I
nd
o
ne
s
ia
n J
o
urna
l o
f
E
lect
rica
l En
g
ineering
a
nd
Co
m
p
u
t
er
Science
Vo
l.
10
,
No
.
2
,
May
201
8
,
p
p
.
5
6
2
~5
6
8
I
SS
N:
2502
-
4752
,
DOI
: 1
0
.
1
1
5
9
1
/
i
j
ee
cs
.
v
1
0
.
i2
.
p
p
562
-
5
6
8
562
J
o
ur
na
l ho
m
ep
a
g
e
:
h
ttp
:
//ia
e
s
co
r
e.
co
m/jo
u
r
n
a
ls
/in
d
ex
.
p
h
p
/
ijeec
s
Dev
elo
p
m
e
nt
o
f
E
ng
lish Handw
ritt
en Recog
nition
U
sing
Deep N
eura
l N
et
w
o
rk
T
eddy
Su
ry
a
G
un
a
w
a
n*
1
,
Ah
m
a
d F
a
k
hrur
Ra
zi
M
o
hd
No
o
r
2
,
M
ira
K
a
rt
i
w
i
3
1
,
2
De
p
a
rtm
e
n
t
o
f
El
e
c
tri
c
a
l
a
n
d
C
o
m
p
u
ter E
n
g
in
e
e
rin
g
,
K
u
ll
iy
y
a
h
o
f
En
g
in
e
e
rin
g
,
M
a
lay
sia
3
De
p
a
rtm
e
n
t
o
f
In
f
o
r
m
a
ti
o
n
S
y
st
e
m
s,
Ku
ll
iy
y
a
h
o
f
ICT
,
In
tern
a
ti
o
n
a
l
Isla
m
ic Un
iv
e
rsit
y
M
a
la
y
sia
,
M
a
la
y
sia
Art
icle
I
nfo
AB
ST
RAC
T
A
r
ticle
his
to
r
y:
R
ec
eiv
ed
N
ov
13
,
2
0
1
7
R
ev
i
s
ed
J
an
2
6
,
2
0
1
8
A
cc
ep
ted
F
eb
15
,
2
0
1
8
Du
e
to
th
e
a
d
v
a
n
c
e
d
in
G
P
U an
d
CP
U,
i
n
re
c
e
n
t
y
e
a
rs,
De
e
p
Ne
u
ra
l
Ne
tw
o
r
k
(DN
N)
b
e
c
o
m
e
s
p
o
p
u
lar
to
b
e
u
ti
li
z
e
d
b
o
th
a
s
f
e
a
tu
re
e
x
tra
c
ti
o
n
a
n
d
c
las
si
f
ier.
T
h
is
p
a
p
e
r
a
ims
to
d
e
v
e
lo
p
o
f
f
li
n
e
h
a
n
d
w
rit
ten
re
c
o
g
n
it
i
o
n
sy
ste
m
u
sin
g
DN
N.
F
irst,
tw
o
p
o
p
u
lar
En
g
li
sh
d
ig
it
s
a
n
d
letters
d
a
tab
a
se
,
i.
e
.
M
NIST
a
n
d
EM
NIST
,
w
e
re
se
l
e
c
ted
to
p
ro
v
i
d
e
d
a
tas
e
t
f
o
r
tra
in
in
g
a
n
d
tes
ti
n
g
p
h
a
se
o
f
DN
N.
A
lt
o
g
e
th
e
r,
th
e
re
a
re
1
0
d
ig
it
s
[
0
-
9
]
a
n
d
5
2
letters
[
a
-
z
,
A
-
Z]
.
T
h
e
p
ro
p
o
se
d
DN
N
u
s
e
d
sta
c
k
e
d
t
w
o
a
u
to
e
n
c
o
d
e
r
lay
e
r
s
a
n
d
o
n
e
so
f
t
m
a
x
la
y
e
r.
Re
c
o
g
n
it
io
n
a
c
c
u
r
a
c
y
f
o
r
En
g
li
sh
d
ig
it
s
a
n
d
letters
is
9
7
.
7
%
a
n
d
8
8
.
8
%
,
re
sp
e
c
ti
v
e
ly
.
P
e
rf
o
rm
a
n
c
e
c
o
m
p
a
riso
n
w
it
h
o
th
e
r
stru
c
tu
re
o
f
n
e
u
ra
l
n
e
tw
o
rk
s
re
v
e
a
led
th
a
t
th
e
w
e
i
g
h
ted
a
v
e
ra
g
e
r
e
c
o
g
n
it
io
n
ra
t
e
f
o
r
p
a
tt
e
rn
n
e
t,
f
e
e
d
f
o
rwa
rd
n
e
t,
a
n
d
p
ro
p
o
se
d
DN
N
w
e
re
8
0
.
3
%
,
6
8
.
3
%
,
a
n
d
9
0
.
4
%
,
re
sp
e
c
ti
v
e
l
y
.
It
sh
o
w
s
th
a
t
o
u
r
p
r
o
p
o
se
d
sy
ste
m
is
a
b
le
to
re
c
o
g
n
ize
h
a
n
d
w
rit
ten
E
n
g
li
sh
d
ig
it
s a
n
d
let
ters
w
it
h
h
ig
h
a
c
c
u
ra
c
y
.
K
ey
w
o
r
d
s
:
Dee
p
n
eu
r
al
n
e
t
w
o
r
k
E
MN
I
ST
Han
d
w
r
itte
n
r
ec
o
g
n
it
io
n
MN
I
ST
Neu
r
al
n
et
w
o
r
k
Co
p
y
rig
h
t
©
2
0
1
8
In
stit
u
te o
f
A
d
v
a
n
c
e
d
E
n
g
i
n
e
e
rin
g
a
n
d
S
c
ien
c
e
.
Al
l
rig
h
ts re
se
rv
e
d
.
C
o
r
r
e
s
p
o
nd
ing
A
uth
o
r
:
T
ed
d
y
S
u
r
y
a
G
u
n
a
w
an
,
De
p
a
rtme
n
t
o
f
El
e
c
tri
c
a
l
a
n
d
Co
m
p
u
ter E
n
g
in
e
e
rin
g
,
Ku
ll
iy
y
a
h
o
f
En
g
in
e
e
rin
g
,
M
a
la
y
sia
.
.
E
m
ail:
ts
g
u
n
a
w
a
n
@iiu
m
.
e
d
u
.
m
y
1.
I
NT
RO
D
UCT
I
O
N
Han
d
w
r
iti
n
g
s
ar
e
th
e
m
o
s
t
co
m
m
o
n
s
ta
n
d
ar
d
an
d
r
eg
u
l
ar
m
ed
i
u
m
t
h
at
h
u
m
a
n
b
ein
g
s
u
s
e
in
co
m
m
u
n
icati
n
g
.
I
t
is
a
ls
o
a
n
ef
f
ec
tiv
e
an
d
e
f
f
icie
n
t
w
a
y
to
r
ec
o
r
d
in
f
o
r
m
atio
n
e
v
en
w
it
h
th
e
i
n
tr
o
d
u
ctio
n
o
f
n
e
w
tec
h
n
o
lo
g
ies,
li
k
e
s
w
i
f
tk
e
y
k
e
y
b
o
ar
d
,
s
o
u
n
d
co
m
m
a
n
d
,
etc.
Gen
er
all
y
,
h
a
n
d
w
r
iti
n
g
r
e
co
g
n
itio
n
s
y
s
te
m
i
s
a
m
ec
h
a
n
i
s
m
t
h
at
u
s
ed
to
r
ec
o
g
n
ize
h
u
m
a
n
h
a
n
d
w
r
itte
n
in
an
y
la
n
g
u
a
g
es
eit
h
er
f
r
o
m
s
ca
n
n
ed
h
a
n
d
w
r
itte
n
i
m
a
g
e
o
r
r
ea
l
tim
e
h
an
d
w
r
iti
n
g
u
s
in
g
s
t
y
l
u
s
p
en
o
n
elec
tr
o
n
ic
d
ev
ice
w
h
ic
h
also
ca
n
b
e
ca
lled
as
o
f
f
li
n
e
an
d
o
n
lin
e
h
a
n
d
w
r
iti
n
g
r
esp
ec
ti
v
e
l
y
[
1
]
.
