
9
8
9
.
1
7
9
2
.
5
9
9
3
.
3
3
9
6
.
6
7
9
1
.
2
4
9
3
.
4
8
9
7
.
8
9
0
.
2
2
8
7
.
6
2
8
9
.
4
5
9
1
.
6
7
9
1
.
3
5
9
4
.
6
3
9
6
.
6
7
9
2
.
8
7
9
4
.
2
4
9
7
.
7
8
0
.
4
2
8
8
.
4
8
9
0
.
2
4
9
1
.
6
7
8
9
.
7
5
9
2
.
4
4
9
5
.
0
0
8
9
.
5
9
9
2
.
3
1
9
5
.
5
6
8
0
.
1
2
8
6
.
4
1
8
7
.
5
2
9
0
.
8
3
8
8
.
9
3
8
9
.
9
3
9
3
.
3
3
9
0
.
8
8
9
3
.
9
7
9
7
.
7
8
0
.
2
2
8
6
.
6
7
8
7
.
8
6
9
0
.
8
3
8
9
.
8
2
9
2
.
1
2
9
3
.
3
3
9
1
.
7
3
9
4
.
6
6
9
7
.
7
8
0
.
4
2
8
5
.
4
1
8
8
.
3
8
9
0
.
0
0
9
0
.
9
1
9
1
.
5
3
9
3
.
3
3
8
7
.
2
4
8
9
.
1
5
9
3
.
3
3
T
ab
le
2
.
T
r
u
e
p
o
s
itiv
e
r
ate
u
s
in
g
d
ec
is
io
n
tr
ee
,
r
an
d
o
m
f
o
r
es
t,
an
d
s
o
f
tm
a
x
class
if
ier
R
a
d
i
u
s
(r)
R
e
g
u
l
a
r
i
z
a
t
i
o
n
p
a
r
a
m
e
t
e
r
(
ε)
O
R
L
JA
F
F
E
Y
A
LE
D
e
c
i
s
i
o
n
Tr
e
e
R
a
n
d
o
m
F
o
r
e
st
S
o
f
t
ma
x
D
e
c
i
s
i
o
n
Tr
e
e
R
a
n
d
o
m
F
o
r
e
st
S
o
f
t
ma
x
D
e
c
i
s
i
o
n
Tr
e
e
R
a
n
d
o
m
F
o
r
e
st
S
o
f
t
ma
x
2
0
.
1
2
9
5
.
2
5
9
5
.
8
7
9
6
.
6
7
9
4
.
2
8
9
5
.
5
2
9
8
.
3
3
9
5
.
1
7
9
5
.
2
8
9
7
.
7
8
0
.
2
2
9
4
.
5
6
9
4
.
6
4
9
5
.
8
3
9
3
.
4
6
9
6
.
5
2
9
8
.
3
3
9
5
.
4
5
9
6
.
4
3
9
7
.
7
8
0
.
4
2
9
5
.
4
3
9
6
.
8
8
9
8
.
5
7
9
5
.
8
2
9
6
.
9
4
9
9
.
6
1
9
5
.
6
3
9
6
.
9
3
9
8
.
6
5
4
0
.
1
2
8
6
.
2
3
8
8
.
1
9
9
0
.
2
4
9
3
.
2
5
9
3
.
3
3
9
6
.
6
7
9
1
.
2
4
9
3
.
4
8
9
8
.
2
1
0
.
2
2
8
7
.
6
2
8
9
.
4
5
9
1
.
6
7
9
1
.
3
5
9
4
.
6
3
9
6
.
6
7
9
3
.
5
2
9
5
.
2
6
9
7
.
7
8
0
.
4
2
8
8
.
4
8
9
0
.
2
4
9
1
.
6
7
9
0
.
5
7
9
3
.
6
4
9
5
.
4
2
9
0
.
2
5
9
3
.
4
3
9
6
.
2
5
8
0
.
1
2
8
6
.
4
1
8
8
.
2
2
9
0
.
8
3
8
8
.
9
3
8
9
.
9
3
9
4
.
3
6
9
1
.
6
8
9
4
.
7
2
9
7
.
7
8
0
.
2
2
8
6
.
9
2
8
7
.
8
6
9
0
.
8
3
8
9
.
8
2
9
2
.
1
2
9
4
.
3
6
9
2
.
1
3
9
4
.
8
9
9
7
.
7
8
0
.
4
2
8
5
.
4
1
8
8
.
3
8
9
1
.
6
2
9
0
.
9
1
9
1
.
5
3
9
4
.
3
6
8
7
.
2
4
8
9
.
1
5
9
4
.
7
3
T
ab
le
3
.
Fals
e
p
o
s
itiv
e
r
ate
u
s
i
n
g
d
ec
is
io
n
tr
ee
,
r
an
d
o
m
f
o
r
es
t,
an
d
s
o
f
tm
a
x
c
lass
if
ier
R
a
d
i
u
s
(r)
R
e
g
u
l
a
r
i
z
a
t
i
o
n
p
a
r
a
m
e
t
e
r
(
ε)
O
R
L
JA
F
F
E
Y
A
LE
D
e
c
i
s
i
o
n
Tr
e
e
R
a
n
d
o
m
F
o
r
e
st
S
o
f
t
ma
x
D
e
c
i
s
i
o
n
Tr
e
e
R
a
n
d
o
m
F
o
r
e
st
S
o
f
t
ma
x
D
e
c
i
s
i
o
n
Tr
e
e
R
a
n
d
o
m
F
o
r
e
st
S
o
f
t
ma
x
2
0
.
1
2
0
.
0
6
0
.
0
4
0
.
0
3
0
.
0
6
0
.
0
5
0
.
0
2
0
.
0
7
0
.
0
4
0
.
0
3
0
.
2
2
0
.
0
8
0
.
0
7
0
.
0
5
0
.
0
7
0
.
0
5
0
.
0
2
0
.
0
8
0
.
0
6
0
.
0
3
0
.
4
2
0
.
0
6
0
.
0
3
0
.
0
2
0
.
0
6
0
.
0
4
0
.
0
1
0
.
0
7
0
.
0
4
0
.
0
2
4
0
.
1
2
0
.
0
9
0
.
0
8
0
.
0
7
0
.
0
9
0
.
0
8
0
.
0
4
0
.
0
8
0
.
0
5
0
.
0
3
0
.
2
2
0
.
0
5
0
.
0
4
0
.
0
6
0
.
0
9
0
.
0
7
0
.
0
4
0
.
0
7
0
.
0
6
0
.
0
3
0
.
4
2
0
.
0
9
0
.
0
8
0
.
0
6
0
.
0
8
0
.
0
7
0
.
0
5
0
.
0
9
0
.
0
8
0
.
0
4
8
0
.
1
2
0
.
0
9
0
.
0
9
0
.
0
7
0
.
0
8
0
.
0
6
0
.
0
5
0
.