B
esid
es,
t
h
e
ap
p
licatio
n
o
f
th
is
s
y
s
te
m
ca
n
b
e
ca
t
eg
o
r
ized
in
to
th
r
ee
w
h
ic
h
ar
e
n
u
m
er
al,
c
h
ar
ac
ter
an
d
cu
r
s
i
v
e
w
o
r
d
.
I
t
is
w
id
el
y
u
s
ed
in
n
u
m
er
o
u
s
ap
p
licatio
n
s
s
u
c
h
as
lan
g
u
ag
e
tr
an
s
latio
n
,
b
an
k
c
h
eq
u
e
s
an
d
k
e
y
w
o
r
d
s
p
o
ttin
g
[
2
,
3
]
.
Fig
u
r
e
1
.
A
T
y
p
ical
Ha
n
d
w
r
itt
en
R
ec
o
g
n
it
io
n
S
y
s
te
m
Evaluation Warning : The document was created with Spire.PDF for Python.
I
n
d
o
n
esia
n
J
E
lec
E
n
g
&
C
o
m
p
Sci
I
SS
N:
2502
-
4752
Dev
elo
p
men
t o
f E
n
g
lis
h
Ha
n
d
w
r
itten
R
ec
o
g
n
itio
n
Usi
n
g
Dee
p
.
.
.
(
Ted
d
y
s
u
r
ya
Gu
n
a
w
a
n
)
563
As
s
h
o
w
n
in
Fi
g
u
r
e
1
,
th
e
co
m
m
o
n
p
r
o
ce
s
s
es
o
f
h
a
n
d
w
r
iti
n
g
r
ec
o
g
n
itio
n
s
y
s
te
m
ar
e
im
a
g
e
ac
q
u
is
itio
n
,
p
r
ep
r
o
ce
s
s
in
g
,
s
e
g
m
e
n
tat
io
n
,
f
ea
tu
r
e
e
x
tr
ac
tio
n
an
d
cla
s
s
i
f
icatio
n
[
2
]
.
I
m
ag
e
ac
q
u
is
itio
n
is
th
e
f
ir
s
t
s
tep
to
g
e
t
an
i
m
a
g
e
f
o
r
m
o
f
h
a
n
d
w
r
itte
n
th
at
w
i
ll
ac
t
as
in
p
u
t
to
p
r
ep
r
o
ce
s
s
in
g
b
y
u
s
i
n
g
s
ca
n
n
er
.
T
h
e
p
r
ep
r
o
ce
s
s
in
g
s
tep
i
s
to
r
e
m
o
v
e
n
o
is
e
o
r
d
is
to
r
tio
n
s
o
f
th
e
s
c
an
n
ed
i
m
a
g
e
w
h
ich
u
s
ab
le
f
o
r
f
u
r
t
h
er
p
r
o
ce
s
s
[
3
,
4]
.
On
e
o
f
th
e
p
r
o
ce
s
s
in
p
r
ep
r
o
ce
s
s
in
g
i
s
th
r
e
s
h
o
ld
i
n
g
w
h
ic
h
co
n
v
er
t sca
n
n
ed
i
m
a
g
e
i
n
to
b
in
ar
y
i
m
a
g
e.
Nex
t
,
s
eg
m
e
n
tatio
n
i
s
u
s
ed
to
d
iv
id
e
ea
ch
w
o
r
d
in
to
s
u
b
-
i
m
a
g
e
s
w
h
er
e
th
is
s
tag
e
is
v
er
y
i
m
p
o
r
ta
n
t
s
tep
p
ar
ticu
lar
l
y
f
o
r
co
n
ti
n
u
o
u
s
h
a
n
d
w
r
iti
n
g
in
o
r
d
er
to
ex
tr
ac
t f
ea
t
u
r
es
f
r
o
m
ea
ch
i
m
a
g
e
o
f
ch
ar
ac
ter
w
h
ic
h
w
ill b
e
p
er
f
o
r
m
i
n
f
ea
t
u
r
e
ex
tr
ac
tio
n
p
r
o
ce
s
s
[
1
]
.
T
h
en
p
r
o
ce
ed
s
w
ith
f
ea
t
u
r
e
ex
tr
ac
tio
n
w
h
er
e
it
w
ill
e
x
tr
ac
t
ev
er
y
c
h
ar
ac
ter
is
tic
o
f
th
e
f
ea
tu
r
es
i
n
ea
ch
i
m
a
g
e.
T
h
is
f
ea
t
u
r
e
is
v
er
y
u
s
e
f
u
l
f
o
r
class
i
f
icatio
n
in
th
e
last
s
tep
.
T
h
er
e
ar
e
m
an
y
clas
s
i
f
icatio
n
tech
n
i
q
u
es
s
u
ch
as
K
-
Nea
r
es
t
Neig
h
b
o
u
r
(
KNN)
,
Ne
u
r
al
Net
w
o
r
k
an
d
Su
p
p
o
r
t
Vec
to
r
Ma
ch
i
n
e
(
SVM
)
w
h
er
e
t
h
ese
clas
s
i
f
ier
s
h
a
v
e
d
if
f
er
en
t
ap
p
r
o
ac
h
to
r
ec
o
g
n
ize
t
h
e
i
m
a
g
e
[
4
]
.
Ge
n
er
all
y
,
m
o
s
t
r
e
s
ea
r
ch
er
ev
alu
a
te
th
e
p
er
f
o
r
m
a
n
ce
o
f
t
h
e
s
y
s
te
m
b
ased
o
n
clas
s
i
f
icatio
n
ac
cu
r
ac
y
[
5
,
6
]
.
A
lt
h
o
u
g
h
m
a
n
y
p
ap
er
h
as
b
ee
n
co
n
d
u
cted
o
n
o
f
f
li
n
e
h
an
d
w
r
itte
n
r
ec
o
g
n
it
io
n
,
b
u
t
t
h
e
u
s
e
o
f
Dee
p
Neu
r
al
Ne
t
w
o
r
k
(
DNN)
is
s
til
l
in
th
e
ea
r
l
y
s
ta
g
e.
T
h
er
ef
o
r
e,
th
e
o
b
j
ec
tiv
e
o
f
th
is
p
ap
er
is
to
d
ev
elo
p
o
f
f
lin
e
h
an
d
w
r
itte
n
r
ec
o
g
n
i
tio
n
u
s
i
n
g
DNN.
W
e
w
ill
u
s
e
t
w
o
p
o
p
u
lar
d
atab
ase,
in
c
lu
d
i
n
g
MN
I
S
T
[
7
]
an
d
E
MN
I
ST
[
8
]
d
u
e
to
th
e
clea
n
d
ata
p
r
o
v
id
ed
.
T
h
u
s
,
t
h
e
al
g
o
r
ith
m
s
t
h
at
w
i
ll
b
e
u
s
ed
i
n
t
h
is
p
r
o
j
e
ct
s
tar
t
w
i
th
f
ea
t
u
r
e
ex
tr
ac
tio
n
w
h
er
e
w
e
w
ill
u
s
e
i
m
a
g
e
p
ix
el
as
f
ea
t
u
r
e
in
p
u
t
t
o
class
if
ier
[
9
]
as
DNN
w
ill
h
an
d
le
b
o
th
f
ea
t
u
r
e
ex
tr
ac
tio
n
a
n
d
clas
s
i
f
ier
.
Fi
n
all
y
,
w
e
w
i
ll
ev
a
lu
ate
th
e
p
er
f
o
r
m
an
ce
o
f
t
h
is
p
r
o
p
o
s
ed
p
r
o
j
ec
t
b
ased
o
n
co
n
f
u
s
io
n
m
atr
i
x
to
f
i
n
d
th
e
cl
ass
i
f
icatio
n
ac
cu
r
ac
y
.
2.
DE
S
I
G
N
O
F
P
RO
P
O
SE
D
H
ANDWR
I
T
T
E
N
R
E
CO
G
N
I
T
I
O
N
SY
ST
E
M
Fig
u
r
e
2
s
h
o
w
s
th
e
p
r
o
p
o
s
ed
h
an
d
w
r
itte
n
r
ec
o
g
n
itio
n
s
y
s
te
m
.
T
h
e
p
r
o
ce
s
s
in
g
s
tep
in
cl
u
d
es
i
m
ag
e
th
r
es
h
o
ld
in
g
,
ch
ar
ac
ter
t
h
i
n
n
in
g
u
s
in
g
m
o
r
p
h
o
lo
g
ical
o
p
er
atio
n
,
s
lan
t
co
r
r
ec
tio
n
,
an
d
f
in
al
l
y
i
m
a
g
e
s
eg
m
e
n
tatio
n
.
T
h
e
p
ix
el
v
al
u
e
f
r
o
m
i
m
a
g
e
s
e
g
m
e
n
tat
io
n
w
il
l
b
e
tr
ea
ted
as
th
e
in
co
m
i
n
g
in
p
u
t
to
th
e
DNN,
in
w
h
ic
h
b
o
th
f
ea
t
u
r
e
ex
tr
ac
tio
n
an
d
class
i
f
icat
io
n
w
ill b
e
co
n
d
u
cted
.
Fig
u
r
e
2
.
P
r
o
p
o
s
ed
Han
d
w
r
itt
en
R
ec
o
g
n
it
io
n
S
y
s
te
m
2
.
1
F
ea
t
ure
E
x
t
ra
ct
io
n a
nd
Cla
s
s
if
ier
Us
ing
Dee
p Ne
ura
l N
e
t
w
o
rk
(
DNN)
I
n
th
i
s
p
r
o
j
ec
t,
w
e
o
n
l
y
u
s
ed
th
e
p
i
x
el
v
a
lu
e
o
f
ea
ch
s
e
g
m
en
ted
i
m
a
g
es
as
f
ea
t
u
r
e
in
p
u
t
to
Dee
p
Neu
r
al
Net
w
o
r
k
(
DNN)
.
T
h
e
i
m
ag
e
s
ize
f
r
o
m
d
ataset
ar
e
2
8
-
by
-
2
8
p
ix
els
in
g
r
a
y
s
ca
le
w
h
i
ch
m
ea
n
s
o
n
e
d
i
g
i
t
in
o
n
e
i
m
ag
e.
W
h
en
w
e
ar
e
wo
r
k
in
g
w
it
h
i
m
a
g
es,
t
h
e
f
ea
t
u
r
es
w
e
u
s
e
is
t
h
e
r
a
w
p
i
x
el
v
al
u
es.
Si
n
ce
t
h
e
i
m
a
g
e
h
as
2
8
x
2
8
p
ix
els
a
s
ill
u
s
tr
ated
in
Fi
g
u
r
e
3
th
u
s
it
eq
u
al
to
7
8
4
p
ix
els,
s
o
w
e
h
a
v
e
7
8
4
in
p
u
t
f
ea
tu
r
es
f
o
r
ea
ch
i
m
a
g
e.
I
n
2
0
0
6
,
d
ee
p
b
elief
n
et
w
o
r
k
w
as
i
n
tr
o
d
u
ce
d
b
y
[
1
0
]
w
h
i
ch
f
o
r
m
s
a
D
NN
m
o
d
el
co
m
p
o
s
ed
o
f
R
estricte
d
B
o
ltz
m
an
n
Ma
c
h
i
n
es
(
R
B
M
)
.
I
t
is
tr
ain
ed
i
n
u
n
s
u
p
er
v
i
s
ed
f
a
s
h
io
n
,
o
n
e
la
y
er
at
a
ti
m
e
f
r
o
m
th
e
lo
w
es
t
to
th
e
h
i
g
h
est
la
y
er
.
An
o
th
er
v
ar
iatio
n
,
i.e
.
d
ee
p
f
ee
d
-
f
o
r
w
ar
d
n
et
w
o
r
k
s
,
w
er
e
e
f
f
ec
tiv
el
y
tr
ai
n
ed
u
s
in
g
th
e
s
a
m
e
id
ea
b
y
f
ir
s
t
p
r
e
-
t
r
ain
in
g
ea
c
h
la
y
er
a
s
a
R
B
M,
th
e
n
f
i
n
e
-
tu
n
i
n
g
b
y
b
ac
k
p
r
o
p
ag
atio
n
[
1
1
]
.
No
w
ad
a
y
s
,
ANNs
w
it
h
d
ee
p
s
tr
u
ct
u
r
e
ar
e
tr
ain
ed
o
n
p
o
w
e
r
f
u
l
GP
U
m
ac
h
i
n
es,
o
v
er
co
m
in
g
b
o
th
r
eso
u
r
ce
s
an
d
tr
ain
i
n
g
t
i
m
e
li
m
i
ts
.
Oth
er
t
y
p
e
s
o
f
DNN
i
n
cl
u
d
es
r
ec
u
r
r
en
t
n
e
u
r
al
n
et
w
o
r
k
s
(
R
NN)
an
d
co
n
v
e
n
tio
n
a
l
n
e
u
r
al
n
et
w
o
r
k
s
(
C
NN)
.
An
R
N
N
is
s
i
m
ilar
to
an
A
NN
ex
ce
p
t
t
h
at
it
allo
w
s
a
s
elf
-
co
n
n
ec
ted
h
id
d
en
la
y
er
th
at
as
s
o
ciate
s
w
it
h
a
ti
m
e
d
ela
y
.
A
C
NN
is
s
i
m
il
ar
to
an
o
r
d
in
ar
y
f
ee
d
f
o
r
w
ar
d
n
eu
r
al
n
et
w
o
r
k
as
th
e
o
u
tp
u
t
o
f
ea
ch
la
y
er
is
a
Evaluation Warning : The document was created with Spire.PDF for Python.
I
SS
N
:
2
5
0
2
-
4752
I
n
d
o
n
esia
n
J
E
lec
E
n
g
&
C
o
m
p
Sci,
Vo
l.
10
,
No
.
2
,
Ma
y
2
0
1
8
:
5
6
2
–
5
6
8
564
co
m
b
i
n
atio
n
o
f
th
e
in
p
u
t,
w
eig
h
t
m
atr
i
x
,
an
d
th
e
b
ias
v
ec
to
r
f
o
llo
w
ed
b
y
a
n
o
n
-
li
n
ea
r
tr
an
s
f
o
r
m
at
io
n
.
Ho
w
e
v
er
,
C
NN
h
as
lo
ca
l
co
n
n
ec
t
iv
i
t
y
,
in
w
h
ic
h
a
C
NN
u
s
e
s
s
o
m
e
s
m
all
f
il
ter
s
,
s
lid
e
s
it
ac
r
o
s
s
a
ll
s
u
b
-
r
eg
io
n
s
o
f
th
e
in
p
u
t
m
atr
i
x
a
n
d
ag
g
r
eg
a
tes
al
l
r
es
u
lts
.
Me
an
s
th
at,
it
ta
k
es
ad
v
a
n
ta
g
e
s
o
f
t
h
e
co
n
v
o
lu
tio
n
o
p
er
atio
n
b
et
w
ee
n
t
h
e
f
ilter
s
a
n
d
th
e
i
n
p
u
t.
A
C
NN
is
n
o
r
m
a
ll
y
co
n
s
is
ted
o
f
s
e
v
er
al
t
y
p
e
s
o
f
la
y
er
s
,
i
n
cl
u
d
in
g
co
n
v
o
lu
tio
n
al
la
y
er
,
p
o
o
lin
g
la
y
er
,
an
d
f
u
ll
y
-
co
n
n
ec
ted
la
y
er
.