0
8
0
.
0
7
0
.
0
3
0
.
2
2
0
.
0
9
0
.
0
9
0
.
0
7
0
.
0
8
0
.
0
7
0
.
0
5
0
.
0
7
0
.
0
6
0
.
0
3
0
.
4
2
0
.
0
9
0
.
0
8
0
.
0
6
0
.
0
7
0
.
0
4
0
.
0
5
0
.
0
8
0
.
0
7
0
.
0
5
T
ab
le
4
.
Ar
ea
u
n
d
e
r
cu
r
v
e
u
s
i
n
g
d
ec
is
io
n
tr
ee
,
r
an
d
o
m
f
o
r
es
t,
an
d
s
o
f
tm
a
x
class
if
ier
R
a
d
i
u
s
(r)
R
e
g
u
l
a
r
i
z
a
t
i
o
n
p
a
r
a
m
e
t
e
r
(
ε)
O
R
L
JA
F
F
E
Y
A
LE
D
e
c
i
s
i
o
n
Tr
e
e
R
a
n
d
o
m
F
o
r
e
st
so
f
t
m
a
x
D
e
c
i
s
i
o
n
Tr
e
e
R
a
n
d
o
m
F
o
r
e
st
so
f
t
m
a
x
D
e
c
i
s
i
o
n
Tr
e
e
R
a
n
d
o
m
F
o
r
e
st
so
f
t
m
a
x
2
0
.
1
2
9
6
.
5
3
9
7
.
8
4
9
8
.
7
3
9
5
.
9
4
9
6
.
4
5
9
9
.
5
3
9
6
.
6
7
9
7
.
5
9
9
9
.
2
4
0
.
2
2
9
5
.
4
1
9
5
.
2
1
9
7
.
8
1
9
5
.
6
3
9
7
.
2
8
9
9
.
5
3
9
6
.
5
3
9
7
.
2
8
9
9
.
2
4
0
.
4
2
9
6
.
8
2
9
7
.
2
3
9
9
.
5
2
9
7
.
9
5
9
8
.
9
2
9
9
.
7
8
9
6
.
9
2
9
7
.
6
3
9
9
.
6
3
4
0
.
1
2
8
7
.
4
2
8
8
.
4
4
9
3
.
7
2
9
3
.
4
8
9
5
.
4
7
9
8
.
7
5
9
3
.
6
7
9
6
.
9
8
9
8
.
7
6
0
.
2
2
8
8
.
8
5
9
0
.
2
8
9
5
.
3
8
9
4
.
7
8
9
5
.
5
8
9
8
.
7
5
9
4
.
2
7
9
8
.
8
3
9
9
.
2
4
0
.
4
2
8
9
.
5
5
9
1
.
6
8
9
5
.
3
8
9
2
.
4
6
9
4
.
6
4
9
8
.
4
3
9
4
.
4
4
9
6
.
1
6
9
8
.
6
3
8
0
.
1
2
8
7
.
2
9
8
8
.
7
3
9
4
.
6
8
9
3
.
3
7
9
6
.
3
4
9
8
.
4
7
9
2
.
3
7
9
4
.
3
1
9
9
.
2
1
0
.
2
2
8
7
.
7
6
8
8
.
4
3
9
4
.
6
8
9
2
.
2
8
9
5
.
7
6
9
8
.
4
7
9
3
.
6
6
9
7
.
4
4
9
9
.
2
1
0
.
4
2
8
6
.
4
5
8
9
.
9
2
9
3
.
7
2
9
4
.
1
9
9
6
.
2
9
9
8
.
4
7
9
1
.
3
4
9
5
.
5
2
9
8
.
7
8
T
h
e
R
OC
c
u
r
v
e
s
f
o
r
t
h
e
p
r
o
p
o
s
e
d
t
e
c
h
n
i
q
u
e
w
i
t
h
s
o
f
t
m
a
x
c
l
a
s
s
i
f
ie
r
s
a
r
e
d
e
p
i
c
t
e
d
i
n
F
i
g
u
r
e
6
.
T
h
e
p
e
r
f
o
r
m
a
n
c
e
c
o
m
p
a
r
i
s
o
n
o
f
t
h
e
s
u
g
g
e
s
t
e
d
m
e
t
h
o
d
w
it
h
s
o
m
e
o
f
t
h
e
e
x
i
s
t
i
n
g
m
e
t
h
o
d
s
i
s
e
n
ca
p
s
u
l
a
t
e
d
i
n
T
a
b
le
5
.
Fro
m
th
e
v
alu
es o
f
T
ab
le
5
,
it is
n
o
ticed
th
at
th
e
p
r
o
p
o
s
ed
f
a
ce
r
ec
o
g
n
itio
n
ap
p
r
o
ac
h
g
iv
es a
g
o
o
d
r
ec
o
g
n
itio
n
r
ate
th
an
s
o
m
e
o
f
th
e
ea
r
lier
a
p
p
r
o
ac
h
es.
I
m
ag
e
d
ata
au
g
m
en
tatio
n
en
la
r
g
es th
e
s
ize
o
f
th
e
tr
ain
in
g
d
a
taset b
y
p
r
o
d
u
cin
g
th
e
m
o
d
if
ie
d
v
er
s
io
n
s
o
f
th
e
im
ag
es
in
th
e
d
ataset.
Data
au
g
m
en
tatio
n
ca
n
im
p
r
o
v
e
th
e
p
er
f
o
r
m
an
ce
o
f
th
e
p
r
o
p
o
s
ed
f
ac
e
r
ec
o
g
n
iti
o
n
s
y
s
tem
.
W
e
ca
r
r
ied
o
u
t
h
o
r
izo
n
tal
a
n
d
v
e
r
tical
f
lip
p
in
g
,
r
o
tatio
n
b
y
4
5
d
eg
r
ee
s
,
an
d
zo
o
m
i
n
g
f
o
r
g
en
er
atin
g
ad
d
itio
n
al
im
ag
es
.
T
h
e
r
ec
o
g
n
itio
n
r
ate
o
f
th
e
p
r
o
p
o
s
ed
m
eth
o
d
with
a
n
d
with
o
u
t
th
e
d
ata
au
g
m
en
tatio
n
is
g
iv
en
in
T
ab
l
e
6
.
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
A
n
effec
tive
fa
ce
r
ec
o
g
n
itio
n
meth
o
d
u
s
in
g
g
u
id
e
d
ima
g
e
fil
ter a
n
d
co
n
v
o
lu
tio
n
a
l…
(
Ya
lla
ma
n
d
a
ia
h
S
)
1705
(
a)
(
b
)
(
c)
Fig
u
r
e
6
.
R
OC
cu
r
v
es f
o
r
(
a)
OR
L
(
b
)
J
AFFE
an
d
(
c)
YAL
E
d
atab
ases
f
o
r
=
2
an
d
=
0
.