Fig
u
r
e
3
.
E
x
a
m
p
le
o
f
P
ix
el
Va
lu
e
o
f
a
n
Im
a
g
e
as
Fe
at
u
r
e
In
p
u
t to
DNN
2
.
2
Select
ed
H
a
nd
w
rit
t
en
I
m
a
g
e
Da
t
a
ba
s
es
T
h
e
d
atab
ase
u
s
ed
f
o
r
th
is
r
es
ea
r
ch
h
av
e
t
w
o
ca
teg
o
r
ies
w
h
i
ch
ar
e
d
ig
it
an
d
alp
h
ab
et.
Dig
it
d
atab
ase
is
f
r
o
m
th
e
MN
I
ST
d
atab
ase
o
f
h
a
n
d
w
r
i
tten
d
i
g
its
w
h
er
e
t
h
is
d
atab
ase
co
n
ta
in
s
6
0
0
0
0
e
x
a
m
p
le
s
o
f
tr
ai
n
in
g
s
et
an
d
1
0
0
0
0
ex
a
m
p
les
o
f
te
s
t
s
et
[
4
]
.
T
h
e
d
ig
it
s
ar
e
all
a
u
n
i
f
o
r
m
s
ize
a
n
d
ar
e
ce
n
tr
ed
in
t
h
e
i
m
a
g
e.
E
ac
h
i
m
a
g
e
is
ce
n
tr
ed
in
a
2
8
x
2
8
i
m
ag
e
b
y
co
m
p
u
ti
n
g
t
h
e
ce
n
tr
e
o
f
m
a
s
s
o
f
th
e
p
i
x
els,
a
n
d
tr
an
s
lat
in
g
t
h
e
i
m
ag
e
s
o
th
at
t
h
e
ce
n
tr
e
o
f
m
a
s
s
i
s
at
th
e
ce
n
tr
e
o
f
th
e
i
m
ag
e.
E
ac
h
i
m
ag
e
i
s
a
b
in
ar
y
i
m
ag
e
o
f
a
h
an
d
w
r
i
tten
d
i
g
it
an
d
co
n
s
is
ts
o
f
3
0
,
0
0
0
p
atter
n
s
f
r
o
m
t
w
o
d
atab
ase
s
,
o
n
e
f
r
o
m
C
e
n
s
u
s
B
u
r
ea
u
e
m
p
lo
y
ee
s
a
n
d
t
h
e
o
th
er
f
r
o
m
h
i
g
h
-
s
ch
o
o
l
s
t
u
d
en
t
s
,
f
r
o
m
ab
o
u
t
2
5
0
d
if
f
er
en
t
w
r
iter
s
i
n
to
tal.
T
h
e
test
s
et
co
n
s
is
ts
o
f
5
,
0
0
0
p
atter
n
s
f
r
o
m
ea
c
h
d
atab
ase.
T
h
e
d
atasets
ar
e
lab
elled
to
en
ab
le
ea
s
y
v
er
i
f
icati
o
n
.
T
h
e
s
a
m
p
le
o
f
tr
ai
n
i
n
g
i
m
ag
e
f
o
r
MN
I
ST
as
s
h
o
w
n
in
F
ig
u
r
e
4.
Ho
w
e
v
er
,
w
e
o
n
l
y
u
s
ed
o
n
e
t
y
p
e
o
f
d
ataset
s
f
r
o
m
t
h
e
l
i
s
t
w
h
ich
is
E
MI
NST
L
etter
s
f
o
r
th
is
p
r
o
p
o
s
ed
p
r
o
j
ec
t.
T
h
e
E
MN
I
ST
L
etter
s
d
ataset
m
er
g
es
a
b
alan
ce
d
s
et
o
f
t
h
e
u
p
p
er
ca
s
e
an
d
lo
w
er
ca
s
e
o
f
h
an
d
w
r
itte
n
letter
s
in
to
a
s
i
n
g
l
e
2
6
-
class
tas
k
.
T
h
e
tr
ain
i
n
g
s
a
m
p
le
f
o
r
o
u
r
p
r
o
p
o
s
ed
p
r
o
j
ec
t is 1
2
4
8
0
0
an
d
test
s
a
m
p
le
is
2
0
8
0
0
.
So
m
e
o
f
tr
ai
n
in
g
i
m
a
g
es sa
m
p
le
ar
e
s
h
o
w
n
in
Fi
g
u
r
e
5
.
Fig
u
r
e
4
.
MN
I
ST
T
r
ain
in
g
I
m
ag
e
Sa
m
p
le
s
Fig
u
r
e
5
.
E
MN
I
ST
T
r
ain
in
g
I
m
ag
e
Sa
m
p
le
s
I
n
ad
d
itio
n
,
letter
d
atab
ase
u
s
ed
f
o
r
th
i
s
p
r
o
j
ec
t
is
f
r
o
m
E
x
t
en
d
ed
Mo
d
if
ied
NI
ST
(
E
MN
I
ST
)
.
B
o
th
E
MN
I
ST
an
d
MN
I
S
T
w
a
s
d
er
iv
ed
f
r
o
m
NI
ST
Sp
ec
ial
Data
b
ase
1
9
[
8
]
.
T
h
e
co
n
v
er
s
io
n
o
f
i
m
a
g
es
f
r
o
m
t
h
e
o
r
ig
in
al
u
s
ed
t
h
e
s
a
m
e
m
et
h
o
d
o
lo
g
y
as
i
n
[
4
]
.
T
h
er
e
ar
e
s
i
x
d
if
f
er
en
t
s
p
lit
s
p
r
o
v
id
ed
in
t
h
is
d
atase
t.
A
s
h
o
r
t
s
u
m
m
ar
y
o
f
t
h
e
d
ataset
i
s
p
r
o
v
id
ed
as f
o
llo
w
s
:
•
E
MN
I
ST
B
y
C
las
s
: 8
1
4
,
2
5
5
ch
ar
ac
ter
s
.
6
2
u
n
b
ala
n
ce
d
class
e
s
.
Evaluation Warning : The document was created with Spire.PDF for Python.
I
n
d
o
n
esia
n
J
E
lec
E
n
g
&
C
o
m
p
Sci
I
SS
N:
2502
-
4752
Dev
elo
p
men
t o
f E
n
g
lis
h
Ha
n
d
w
r
itten
R
ec
o
g
n
itio
n
Usi
n
g
Dee
p
.
.
.
(
Ted
d
y
s
u
r
ya
Gu
n
a
w
a
n
)
565
•
E
MN
I
ST
B
y
Me
r
g
e:
8
1
4
,
2
5
5
c
h
ar
ac
ter
s
.
4
7
u
n
b
alan
ce
d
clas
s
es.
•
E
MN
I
ST
B
alan
ce
d
:
1
3
1
,
6
0
0
ch
ar
ac
ter
s
.
4
7
b
alan
ce
d
class
e
s
.
•
E
MN
I
ST
L
etter
s
: 1
4
5
,
6
0
0
ch
ar
ac
ter
s
.
2
6
b
alan
ce
d
class
es.
•
E
MN
I
ST
Dig
its
: 2
8
0
,
0
0
0
ch
ar
ac
ter
s
.
1
0
b
alan
ce
d
class
es.
•
E
MN
I
ST
MN
I
S
T
: 7
0
,
0
0
0
ch
ar
ac
ter
s
.
1
0
b
alan
ce
d
class
es.
3.
I
M
P
L
E
M
E
NT
AT
I
O
N
O
F
D
E
E
P
N
E
URA
L
NE
T
WO
RK
I
n
t
h
is
p
r
o
p
o
s
ed
p
r
o
j
ec
t,
Dee
p
Neu
r
al
Net
w
o
r
k
(
DNN)
i
s
u
s
ed
as
f
ea
tu
r
e
ex
tr
ac
tio
n
a
n
d
cl
ass
i
f
ier
o
f
th
e
h
a
n
d
w
r
itte
n
s
y
s
te
m
.