4
2
T
ab
le
5
.
C
o
m
p
a
r
is
o
n
o
f
r
ec
o
g
n
itio
n
r
ate
(
%)
with
s
o
m
e
o
f
t
h
e
ex
is
tin
g
tech
n
iq
u
es
M
e
t
h
o
d
D
a
t
a
b
a
s
e
O
R
L
JA
F
F
E
Y
A
LE
C
LD
A
[
5
]
9
4
.
0
6
P
C
A
[
8
]
8
9
.
5
0
F
LD
[
8
]
9
1
.
0
0
P
C
A
i
m
a
g
e
r
e
c
o
n
s
t
r
u
c
t
i
o
n
+
LD
A
[
1
4
]
9
7
.
2
0
G
F
D
B
N
[
2
0
]
9
4
.
9
8
D
I
W
TLB
P
[
2
2
]
9
7
.
0
0
D
S
D
S
A
[
2
4
]
9
8
.
0
0
Pr
o
p
o
sed
9
8
.
3
3
F
LLEPC
A
[
1
0
]
9
4
.
9
8
S
i
n
g
l
e
2
D
N
N
R
W
[
1
5
]
9
7
.
0
0
P
S
O
[
1
7
]
9
8
.
8
0
Pr
o
p
o
sed
9
9
.
5
3
O
P
R
[
1
8
]
9
4
.
1
5
P
LR
[
1
9
]
9
6
.
2
3
R
D
C
D
L
[
2
3
]
9
7
.
2
2
D
S
D
S
A
[
2
4
]
9
8
.
1
6
Pr
o
p
o
sed
9
8
.
6
5
T
ab
le
6
.
R
ec
o
g
n
itio
n
r
ate
o
f
th
e
p
r
o
p
o
s
ed
m
eth
o
d
with
an
d
with
o
u
t
d
ata
au
g
m
en
tatio
n
D
a
t
a
b
a
s
e
O
R
L
JA
F
F
E
Y
A
LE
W
i
t
h
o
u
t
D
a
t
a
A
u
g
me
n
t
a
t
i
o
n
9
8
.
3
3
%
9
9
.
5
3
%
9
8
.
6
5
%
W
i
t
h
D
a
t
a
A
u
g
me
n
t
a
t
i
o
n
9
9
.
1
7
%
9
9
.
7
4
%
9
9
.
3
2
%
5.
CO
NCLU
SI
O
N
I
n
th
is
w
o
r
k
,
a
f
ac
e
r
ec
o
g
n
itio
n
a
p
p
r
o
ac
h
u
s
in
g
a
g
u
id
ed
i
m
ag
e
f
ilter
an
d
a
co
n
v
o
lu
tio
n
al
n
eu
r
al
n
etwo
r
k
was
p
r
o
p
o
s
ed
.
First,
th
e
f
ac
e
r
eg
io
n
o
f
th
e
test
im
ag
e
is
ex
tr
ac
ted
u
s
in
g
th
e
Vio
la
-
J
o
n
es
alg
o
r
ith
m
an
d
th
en
r
esized
.
T
h
e
r
esize
d
im
ag
e
was
p
ass
ed
th
r
o
u
g
h
a
g
u
id
ed
im
ag
e
f
ilter
.
A
g
u
id
ed
im
ag
e
f
ilter
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
.
3
,
Sep
tem
b
er
2
0
2
1
:
1
6
9
9
-
1
7
0
7
1706
s
m
o
o
th
en
s
th
e
f
ac
e
im
ag
e,
th
en
th
e
p
r
o
p
o
s
ed
C
NN
wa
s
u
ti
lized
to
ex
tr
ac
t
th
e
f
ea
tu
r
es
an
d
class
if
y
th
e
in
p
u
t
f
ac
e
im
ag
e.
He
r
e,
th
e
s
o
f
tm
ax
class
if
ier
,
wh
ich
g
iv
es
g
o
o
d
r
esu
lts
th
an
th
e
d
ec
is
io
n
tr
ee
an
d
r
a
n
d
o
m
f
o
r
est,
was
u
s
ed
in
C
NN’
s
clas
s
if
ier
s
ec
tio
n
.
T
h
e
ca
p
a
b
ilit
y
o
f
th
e
p
r
o
p
o
s
ed
ap
p
r
o
ac
h
was
co
m
p
ar
ed
with
s
o
m
e
o
f
th
e
ea
r
lier
m
eth
o
d
s
.
Fro
m
th
e
co
m
p
ar
ativ
e
r
esu
lts
,
it
is
f
o
u
n
d
th
at
t
h
e
s
u
g
g
es
ted
tech
n
iq
u
e
p
r
o
d
u
ce
s
b
etter
r
esu
lts
th
an
s
o
m
e
o
f
th
e
ex
is
tin
g
m
eth
o
d
s
.
B
ased
o
n
th
e
ex
p
er
im
en
tal
r
esu
lts
,
it
is
c
o
n
clu
d
e
d
th
at
o
u
r
p
r
o
p
o
s
ed
m
eth
o
d
ca
n
b
e
u
s
ed
f
o
r
f
ac
e
r
ec
o
g
n
itio
n
.
RE
F
E
R
E
NC
E
S
[1
]
Ya
n
g
,
M
in
g
-
Hs
u
a
n
,
Da
v
i
d
J.
Krie
g
m
a
n
,
a
n
d
Na
re
n
d
ra
Ah
u
ja,
"
De
te
c
ti
n
g
fa
c
e
s
in
ima
g
e
s:
A
su
rv
e
y
,
"
IEE
E
T
ra
n
sa
c
ti
o
n
s
o
n
p
a
tt
e
rn
a
n
a
lys
is
a
n
d
ma
c
h
i
n
e
i
n
telli
g
e
n
c
e
,
v
o
l.
2
4
,
n
o
.
1
,
p
p
.
34
-
58
,
2
0
0
2
,
d
o
i
:
1
0
.
1
1
0
9
/
3
4
.
9
8
2
8
8
3
.
[2
]
He
,
Ka
i
m
in
g
,
Jia
n
S
u
n
,
a
n
d
Xia
o
o
u
Tan
g
,
"
G
u
id
e
d
ima
g
e
fil
teri
n
g
,
"
E
u
ro
p
e
a
n
c
o
n
fer
e
n
c
e
o
n
c
o
mp
u
ter
v
isio
n
.
S
p
rin
g
e
r,
Be
rli
n
,
He
id
e
l
b
e
rg
,
2
0
1
0
,
d
o
i:
1
0
.
1
1
0
9
/T
P
A
M
I.
2
0
1
2
.
2
1
3
.