W
e
co
n
d
u
cted
tr
ain
i
n
g
an
d
test
i
n
g
p
h
as
e
o
f
DNN
o
n
th
e
d
i
f
f
er
e
n
t
s
a
m
p
les.
3
.
1
T
ra
ini
ng
P
ha
s
e
o
f
t
he
D
NN
Dee
p
Neu
r
al
Net
w
o
r
k
(
DNN)
is
a
n
et
w
o
r
k
th
a
t
co
n
s
i
s
t
o
f
m
an
y
h
id
d
en
la
y
er
s
w
i
th
d
if
f
er
e
n
t
n
u
m
b
er
o
f
n
e
u
r
o
n
i
n
ea
ch
la
y
er
.
Fo
r
th
is
r
esear
c
h
,
a
s
tac
k
ed
au
to
en
co
d
er
s
f
o
r
h
an
d
w
r
iti
n
g
r
ec
o
g
n
itio
n
is
u
s
ed
to
tr
ai
n
m
u
ltip
le
la
y
er
s
.
T
h
e
n
u
m
b
er
h
id
d
en
la
y
er
u
s
ed
is
th
r
ee
,
in
c
l
u
d
in
g
t
w
o
h
id
d
en
la
y
er
s
an
d
o
n
e
s
o
f
t
m
a
x
la
y
er
in
w
h
ic
h
t
h
ese
t
h
r
ee
la
y
er
s
w
i
ll
b
e
s
tack
ed
to
g
e
th
er
i
n
o
r
d
er
to
f
o
r
m
d
ee
p
n
et
w
o
r
k
.
T
h
e
f
ir
s
t
an
d
s
ec
o
n
d
la
y
er
w
il
l
b
e
tr
ain
ed
w
it
h
o
u
t
u
s
i
n
g
lab
el
f
r
o
m
tr
ai
n
i
n
g
d
ata
w
h
ic
h
m
ea
n
s
u
n
s
u
p
er
v
i
s
ed
f
as
h
io
n
[
1
2
]
.
So
f
tm
a
x
la
y
er
d
if
f
e
r
f
r
o
m
h
id
d
en
la
y
er
in
wh
ich
w
e
tr
ai
n
ed
th
i
s
la
y
er
w
it
h
s
u
p
er
v
is
ed
f
a
s
h
io
n
u
s
in
g
la
b
els
f
r
o
m
tr
ain
in
g
d
ataset.
T
h
is
au
to
e
n
co
d
er
u
s
e
s
r
e
g
u
lar
izer
s
to
lear
n
a
s
p
ar
s
e
r
ep
r
esen
tatio
n
i
n
t
h
e
f
i
r
s
t
la
y
er
.
L
2
W
eig
h
tR
e
g
u
lar
izatio
n
co
n
tr
o
ls
th
e
i
m
p
ac
t
o
f
a
n
L
2
r
eg
u
la
r
izer
f
o
r
th
e
w
ei
g
h
t
s
o
f
th
e
n
e
t
w
o
r
k
(
a
n
d
n
o
t
th
e
b
iases
)
.
Sp
ar
s
it
y
R
e
g
u
lar
izat
io
n
co
n
tr
o
ls
t
h
e
i
m
p
ac
t
o
f
a
s
p
ar
s
it
y
r
eg
u
lar
izer
,
w
h
ic
h
att
e
m
p
ts
to
en
f
o
r
ce
a
co
n
s
tr
ain
t
o
n
t
h
e
s
p
ar
s
i
t
y
o
f
t
h
e
o
u
tp
u
t
f
r
o
m
t
h
e
h
id
d
en
la
y
er
.
No
te
t
h
at,
t
h
i
s
i
s
d
i
f
f
er
e
n
t
f
r
o
m
ap
p
l
y
i
n
g
a
s
p
ar
s
it
y
r
eg
u
lar
izer
to
th
e
w
ei
g
h
t
s
.
Sp
ar
s
it
y
P
r
o
p
o
r
tio
n
is
a
p
ar
am
eter
o
f
t
h
e
s
p
ar
s
it
y
r
e
g
u
lar
izer
.
I
t
co
n
tr
o
ls
th
e
s
p
ar
s
it
y
o
f
th
e
o
u
tp
u
t
f
r
o
m
th
e
h
id
d
en
la
y
er
.
A
lo
w
v
alu
e
f
o
r
Sp
ar
s
it
y
P
r
o
p
o
r
tio
n
u
s
u
all
y
lead
s
to
ea
ch
n
eu
r
o
n
in
t
h
e
h
id
d
en
la
y
er
s
p
e
cializin
g
b
y
o
n
l
y
g
i
v
i
n
g
a
h
i
g
h
o
u
tp
u
t
f
o
r
a
s
m
all
n
u
m
b
er
o
f
tr
ain
in
g
e
x
a
m
p
le
s
.
Fo
r
ex
a
m
p
le,
if
Sp
ar
s
it
y
P
r
o
p
o
r
tio
n
is
s
et
to
0
.
1
,
th
is
is
eq
u
i
v
alen
t
to
s
a
y
i
n
g
t
h
at
ea
ch
n
e
u
r
o
n
in
t
h
e
h
id
d
en
la
y
er
s
h
o
u
ld
h
av
e
a
n
a
v
er
ag
e
o
u
tp
u
t
o
f
0
.
1
o
v
er
t
h
e
tr
ain
i
n
g
ex
a
m
p
le
s
.
T
h
is
v
a
lu
e
m
u
s
t
b
e
b
et
w
ee
n
0
an
d
1
.
T
h
e
id
ea
l
v
alu
e
v
ar
ies
d
ep
en
d
in
g
o
n
t
h
e
n
a
tu
r
e
o
f
th
e
p
r
o
b
le
m
.
T
h
e
co
n
f
ig
u
r
atio
n
u
s
ed
f
o
r
d
ig
its
an
d
let
ter
s
th
at
a
f
f
ec
t t
h
e
p
er
f
o
r
m
a
n
ce
o
f
o
u
r
test
ar
e
illu
s
tr
ated
in
T
ab
le
1
.
T
ab
le
1
.
C
o
n
f
ig
u
r
atio
n
o
f
DN
N
C
l
a
ss
N
u
mb
e
r
o
f
c
h
a
r
a
c
t
e
r
N
u
mb
e
r
o
f
h
i
d
d
e
n
l
a
y
e
r
N
u
mb
e
r
o
f
n
o
d
e
s
i
n
i
n
p
u
t
l
a
y
e
r
N
u
mb
e
r
o
f
n
e
u
r
o
n
s
i
n
1
st
h
i
d
d
e
n
l
a
y
e
r
N
u
mb
e
r
o
f
n
e
u
r
o
n
s
i
n
2
nd
h
i
d
d
e
n
l
a
y
e
r
N
u
mb
e
r
o
f
n
e
u
r
o
n
s
i
n
3
rd
h
i
d
d
e
n
l
a
y
e
r
D
i
g
i
t
s
10
3
7
8
4
3
0
0
50
10
L
e
t
t
e
r
s
26
3
7
8
4
3
0
0
50
27
W
e
u
s
ed
s
m
aller
n
u
m
b
er
o
f
n
eu
r
o
n
s
i
n
f
ir
s
t
h
id
d
en
la
y
er
co
m
p
ar
ed
to
in
p
u
t
o
f
th
e
DN
N.
Sin
ce
o
u
r
d
ig
its
i
m
a
g
e
s
a
m
p
le
w
ill
h
a
v
e
7
8
4
-
by
-
6
0
0
0
0
w
h
ile
le
tter
s
i
m
ag
e
s
a
m
p
le
co
n
s
i
s
t
o
f
7
8
4
-
by
-
1
2
4
8
0
0
as
in
p
u
t
to
DNN,
w
e
s
et
n
u
m
b
er
o
f
n
e
u
r
o
n
s
f
o
r
f
ir
s
t
h
id
d
en
la
y
er
i
s
3
0
0
.
Seco
n
d
la
y
er
w
ill
h
a
v
e
1
0
0
w
h
ile
1
0
o
r
2
7
f
o
r
s
o
f
t
m
ax
la
y
er
.
L
es
s
n
u
m
b
er
o
f
n
e
u
r
o
n
s
w
i
ll
m
ak
e
t
h
e
au
to
en
co
d
er
lear
n
s
s
m
al
ler
an
d
co
m
p
r
ess
ed
r
ep
r
esen
tatio
n
o
f
t
h
e
i
n
p
u
t
o
f
ev
er
y
la
y
er
.
I
n
th
is
tr
ai
n
in
g
p
r
o
ce
s
s
,
th
e
in
p
u
t
o
f
ea
c
h
la
y
er
w
ill
u
s
e
e
x
tr
a
ct
f
ea
t
u
r
e
f
r
o
m
p
r
ev
io
u
s
en
co
d
er
as tr
ain
i
n
g
d
ata
to
tr
ain
t
h
e
la
y
er
.
Fig
u
r
e
6
.
Stack
ed
L
a
y
er
s
o
f
D
NN
Evaluation Warning : The document was created with Spire.PDF for Python.
I
SS
N
:
2
5
0
2
-
4752
I
n
d
o
n
esia
n
J
E
lec
E
n
g
&
C
o
m
p
Sci,
Vo
l.
10
,
No
.
2
,
Ma
y
2
0
1
8
:
5
6
2
–
5
6
8
566
Af
ter
tr
ai
n
ed
t
h
r
ee
s
ep
ar
ate
l
a
y
er
s
o
f
a
D
NN,
w
e
w
i
ll
s
ta
ck
t
h
o
s
e
t
h
r
ee
la
y
er
s
to
g
et
h
er
to
f
o
r
m
a
d
ee
p
n
et
w
o
r
k
,
as
s
h
o
w
n
i
n
Fi
g
u
r
e
6
.
Fu
r
t
h
er
m
o
r
e,
w
e
co
m
p
u
te
th
e
r
es
u
lts
w
i
th
te
s
ti
n
g
d
ataset
u
s
in
g
t
h
e
f
u
ll
d
ee
p
n
et
w
o
r
k
f
o
r
m
ed
.
I
n
o
r
d
er
to
in
cr
ea
s
e
th
e
p
er
f
o
r
m
a
n
c
e
o
f
d
ee
p
n
et
w
o
r
k
,
w
e
t
u
n
e
th
e
d
ee
p
n
et
w
o
r
k
b
y
r
etr
ain
in
g
it
u
s
i
n
g
tr
ain
i
n
g
d
ataset
in
s
u
p
er
v
i
s
ed
f
a
s
h
io
n
w
h
i
ch
m
ea
n
s
i
n
cl
u
d
in
g
tr
ai
n
in
g
la
b
el
d
ata.
3
.
2
T
esting
P
ha
s
e
o
f
t
he
DNN
T
esti
n
g
is
t
h
e
last
p
ar
t
o
f
h
an
d
w
r
iti
n
g
r
ec
o
g
n
it
io
n
in
o
r
d
er
to
ev
alu
ate
t
h
e
DNN.
T
o
test
th
e
n
et
w
o
r
k
,
w
e
n
ee
d
to
h
av
e
te
s
t
d
ataset
a
lo
n
g
w
i
th
te
s
t
lab
el
o
f
i
m
a
g
es
.
W
ith
th
e
f
u
ll
d
ee
p
n
et
w
o
r
k
f
o
r
m
ed
an
d
tr
ai
n
ed
,
w
e
test
th
e
n
e
t
w
o
r
k
u
s
i
n
g
te
s
t
d
ataset.
B
ased
o
n
t
h
e
test
,
w
e
co
m
p
u
te
t
h
e
r
e
s
u
l
ts
o
f
t
h
e
s
y
s
te
m
s
u
c
h
a
s
r
ec
o
g
n
itio
n
ac
cu
r
ac
y
,
p
er
f
o
r
m
an
ce
a
n
d
p
er
ce
n
tag
e
er
r
o
r
.
R
es
u
lts
o
f
h
a
n
d
w
r
i
tin
g
r
ec
o
g
n
itio
n
s
y
s
te
m
u
s
in
g
DNN
d
iv
id
ed
in
to
t
w
o
s
in
c
e
d
if
f
er
en
t
in
d
atab
ase.
On
e
f
o
r
d
ig
it
r
ec
o
g
n
itio
n
an
d
t
h
e
o
th
er
f
o
r
letter
r
ec
o
g
n
itio
n
.
Ou
r
test
p
er
f
o
r
m
a
n
ce
is
b
ased
o
n
th
e
co
n
f
u
s
io
n
m
atr
i
x
.
4.
RE
SU
L
T
S AN
D
D
I
SCU
SS
I
O
N
I
n
th
i
s
s
ec
tio
n
,
t
h
e
ex
p
er
i
m
e
n
tal
s
et
u
p
alo
n
g
w
it
h
th
e
tes
t
in
g
o
f
th
e
tr
ai
n
ed
DNN
o
n
t
h
e
s
elec
ted
d
ataset
w
ill b
e
elab
o
r
ated
.
4
.
1
E
x
peri
m
e
nta
l Set
up
A
ll
ex
p
er
i
m
en
t
w
ill
b
e
co
n
d
u
cted
u
s
i
n
g
a
L
en
o
v
o
G5
0
0
s
lap
to
p
w
ith
th
e
f
o
llo
w
i
n
g
s
p
e
cif
icatio
n
s
,
I
n
tel
P
en
ti
u
m
2
0
2
0
M,
8
G
B
R
A
M,
W
i
n
d
o
w
s
1
0
P
r
o
o
p
er
atin
g
s
y
s
te
m
s
.
T
h
e
s
o
f
t
w
ar
e
u
s
ed
f
o
r
t
h
i
s
ex
p
er
i
m
e
n
t
is
M
A
T
L
A
B
R
2
0
1
7
a.
T
h
is
v
er
s
io
n
w
a
s
s
u
p
p
o
r
ted
w
it
h
Neu
r
al
Net
w
o
r
k
T
o
o
l
b
o
x
1
0
.
0
w
h
er
e
t
h
e
to
o
lb
o
x
is
v
er
y
i
m
p
o
r
tan
t
a
n
d
u
s
e
f
u
l
f
o
r
t
h
is
p
r
o
j
ec
t.
B
esid
es,
R
2
0
1
7
a
also
eq
u
ip
p
ed
w
i
t
h
I
m
a
g
e
P
r
o
ce
s
s
in
g
1
0
.
0
an
d
C
o
m
p
u
ter
Vis
io
n
T
o
o
lb
o
x
7
.
3
.
T
h
e
g
o
al
o
f
b
o
th
to
o
lb
o
x
es
is
to
s
i
m
p
li
f
y
all
p
r
o
ce
s
s
es
f
o
r
p
r
ep
r
o
ce
s
s
in
g
,
f
ea
t
u
r
e
ex
tr
ac
t
i
o
n
,
as
w
e
ll a
s
DNN.
4
.
2
DNN
T
esti
ng
P
er
f
o
r
m
a
nce
o
n Dig
it
s
a
nd
L
et
t
er
s
Cla
s
s
es
Usi
n
g
co
n
f
u
s
io
n
m
a
tr
ix
,
w
e
o
b
tain
ed
th
e
p
er
f
o
r
m
a
n
ce
m
etr
ics
o
f
h
a
n
d
w
r
iti
n
g
r
ec
o
g
n
i
tio
n
u
s
i
n
g
DNN
as s
h
o
w
n
in
T
ab
le
2
.