[3
]
Kirb
y
,
M
ich
a
e
l,
a
n
d
Law
re
n
c
e
S
iro
v
ich
,
"
A
p
p
li
c
a
ti
o
n
o
f
t
h
e
Ka
rh
u
n
e
n
-
Lo
e
v
e
p
r
o
c
e
d
u
re
f
o
r
t
h
e
c
h
a
r
a
c
teriz
a
ti
o
n
o
f
h
u
m
a
n
fa
c
e
s,
"
IEE
E
T
ra
n
sa
c
ti
o
n
s
o
n
Pa
tt
e
r
n
a
n
a
lys
is
a
n
d
M
a
c
h
i
n
e
in
telli
g
e
n
c
e
,
v
o
l.
1
2
,
n
o
.
1
,
p
p
.
1
0
3
-
1
0
8
,
1
9
9
0
.
[4
]
Tu
rk
M
a
tt
h
e
w,
a
n
d
Ale
x
P
e
n
tl
a
n
d
,
"
E
ig
e
n
fa
c
e
s fo
r
re
c
o
g
n
it
i
o
n
,
"
J
o
u
rn
a
l
o
f
c
o
g
n
it
ive
n
e
u
ro
sc
ien
c
e
,
v
o
l.
3
,
n
o
.
1
p
p
.
71
-
86
,
1
9
9
1
,
d
o
i:
1
0
.
1
1
6
2
/j
o
c
n
.
1
9
9
1
.
3
.
1
.
7
1
.
[5
]
Zh
a
o
,
Wen
y
i,
A
Krish
n
a
sw
a
m
y
,
Ra
m
a
Ch
e
ll
a
p
p
a
,
Da
n
iel
L.
S
we
ts,
a
n
d
Jo
h
n
Wen
g
,
"
Disc
rimin
a
n
t
a
n
a
ly
sis
o
f
p
rin
c
i
p
a
l
c
o
m
p
o
n
e
n
ts
fo
r
fa
c
e
re
c
o
g
n
i
ti
o
n
,
"
Fa
c
e
Rec
o
g
n
i
ti
o
n
.
S
p
r
in
g
e
r,
Be
rli
n
,
He
id
e
lb
e
r
g
,
p
p
.
73
-
85
,
1
9
9
8
,
d
o
i
:
1
0
.
1
1
0
9
/A
F
G
R.
1
9
9
8
.
6
7
0
9
7
1
.
[6
]
Lu
,
G
u
i
-
F
u
,
Jia
n
Zo
u
,
a
n
d
Y
o
n
g
Wan
g
,
"I
n
c
re
m
e
n
tal
c
o
m
p
lete
L
DA
f
o
r
fa
c
e
re
c
o
g
n
it
io
n
,
"
Pa
tt
e
rn
Rec
o
g
n
i
ti
o
n
,
v
o
l.
4
5
,
n
o
.
7
,
p
p
.
2
5
1
0
-
2
5
2
1
,
2
0
1
2
,
d
o
i:
1
0
.
1
0
1
6
/j
.
p
a
tco
g
.
2
0
1
2
.
0
1
.
0
1
8
.
[7
]
D.
A.
M
e
e
d
e
n
i
y
a
,
a
n
d
D.
A.
A.
C.
Ra
tn
a
we
e
ra
,
"
En
h
a
n
c
e
d
fa
c
e
re
c
o
g
n
it
io
n
t
h
ro
u
g
h
v
a
riati
o
n
o
f
p
ri
n
c
i
p
le
c
o
m
p
o
n
e
n
t
a
n
a
ly
sis
(
P
CA),
"
2
0
0
7
In
ter
n
a
ti
o
n
a
l
Co
n
fer
e
n
c
e
o
n
I
n
d
u
stri
a
l
a
n
d
In
fo
rm
a
t
io
n
S
y
ste
ms
,
IEE
E,
2
0
0
7
,
d
o
i:
1
0
.
1
1
0
9
/ICIINF
S
.
2
0
0
7
.
4
5
7
9
2
0
0
.
[8
]
Ba
n
sa
l,
Ab
h
is
h
e
k
,
Ka
p
i
l
M
e
h
ta,
a
n
d
S
a
h
il
Aro
ra
,
"
F
a
c
e
re
c
o
g
n
i
t
i
o
n
u
sin
g
P
CA
a
n
d
LDA
a
lg
o
rit
h
m
,
"
2
0
1
2
se
c
o
n
d
in
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
ti
n
g
&
Co
m
mu
n
ica
t
io
n
T
e
c
h
n
o
l
o
g
ies
.
I
EE
E,
2
0
1
2
,
d
o
i
:
1
0
.
1
1
0
9
/ACCT
.
2
0
1
2
.
5
2
.
[9
]
Be
lh
u
m
e
u
r,
P
e
ter
N.
,
J
o
ã
o
P
.
H
e
sp
a
n
h
a
,
a
n
d
Da
v
id
J.
Krie
g
m
a
n
,
"
Ei
g
e
n
fa
c
e
s
v
s.
f
ish
e
rfa
c
e
s:
Re
c
o
g
n
i
ti
o
n
u
sin
g
c
las
s sp
e
c
ifi
c
li
n
e
a
r
p
ro
jec
ti
o
n
,
"
I
EE
E
T
ra
n
sa
c
ti
o
n
s o
n
p
a
tt
e
rn
a
n
a
lys
is a
n
d
m
a
c
h
i
n
e
in
telli
g
e
n
c
e
,
v
o
l.
1
9
,
n
o
.
7
,
p
p
.
711
-
7
2
0
,
1
9
9
7
,
d
o
i:
1
0
.
1
0
0
7
/BF
b
0
0
1
5
5
2
2
.
[1
0
]
An
,
G
a
o
Yu
n
,
a
n
d
Qi
u
Qi
R
u
a
n
,
"
No
v
e
l
m
a
th
e
m
a
ti
c
a
l
m
o
d
e
l
fo
r
e
n
h
a
n
c
e
d
f
ish
e
r'
s
li
n
e
a
r
d
isc
ri
m
in
a
n
t
a
n
d
it
s
a
p
p
li
c
a
ti
o
n
t
o
fa
c
e
re
c
o
g
n
it
i
o
n
,
"
1
8
t
h
I
n
ter
n
a
t
io
n
a
l
Co
n
fer
e
n
c
e
o
n
P
a
tt
e
rn
Rec
o
g
n
it
io
n
(IC
PR
'0
6
)
,
IEE
E
,
v
o
l.
2
,
2
0
0
6
,
p
p
.
5
2
4
-
5
2
7
,
d
o
i:
1
0
.
1
1
0
9
/I
CP
R.
2
0
0
6
.
8
7
3
.
[1
1
]
Zh
o
u
,
Wei
,
Xia
o
r
o
n
g
P
u
,
a
n
d
Zi
m
in
g
Zh
e
n
g
,
"
P
a
rts
-
b
a
se
d
h
o
li
stic
fa
c
e
re
c
o
g
n
it
i
o
n
wit
h
RBF
n
e
u
ra
l
n
e
t
wo
rk
s,
"
In
ter
n
a
ti
o
n
a
l
S
y
mp
o
siu
m
o
n
Ne
u
r
a
l
Ne
tw
o
rk
s
,
S
p
ri
n
g
e
r,
Be
rli
n
,
He
id
e
l
b
e
rg
,
2
0
0
6
,
d
o
i:
1
0
.
1
0
0
7
/
1
1
7
6
0
0
2
3
_
1
7
.
[1
2
]
Ab
u
sh
a
m
,
E
ima
d
El
d
in
,
Da
v
i
d
Ng
o
,
a
n
d
A
n
d
re
w
Te
o
h
,
"
F
u
s
io
n
o
f
lo
c
a
ll
y
li
n
e
a
r
e
m
b
e
d
d
in
g
a
n
d
p
rin
c
i
p
a
l
c
o
m
p
o
n
e
n
t
a
n
a
ly
sis
f
o
r
fa
c
e
re
c
o
g
n
it
i
o
n
(F
LL
E
P
CA),
"
I
n
ter
n
a
ti
o
n
a
l
Co
n
fer
e
n
c
e
o
n
P
a
tt
e
rn
Rec
o
g
n
it
io
n
a
n
d
Ima
g
e
An
a
lys
is
.
S
p
ri
n
g
e
r,
Be
rli
n
,
He
id
e
l
b
e
rg
,
2
0
0
5
,
d
o
i:
1
0
.
1
0
0
7
/1
1
5
5
2
4
9
9
_
3
7
.
[1
3
]
Ah
o
n
e
n
,
Ti
m
o
,
A
b
d
e
n
o
u
r
Ha
d
id
,
a
n
d
M
a
tt
i
P
ietik
ä
i
n
e
n
,
"F
a
c
e
re
c
o
g
n
i
ti
o
n
wit
h
l
o
c
a
l
b
i
n
a
ry
p
a
tt
e
rn
s,
"
E
u
ro
p
e
a
n
c
o
n
fer
e
n
c
e
o
n
c
o
mp
u
ter
v
isio
n
.
S
p
rin
g
e
r,
Be
rli
n
,
He
id
e
lb
e
r
g
,
2
0
0
4
.
[
1
4
]
A
h
o
n
e
n
T
i
m
o
,
E
s
a
R
a
h
t
u
,
V
i
l
l
e
O
j
a
n
s
i
v
u
,
a
n
d
J
a
n
n
e
H
e
i
k
k
i
l
a
,
"
R
e
c
o
g
n
i
t
i
o
n
o
f
b
l
u
r
r
e
d
f
a
c
e
s
u
s
i
n
g
l
o
c
a
l
p
h
a
s
e
q
u
a
n
t
i
z
a
t
i
o
n
,
"
2
0
0
8
1
9
t
h
i
n
t
e
r
n
a
t
i
o
n
a
l
c
o
n
f
e
r
e
n
c
e
o
n
p
a
t
t
e
r
n
r
e
c
o
g
n
i
t
i
o
n
.
I
E
E
E
,
2
0
0
8
,
d
o
i
:
1
0
.
1
1
0
9
/
I
C
P
R
.
2
0
0
8
.
4
7
6
1
8
4
7
.
[1
5
]
Ch
i
Ho
Ch
a
n
,
M
A
Tah
ir
,
Jo
se
f
Kitt
ler,
a
n
d
M
P
ietik
a
i
n
e
n
,
"
M
u
lt
isc
a
le
lo
c
a
l
p
h
a
se
q
u
a
n
t
iza
ti
o
n
fo
r
r
o
b
u
s
t
c
o
m
p
o
n
e
n
t
-
ba
se
d
fa
c
e
re
c
o
g
n
it
i
o
n
u
si
n
g
k
e
r
n
e
l
fu
sio
n
o
f
m
u
lt
i
p
le
d
e
sc
rip
to
rs,
"
IEE
E
T
ra
n
s
a
c
ti
o
n
s
o
n
Pa
tt
e
rn
An
a
lys
is
a
n
d
M
a
c
h
i
n
e
In
tell
ig
e
n
c
e
,
v
o
l
.
3
5
,
n
o
.
5
,
p
p
.
1
1
6
4
-
1
1
7
7
,
2
0
1
2
,
d
o
i:
1
0
.
1
1
0
9
/T
P
A
M
I.
2
0
1
2
.
1
9
9
.
[1
6
]
Zh
o
u
Ch
a
n
g
j
u
n
,
L
Wan
g
,
Q
Z
h
a
n
g
,
a
n
d
X
Wei
,
"
F
a
c
e
re
c
o
g
n
it
io
n
b
a
se
d
o
n
P
CA
ima
g
e
re
c
o
n
stru
c
t
io
n
a
n
d
LDA,
"
Op
ti
k
,
v
o
l.
1
2
4
,
n
o
.
22
,
p
p
.
5
5
9
9
-
5
6
0
3
,
2
0
1
3
,
d
o
i:
1
0
.
1
0
1
6
/
j
.
ij
leo
.
2
0
1
3
.
0
4
.
1
0
8
.
[1
7
]
Da
i,
Ka
n
k
a
n
,
Jia
n
we
i
Z
h
a
o
,
a
n
d
F
e
il
o
n
g
Ca
o
,
"
A
n
o
v
e
l
d
e
c
o
rre
lat
e
d
n
e
u
ra
l
n
e
two
r
k
e
n
se
m
b
le
a
l
g
o
rit
h
m
fo
r
fa
c
e
re
c
o
g
n
it
i
o
n
,
"
K
n
o
wle
d
g
e
-
B
a
se
d
S
y
ste
ms
,
v
o
l.
89
,
p
p
.
5
4
1
-
5
5
2
,
2
0
1
5
,
d
o
i:
1
0
.
1
0
1
6
/j
.
k
n
o
sy
s.
2
0
1
5
.
0
9
.
0
0
2
.
[1
8
]
P
a
ti
l,
He
m
p
ra
sa
d
Y.
,
As
h
wi
n
G
.
Ko
th
a
ri,
a
n
d
Kish
o
r
M
.
B
h
u
rc
h
a
n
d
i,
"
E
x
p
re
ss
io
n
in
v
a
rian
t
fa
c
e
re
c
o
g
n
i
ti
o
n
u
sin
g
lo
c
a
l
b
in
a
r
y
p
a
tt
e
rn
s
a
n
d
c
o
n
to
u
rlet
tran
sfo
r
m,
"
Op
t
ik
,
v
o
l.
1
2
7
,
n
o
.
5
,
p
p
.
2
6
7
0
-
2
6
7
8
,
2
0
1
6
,
d
o
i
:
1
0
.
1
0
1
6
/
j.
ij
leo
.
2
0
1
5
.
1
1
.
1
8
7
.