I
t
w
a
s
f
o
u
n
d
t
h
at
t
h
e
r
ec
o
g
n
itio
n
r
ate
f
o
r
d
ig
its
i
s
h
ig
h
er
th
a
n
l
etter
s
.
T
h
is
co
u
ld
b
e
d
u
e
to
s
m
aller
n
u
m
b
er
o
f
d
ig
i
ts
(
1
0
)
co
m
p
ar
ed
to
letter
s
(
5
2
)
.
T
h
e
w
ei
g
h
ted
a
v
er
ag
e
o
f
t
h
e
r
ec
o
g
n
i
tio
n
r
ate
w
il
l b
e
ar
o
u
n
d
9
3
.
4
%.
T
ab
le
2
.
P
er
f
o
r
m
a
n
ce
E
v
al
u
ati
o
n
o
f
th
e
T
r
ain
ed
DNN
o
n
T
esti
n
g
Data
s
et
C
l
a
ss
A
c
c
u
r
a
c
y
P
e
r
f
o
r
man
c
e
P
e
r
c
e
n
t
a
g
e
o
f
e
r
r
o
r
Ex
e
c
u
t
i
o
n
T
i
me
D
i
g
i
t
s
9
8
.
5
%
0
.
0
1
0
9
1
.
5
%
3
.
3
9
5
7
6
9
(
s)
L
e
t
t
e
r
s
8
8
.
8
%
0
.
0
3
1
3
1
1
.
2
%
7
.
7
4
0
5
3
4
(
s)
4
.
3
Co
m
pa
riso
n w
it
h o
t
her
ANN
s
t
ruct
ures
I
n
th
i
s
s
ec
t
io
n
,
w
e
p
r
esen
ted
th
e
p
er
f
o
r
m
a
n
ce
e
v
alu
a
tio
n
o
f
d
if
f
er
en
t
ANN
s
tr
u
ct
u
r
es,
in
cl
u
d
in
g
p
atter
n
et,
f
ee
d
f
o
r
w
ar
d
n
et,
an
d
p
r
o
p
o
s
ed
DNN.
T
h
e
p
atter
n
et
an
d
f
ee
d
f
o
r
w
ar
d
n
et
w
er
e
co
n
f
i
g
u
r
ed
w
it
h
1
h
id
d
en
la
y
er
an
d
1
0
n
u
m
b
er
o
f
n
eu
r
o
n
s
in
t
h
e
h
id
d
en
la
y
e
r
.
T
h
e
w
e
ig
h
t
av
er
a
g
e
o
f
th
e
r
ec
o
g
n
itio
n
r
ate
is
ca
lcu
lated
to
ta
k
e
in
to
ac
co
u
n
t
d
if
f
er
e
n
t
n
u
m
b
er
o
f
Di
g
its
(
1
0
)
an
d
L
e
tter
s
(
5
2
)
.
T
ab
le
3
s
h
o
w
s
t
h
e
p
er
f
o
r
m
a
n
ce
co
m
p
ar
is
o
n
o
f
t
h
r
ee
ANN
s
tr
u
ct
u
r
es,
in
w
h
i
ch
it
ca
n
b
e
s
ee
n
th
a
t
o
u
r
p
r
o
p
o
s
ed
alg
o
r
ith
m
p
er
f
o
r
m
s
b
etter
co
m
p
ar
ed
to
t
h
e
o
th
er
A
NN
s
tr
u
ct
u
r
e.
T
ab
le
3
.
R
ec
o
g
n
itio
n
R
ate
o
f
T
h
r
ee
A
NN
Stru
c
tu
r
es
A
N
N
st
r
u
c
t
u
r
e
C
l
a
ss
A
c
c
u
r
a
c
y
W
e
i
g
h
t
e
d
A
v
e
r
a
g
e
p
a
t
t
e
r
n
n
e
t
D
i
g
i
t
s
9
2
.
2
%
8
0
.
3
%
L
e
t
t
e
r
s
7
7
.
9
%
f
e
e
d
f
o
r
w
a
r
d
n
e
t
D
i
g
i
t
s
8
4
.
7
%
6
8
.
3
%
L
e
t
t
e
r
s
6
5
.
2
%
P
r
o
p
o
se
d
D
N
N
D
i
g
i
t
s
9
8
.
5
%
9
0
.
4
%
L
e
t
t
e
r
s
8
8
.
8
%
Evaluation Warning : The document was created with Spire.PDF for Python.
I
n
d
o
n
esia
n
J
E
lec
E
n
g
&
C
o
m
p
Sci
I
SS
N:
2502
-
4752
Dev
elo
p
men
t o
f E
n
g
lis
h
Ha
n
d
w
r
itten
R
ec
o
g
n
itio
n
Usi
n
g
Dee
p
.
.
.
(
Ted
d
y
s
u
r
ya
Gu
n
a
w
a
n
)
567
5.
CO
NCLU
SI
O
NS A
ND
F
UT
URE WO
RK
S
I
n
t
h
is
p
r
o
j
ec
t,
th
e
h
a
n
d
w
r
it
i
n
g
r
ec
o
g
n
itio
n
s
y
s
te
m
is
p
r
o
p
o
s
ed
b
y
u
s
in
g
i
m
a
g
e
p
i
x
els
as
f
ea
t
u
r
e
in
p
u
t
a
n
d
DNN
as
f
ea
t
u
r
e
ex
t
r
ac
tio
n
an
d
class
i
f
ier
.
T
h
e
s
eg
m
en
ted
i
m
a
g
e
s
ize
is
2
8
b
y
2
8
p
ix
els
p
r
o
d
u
cin
g
7
8
4
f
ea
tu
r
e
i
n
p
u
t.
T
h
e
DNN
s
tr
u
ct
u
r
e
i
s
s
tack
ed
a
u
to
e
n
co
d
er
s
w
ith
3
0
0
an
d
5
0
n
e
u
r
o
n
s
a
n
d
s
o
f
t
m
a
x
la
y
er
w
it
h
1
0
n
e
u
r
o
n
s
.
B
o
th
MN
I
ST
an
d
E
MN
I
ST
h
an
d
w
r
itte
n
i
m
ag
e
d
atab
ase
w
er
e
u
s
ed
in
th
e
p
er
f
o
r
m
an
ce
ev
alu
a
tio
n
.
T
h
e
w
ei
g
h
ted
r
ec
o
g
n
i
tio
n
r
ate
f
o
r
b
o
th
d
ig
it
s
an
d
letter
s
class
es
f
o
r
o
u
r
p
r
o
p
o
s
ed
DNN,
p
atter
n
et,
an
d
f
ee
d
f
o
r
w
ar
d
w
er
e
9
0
.
4
%,
8
0
.
3
%,
an
d
6
8
.
3
%.
Fu
tu
r
e
w
o
r
k
s
in
cl
u
d
e
d
if
f
er
en
t
f
ea
t
u
r
e
ex
tr
ac
tio
n
m
et
h
o
d
s
,
d
if
f
er
e
n
t d
ee
p
n
eu
r
al
n
et
w
o
r
k
co
n
f
i
g
u
r
atio
n
,
an
d
d
if
f
er
e
n
t d
atab
ase.
ACNO
WL
E
D
G
M
E
NT
T
h
e
au
th
o
r
s
w
o
u
ld
li
k
e
to
e
x
p
r
ess
th
eir
g
r
atit
u
d
e
to
t
h
e
M
ala
y
s
ia
n
Mi
n
i
s
tr
y
o
f
Hi
g
h
er
E
d
u
ca
tio
n
(
MO
HE
)
,
w
h
ic
h
h
a
s
p
r
o
v
id
ed
f
u
n
d
in
g
f
o
r
th
e
r
e
s
ea
r
ch
t
h
r
o
u
g
h
th
e
Fu
n
d
a
m
en
ta
l
R
e
s
e
ar
ch
Gr
an
t
Sch
e
m
e,
FR
GS1
5
-
194
-
0435.