[1
9
]
Kh
a
n
S
a
ji
d
Ali
,
M
Ish
ti
a
q
,
M
Na
z
ir,
a
n
d
M
S
h
a
h
n
e
e
n
,
"
F
a
c
e
re
c
o
g
n
it
i
o
n
u
n
d
e
r
v
a
ry
i
n
g
e
x
p
re
ss
io
n
s
a
n
d
il
lu
m
in
a
ti
o
n
u
s
in
g
p
a
rti
c
le
sw
a
rm
o
p
ti
m
iza
ti
o
n
,
"
J
o
u
r
n
a
l
o
f
c
o
mp
u
ta
ti
o
n
a
l
sc
ien
c
e
,
v
o
l.
28
,
p
p
.
94
-
100
,
2
0
1
8
,
d
o
i:
1
0
.
1
0
1
6
/
j.
jo
c
s.
2
0
1
8
.
0
8
.
0
0
5
.
[2
0
]
Tai
Yin
g
,
J
Ya
n
g
,
Y
Z
h
a
n
g
,
L
L
u
o
,
J
Qia
n
,
a
n
d
Y
C
h
e
n
,
"
F
a
c
e
re
c
o
g
n
i
ti
o
n
wit
h
p
o
se
v
a
riati
o
n
s
a
n
d
m
isa
li
g
n
m
e
n
t
v
ia
o
rth
o
g
o
n
a
l
p
ro
c
ru
ste
s
re
g
re
ss
io
n
,
"
IEE
E
T
ra
n
s
a
c
ti
o
n
s
o
n
Im
a
g
e
Pro
c
e
ss
in
g
,
v
o
l
.
2
5
,
n
o
.
6
,
p
p
.
2
6
7
3
-
2
6
8
3
,
2
0
1
6
,
d
o
i:
1
0
.
1
1
0
9
/
TIP
.
2
0
1
6
.
2
5
5
1
3
6
2
.
[2
1
]
Li
Ju
n
,
Yu
Ko
n
g
,
H
Z
h
a
o
,
J
Ya
n
g
,
a
n
d
Yu
n
F
u
,
"
Lea
rn
i
n
g
fa
st
lo
w
-
ra
n
k
p
ro
j
e
c
ti
o
n
fo
r
ima
g
e
c
las
sifica
ti
o
n
,
"
IEE
E
T
ra
n
sa
c
ti
o
n
s
o
n
Ima
g
e
Pro
c
e
ss
in
g
,
v
o
l.
2
5
,
n
o
.
10
,
p
p
.
4
8
0
3
-
4
8
1
4
,
2
0
1
6
,
d
o
i:
1
0
.
1
1
0
9
/
TIP
.
2
0
1
6
.
2
5
9
8
6
5
4
.
[2
2
]
Ch
e
n
Yo
n
g
,
T
H
u
a
n
g
,
H
Li
u
,
a
n
d
Di
Zh
a
n
,
"
M
u
lt
i
-
p
o
se
fa
c
e
e
n
se
m
b
le
c
las
sifica
ti
o
n
a
id
e
d
b
y
G
a
b
o
r
fe
a
tu
re
s
a
n
d
d
e
e
p
b
e
li
e
f
n
e
ts,
"
O
p
ti
k
,
v
o
l.
1
2
7
,
n
o
.
2
,
p
p
.
9
4
6
-
9
5
4
,
2
0
1
6
,
d
o
i
:
1
0
.
1
0
1
6
/j
.
ij
leo
.
2
0
1
5
.
1
0
.
1
7
9
.
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
A
n
effec
tive
fa
ce
r
ec
o
g
n
itio
n
meth
o
d
u
s
in
g
g
u
id
e
d
ima
g
e
fil
ter a
n
d
co
n
v
o
lu
tio
n
a
l…
(
Ya
lla
ma
n
d
a
ia
h
S
)
1707
[2
3
]
Yin
,
Xi,
a
n
d
Xia
o
m
i
n
g
Li
u
,
"
M
u
lt
i
-
tas
k
c
o
n
v
o
lu
ti
o
n
a
l
n
e
u
ra
l
n
e
t
wo
rk
fo
r
p
o
se
-
in
v
a
rian
t
fa
c
e
re
c
o
g
n
it
io
n
,
"
IEE
E
T
ra
n
sa
c
ti
o
n
s
o
n
Ima
g
e
Pro
c
e
ss
in
g
,
v
o
l.
2
7
,
n
o
.
2
,
p
p
.
9
6
4
-
9
7
5
,
2
0
1
7
,
d
o
i:
1
0
.
1
1
0
9
/T
IP
.
2
0
1
7
.
2
7
6
5
8
3
0
.
[2
4
]
G
e
n
m
in
g
Din
g
,
Ju
n
Ti
a
n
,
Jin
s
o
n
g
Wu
,
Qia
n
Zh
a
o
,
a
n
d
Li
li
Xie
,
"
E
n
e
rg
y
e
fficie
n
t
h
u
m
a
n
a
c
ti
v
it
y
re
c
o
g
n
it
io
n
u
sin
g
we
a
ra
b
le
se
n
so
rs,
"
2
0
1
8
I
EE
E
W
ire
les
s
Co
mm
u
n
ica
ti
o
n
s
a
n
d
Ne
two
rk
in
g
Co
n
fer
e
n
c
e
W
o
rk
sh
o
p
s
(W
CNCW
)
.
IEE
E,
2
0
1
8
,
d
o
i:
1
0
.
1
1
0
9
/W
CNCW.
2
0
1
8
.
8
3
6
8
9
8
0
.
[2
5
]
Zh
ifan
g
Li
a
o
,
Le
G
a
o
,
Ti
a
n
Zh
o
u
,
Xia
o
p
i
n
g
F
a
n
,
Ya
n
Zh
a
n
g
,
a
n
d
J
in
so
n
g
W
u
,
"
An
o
il
p
a
i
n
ters
re
c
o
g
n
it
i
o
n
m
e
th
o
d
b
a
se
d
o
n
c
l
u
ste
r
m
u
l
ti
p
le
k
e
r
n
e
l
lea
rn
i
n
g
a
lg
o
rit
h
m
,
"
IE
EE
Acc
e
ss
,
v
o
l.
7
,
p
p
.
2
6
8
4
2
-
2
6
8
5
4
,
2
0
1
9
,
d
o
i
:
1
0
.
1
1
0
9
/ACCE
S
S
.
2
0
1
9
.
2
8
9
9
3
8
9
.
[2
6
]
M
u
q
e
e
t,
M
o
h
d
Ab
d
u
l,
a
n
d
Ra
g
h
u
n
a
th
S
.
Ho
lam
b
e
,
"
L
o
c
a
l
b
in
a
r
y
p
a
tt
e
rn
s
b
a
se
d
o
n
d
irec
ti
o
n
a
l
wa
v
e
let
tran
sf
o
rm
fo
r
e
x
p
re
ss
io
n
a
n
d
p
o
se
-
in
v
a
rian
t
fa
c
e
re
c
o
g
n
i
ti
o
n
,
"
Ap
p
li
e
d
Co
m
p
u
ti
n
g
a
n
d
In
f
o
rm
a
ti
c
s
,
v
o
l.
1
5
,
n
o
.
2
,
p
p
.
1
6
3
-
171
,
2
0
1
9
,
d
o
i:
1
0
.
1
0
1
6
/
j.
a
c
i.
2
0
1
7
.
1
1
.
0
0
2
.
[2
7
]
G
e
o
ffre
y
E.
Hi
n
to
n
,
N S
riv
a
sta
v
a
,
A
Kriz
h
e
v
sk
y
,
I
S
u
ts
k
e
v
e
r,
a
n
d
R
R.
S
a
lak
h
u
t
d
in
o
v
,
"
Im
p
ro
v
in
g
n
e
u
ra
l
n
e
two
r
k
s
b
y
p
re
v
e
n
ti
n
g
c
o
-
a
d
a
p
tatio
n
o
f
fe
a
tu
re
d
e
tec
to
rs,
"
a
rXiv
p
re
p
rin
t
a
rXiv:1
2
0
7
.
0
5
8
0
,
2
0
1
2
.
[2
8
]
Taig
m
a
n
Ya
n
iv
,
M
Ya
n
g
,
M
A
R
a
n
z
a
to
,
a
n
d
Li
o
r
W
o
lf
,
"
De
e
p
fa
c
e
:
Clo
si
n
g
th
e
g
a
p
t
o
h
u
m
a
n
-
le
v
e
l
p
e
rfo
rm
a
n
c
e
in
fa
c
e
v
e
rifi
c
a
ti
o
n
,
"
Pro
c
e
e
d
in
g
s
o
f
th
e
IEE
E
c
o
n
fer
e
n
c
e
o
n
c
o
mp
u
ter
v
isio
n
a
n
d
p
a
t
ter
n
re
c
o
g
n
it
io
n
,
2
0
1
4
,
d
o
i:
1
0
.
1
1
0
9
/CVP
R
.
2
0
1
4
.
2
2
0
.
[2
9
]
Jin
so
n
g
Wu
,
S
o
n
g
G
u
o
,
Hu
a
w
e
i
Hu
a
n
g
,
Wi
l
l
iam
Li
u
,
a
n
d
Y
o
n
g
Xia
n
g
,
"
I
n
fo
rm
a
ti
o
n
a
n
d
c
o
m
m
u
n
ica
ti
o
n
s
tec
h
n
o
l
o
g
ies
f
o
r
su
sta
in
a
b
le
d
e
v
e
lo
p
m
e
n
t
g
o
a
ls:
sta
te
-
of
-
t
h
e
-
a
rt,
n
e
e
d
s
a
n
d
p
e
rsp
e
c
ti
v
e
s,
"
IE
EE
C
o
mm
u
n
ic
a
ti
o
n
s
S
u
rv
e
y
s &
T
u
to
ri
a
ls
,
v
o
l
.
2
0
,
n
o
.
3
,
p
p
.
2
3
8
9
-
2
4
0
6
,
2
0
1
8
,
d
o
i:
1
0
.
1
1
0
9
/COM
S
T.
2
0
1
8
.
2
8
1
2
3
0
1
.
[3
0
]
Ra
c
h
a
d
Ata
t,
Li
n
g
ji
a
Li
u
,
Ji
n
so
n
g
W
u
,
G
u
a
n
g
y
u
Li
,
Ch
u
n
x
u
a
n
Y
e
,
a
n
d
Ya
n
g
Yi
,
"
Big
d
a
ta
m
e
e
t
c
y
b
e
r
-
p
h
y
si
c
a
l
sy
ste
m
s: A
p
a
n
o
ra
m
ic su
rv
e
y
,
"
I
EE
E
Acc
e
ss
,
v
o
l
.
6
,
p
p
.
7
3
6
0
3
-
7
3
6
3
6
,
2
0
1
8
.
[3
1
]
G
u
o
ju
n
Li
n
,
M
Ya
n
g
,
J
Ya
n
g
,
L
S
h
e
n
,
a
n
d
W
Xie
,
"
Ro
b
u
st
,
d
isc
ri
m
in
a
ti
v
e
a
n
d
c
o
m
p
re
h
e
n
siv
e
d
ict
io
n
a
r
y
lea
rn
i
n
g
fo
r
fa
c
e
re
c
o
g
n
it
io
n
,
"
Pa
tt
e
rn
Rec
o
g
n
it
io
n
,
v
o
l.
8
1
,
p
p
.
3
4
1
-
3
5
6
,
2
0
1
8
,
d
o
i:
1
0
.
1
0
1
6
/
j.
p
a
tco
g
.
2
0
1
8
.
0
3
.
0
2
1
.
[3
2
]
G
ö
rg
e
l,
P
e
li
n
,
a
n
d
A
h
m
e
t
S
ims
e
k
,
"
F
a
c
e
re
c
o
g
n
it
io
n
v
ia
d
e
e
p
sta
c
k
e
d
d
e
n
o
is
in
g
sp
a
rse
a
u
to
e
n
c
o
d
e
rs
(DSDS
A),
"
A
p
p
li
e
d
M
a
t
h
e
ma
ti
c
s
a
n
d
Co
m
p
u
t
a
ti
o
n
,
v
o
l.
3
5
5
,
p
p
.
3
2
5
-
3
4
2
,
2
0
1
9
,
d
o
i:
1
0
.
1
0
1
6
/
j.
a
m
c
.
2
0
1
9
.
0
2
.
0
7
1
.
[3
3
]
To
m
a
si,
Ca
rlo
,
a
n
d
Ro
b
e
rto
M
a
n
d
u
c
h
i,
"
Bil
a
tera
l
fi
lt
e
rin
g
f
o
r
g
ra
y
a
n
d
c
o
lo
r
ima
g
e
s,
"
S
ixth
i
n
ter
n
a
ti
o
n
a
l
c
o
n
fer
e
n
c
e
o
n
c
o
mp
u
ter
v
isio
n
(I
EE
E
Ca
t
.
No
.
9
8
CH3
6
2
7
1
)
.
IEE
E
,
1
9
9
8
,
d
o
i:
1
0
.
1
1
0
9
/ICCV.
1
9
9
8
.
7
1
0
8
1
5
.
[3
4
]
Du
ra
n
d
,
F
ré
d
o
,
a
n
d
Ju
li
e
Do
rse
y
,
"
F
a
st
b
il
a
tera
l
fil
teri
n
g
fo
r
th
e
d
is
p
l
a
y
o
f
h
i
g
h
-
d
y
n
a
m
ic
-
ra
n
g
e
ima
g
e
s,
"
Pro
c
e
e
d
in
g
s
o
f
t
h
e
2
9
th
a
n
n
u
a
l
c
o
n
fer
e
n
c
e
o
n
Co
mp
u
ter
g
ra
p
h
ics
a
n
d
i
n
ter
a
c
ti
v
e
tec
h
n
iq
u
e
s
,
2
0
0
2
,
d
o
i
:
1
0
.
1
1
4
5
/
5
6
6
6
5
4
.
5
6
6
5
7
4
.
[3
5
]
Ba
e
,
S
o
o
n
m
in
,
S
y
l
v
a
in
P
a
ris,
a
n
d
F
ré
d
o
D
u
ra
n
d
,
"
Two
-
sc
a
le
t
o
n
e
m
a
n
a
g
e
m
e
n
t
fo
r
p
h
o
to
g
ra
p
h
ic
l
o
o
k
,
"
AC
M
T
ra
n
sa
c
ti
o
n
s
o
n
Gr
a
p
h
ics
(T
OG
)
,
v
o
l
.
2
5
,
n
o
.
3
,
p
p
.
6
3
7
-
6
4
5
,
2
0
0
6
,
d
o
i
:
1
0
.
1
1
4
5
/
1
1
4
1
9
1
1
.
1
1
4
1
9
3
5
.
[3
6
]
Zee
v
F
a
rb
m
a
n
,
Ra
a
n
a
n
F
a
tt
a
l,
D
a
n
i
Li
sc
h
in
s
k
i,
a
n
d
Rich
a
rd
S
z
e
li
sk
i
,
"
E
d
g
e
-
p
re
se
rv
in
g
d
e
c
o
m
p
o
si
ti
o
n
s
f
o
r
m
u
lt
i
-
sc
a
le
to
n
e
a
n
d
d
e
tail
m
a
n
ip
u
lati
on,
"
AC
M
T
ra
n
sa
c
ti
o
n
s
o
n
Gr
a
p
h
ics
(T
OG
)
,
v
o
l
.
2
7
,
n
o
.
3
,
p
p
.
1
-
1
0
,
2
0
0
8
,
d
o
i:
1
0
.
1
1
4
5
/
1
3
6
0
6
1
2
.
1
3
6
0
6
6
6
.
[3
7
]
He
,
Ka
imin
g
,
Jia
n
S
u
n
,
a
n
d
Xia
o
o
u
Ta
n
g
,
"
G
u
id
e
d
ima
g
e
fil
teri
n
g
,
"
IEE
E
tr
a
n
sa
c
ti
o
n
s
o
n
p
a
tt
e
rn
a
n
a
lys
is
a
n
d
ma
c
h
in
e
i
n
telli
g
e
n
c
e
,
v
o
l.
3
5
,
n
o
.
6
,
p
p
.
1
3
9
7
-
1
4
0
9
,
2
0
1
2
,
d
o
i:
1
0
.
1
1
0
9
/T
P
AMI
.
2
0
1
2
.
2
1
3
.
[3
8
]
O.
Oliv
e
tt
i,
“
Ora
c
l
e
re
se
a
r
c
h
lab
o
ra
to
ry
fa
c
e
d
a
tab
a
se
o
f
fa
c
e
s,”
[
3
9
]
M
.
J
.
L
y
o
n
s
,
S
.
A
k
a
m
a
t
s
u
,
M
.
Ka
m
a
c
h
i
,
J
.
G
y
o
b
a
,
a
n
d
J
.
B
u
d
y
n
e
k
,
“
T
h
e
j
a
p
a
n
e
s
e
f
e
m
a
le
fa
c
i
a
l
e
x
p
r
e
s
s
i
o
n
(
j
a
ff
e
)
d
a
t
a
b
a
se
,
”
P
r
o
c
e
e
d
i
n
g
s
o
f
t
h
i
r
d
i
n
t
e
r
n
a
t
i
o
n
a
l
c
o
n
f
e
r
e
n
c
e
o
n
a
u
t
o
m
a
t
i
c
f
a
c
e
a
n
d
g
e
s
t
u
r
e
r
e
c
o
g
n
i
t
i
o
n
,
1
9
9
8
,
p
p
.
1
4
-
16
.
[4
0
]
P
.
N.
Be
lh
u
m
e
u
r.
,
J.
P
.
H
e
sp
a
n
h
a
.
,
a
n
d
D.
J.
Krie
g
m
a
n
,
“
Ei
g
e
n
fa
c
e
s
v
s.fi
s
h
e
rfa
c
e
s:
Re
c
o
g
n
it
io
n
u
sin
g
c
l
a
ss
sp
e
c
ifi
c
li
n
e
a
r
p
r
o
jec
ti
o
n
,
”
IEE
E
T
ra
n
sa
c
t
io
n
s
o
n
Pa
tt
e
rn
An
a
lys
is & M
a
c
h
i
n
e
In
telli
g
e
n
c
e
,
v
o
l.
7
,
p
p
.
7
1
1
-
720
,
1
9
9
7
.
B
I
O
G
RAP
H
I
E
S O
F
AUTH
O
RS
Ya
ll
a
m
a
n
d
a
i
a
h
S
re
c
e
iv
e
d
h
is
B.
Tec
h
d
e
g
re
e
fro
m
JN
TUH,
H
y
d
e
ra
b
a
d
,
a
n
d
M
.
Tec
h
d
e
g
re
e
fro
m
JN
TUK,
Ka
k
in
a
d
a
.
Cu
rre
n
tl
y
,
h
e
is
a
Re
se
a
rc
h
S
c
h
o
lar
a
t
th
e
S
ENS
E,
VIT
-
AP
Un
iv
e
rsity
,
In
d
ia.
His res
e
a
rc
h
field
s
in
c
l
u
d
e
Im
a
g
e
P
ro
c
e
ss
in
g
a
n
d
M
a
c
h
i
n
e
L
e
a
rn
in
g
.
Pu
r
n
a
c
h
a
n
d
N
r
e
c
e
iv
e
d
h
is
M
.
Tec
h
d
e
g
re
e
fro
m
VIT
U
n
iv
e
rsi
ty
,
Ve
ll
o
re
,
In
d
ia,
a
n
d
P
h
D
d
e
g
re
e
fro
m
Un
iv
e
rsit
y
o
f
Av
e
i
ro
,
Av
e
iro
P
o
rt
u
g
a
l.
C
u
rre
n
tl
y
,
h
e
is
wo
rk
i
n
g
a
s
As
so
c
iate
P
ro
fe
ss
o
r
a
t
VIT
-
AP
Un
i
v
e
rsity
,
In
d
ia.
His
fiel
d
s
o
f
re
se
a
rc
h
c
o
n
t
a
in
Im
a
g
e
P
ro
c
e
ss
in
g
,
Vi
d
e
o
Pr
o
c
e
ss
in
g
,
a
n
d
P
a
tt
e
rn
Re
c
o
g
n
it
i
o
n
.
Evaluation Warning : The document was created with Spire.PDF for Python.