RE
F
E
R
E
NC
E
S
[1
]
A
.
P
riy
a
,
S
.
M
ish
ra
,
S
.
Ra
j,
S
.
M
a
n
d
a
l,
S
.
Da
tt
a
,
"
On
li
n
e
a
n
d
o
ff
l
in
e
c
h
a
r
a
c
ter
re
c
o
g
n
it
i
o
n
:
A
su
rv
e
y
,
"
in
Co
m
m
u
n
ica
ti
o
n
a
n
d
S
ig
n
a
l
P
ro
c
e
ss
in
g
(ICCS
P
),
2
0
1
6
I
n
tern
a
ti
o
n
a
l
Co
n
f
e
re
n
c
e
o
n
,
p
p
.
0
9
6
7
-
0
9
7
0
,
2
0
1
6
.
[2
]
N.
S
h
a
rm
a
,
T
.
P
a
tn
a
ik
,
B.
Ku
m
a
r,
"
Re
c
o
g
n
it
io
n
f
o
r
Ha
n
d
w
rit
te
n
En
g
li
sh
L
e
tt
e
rs:
A
,
"
In
ter
n
a
ti
o
n
a
l
J
o
u
rn
a
l
o
f
En
g
i
n
e
e
rin
g
a
n
d
I
n
n
o
v
a
ti
v
e
T
e
c
h
n
o
l
o
g
y
(
IJ
EIT
)
,
v
o
l.
2
,
2
0
1
3
.
[3
]
U.
-
V
.
M
a
rti
,
H.
Bu
n
k
e
,
"
T
h
e
I
AM
-
d
a
tab
a
se
:
a
n
En
g
li
sh
se
n
ten
c
e
d
a
tab
a
se
f
o
r
o
ff
li
n
e
h
a
n
d
w
rit
in
g
re
c
o
g
n
it
io
n
,
"
In
ter
n
a
t
io
n
a
l
J
o
u
rn
a
l
o
n
Do
c
u
me
n
t
A
n
a
lys
is a
n
d
Rec
o
g
n
it
i
o
n
,
v
o
l.
5
,
p
p
.
3
9
-
4
6
,
2
0
0
2
.
[4
]
Y.
L
e
Cu
n
,
L
.
Bo
tt
o
u
,
Y.
Be
n
g
io
,
P
.
Ha
f
f
n
e
r,
"
Gr
a
d
ien
t
-
b
a
se
d
lea
rn
i
n
g
a
p
p
li
e
d
to
d
o
c
u
me
n
t
re
c
o
g
n
it
i
o
n
,
"
P
r
o
c
e
e
d
in
g
s o
f
th
e
IEE
E,
v
o
l.
8
6
,
p
p
.
2
2
7
8
-
2
3
2
4
,
1
9
9
8
.
[5
]
D.
Cireş
a
n
,
U.
M
e
ier,
"
M
u
lt
i
-
c
o
lu
mn
d
e
e
p
n
e
u
r
a
l
n
e
tw
o
rk
s
fo
r
o
ff
l
in
e
h
a
n
d
writt
e
n
C
h
in
e
se
c
h
a
ra
c
ter
c
la
ss
if
ica
ti
o
n
,
"
i
n
Ne
u
ra
l
Ne
tw
o
rk
s (IJCN
N),
2
0
1
5
In
ter
n
a
ti
o
n
a
l
Jo
in
t
C
o
n
f
e
re
n
c
e
o
n
,
p
p
.
1
-
6
,
2
0
1
5
.
[
6]
A
.
Yu
a
n
,
G
.
Ba
i,
L
.
Jia
o
,
Y.
Li
u
,
"
Offl
i
n
e
h
a
n
d
writ
ten
E
n
g
li
s
h
c
h
a
ra
c
ter
re
c
o
g
n
it
io
n
b
a
se
d
o
n
c
o
n
v
o
l
u
ti
o
n
a
l
n
e
u
ra
l
n
e
two
rk
,
"
in
D
o
c
u
m
e
n
t
An
a
ly
sis
S
y
ste
m
s
(D
A
S
),
2
0
1
2
1
0
t
h
IA
P
R
In
tern
a
ti
o
n
a
l
W
o
rk
sh
o
p
o
n
,
p
p
.
1
2
5
-
1
2
9
,
2
0
1
2
.
[7
]
Y.
LeCu
n
,
C.
Co
rtes
,
C.
J.
C.
Bu
rg
e
s,
"
T
h
e
M
NIST
Da
tab
a
s
e
o
f
Ha
n
d
w
rit
ten
Dig
it
s,"
Re
tri
e
v
e
d
o
n
:
1
0
De
c
e
m
b
e
r
2
0
1
7
.
[8
]
P
.
J.
G
ro
th
e
r,
K.
K.
Ha
n
a
o
k
a
,
"
NIST
S
p
e
c
ial
Da
tab
a
s
e
1
9
:
Ha
n
d
p
rin
ted
F
o
rm
s
a
n
d
Ch
a
ra
c
ters
Da
tab
a
se
,
"
2016.
[9
]
C.
Ka
e
n
sa
r,
"
A
c
o
mp
a
r
a
ti
v
e
stu
d
y
o
n
h
a
n
d
writi
n
g
d
i
g
it
re
c
o
g
n
it
i
o
n
c
la
ss
if
ier
u
sin
g
n
e
u
r
a
l
n
e
tw
o
rk
,
su
p
p
o
rt
v
e
c
to
r
ma
c
h
in
e
a
n
d
k
-
n
e
a
re
st
n
e
i
g
h
b
o
r,
"
i
n
T
h
e
9
th
In
tern
a
ti
o
n
a
l
Co
n
f
e
r
e
n
c
e
o
n
Co
m
p
u
ti
n
g
a
n
d
In
f
o
rm
a
t
io
n
T
e
c
h
n
o
l
o
g
y
(IC2
IT
2
0
1
3
),
p
p
.
1
5
5
-
1
6
3
,
2
0
1
3
.
[1
0
]
G
.
E.
Hin
to
n
,
S
.
Os
in
d
e
ro
,
Y.
-
W
.
T
e
h
,
"
A
fa
st
lea
rn
in
g
a
lg
o
rit
h
m
f
o
r
d
e
e
p
b
e
li
e
f
n
e
ts,"
N
e
u
ra
l
c
o
mp
u
ta
ti
o
n
,
v
o
l
.
1
8
,
p
p
.
1
5
2
7
-
1
5
5
4
,
2
0
0
6
.
[1
1
]
G
.
E.
Hin
to
n
,
R
.
R.
S
a
lak
h
u
td
in
o
v
,
"
Re
d
u
c
in
g
th
e
d
im
e
n
sio
n
a
li
ty
o
f
d
a
ta
w
it
h
n
e
u
ra
l
n
e
tw
o
rk
s,"
sc
ien
c
e
,
v
o
l.
3
1
3
,
p
p
.
5
0
4
-
5
0
7
,
2
0
0
6
.
[1
2
]
S
.
Na
re
jo
,
E.
P
a
se
ro
,
F
.
Ku
lso
o
m
,
"
EE
G
Ba
s
e
d
E
y
e
S
tate
Cla
ss
i
f
ica
ti
o
n
u
sin
g
De
e
p
Be
li
e
f
Ne
t
wo
rk
a
n
d
S
tac
k
e
d
A
u
to
En
c
o
d
e
r,
"
In
ter
n
a
ti
o
n
a
l
J
o
u
rn
a
l
o
f
El
e
c
trica
l
a
n
d
Co
mp
u
ter
En
g
i
n
e
e
rin
g
(
IJ
ECE
)
,
v
o
l.
6
,
p
p
.
3
1
3
1
,
2
0
1
6
.
Evaluation Warning : The document was created with Spire.PDF for Python.
I
SS
N
:
2
5
0
2
-
4752
I
n
d
o
n
esia
n
J
E
lec
E
n
g
&
C
o
m
p
Sci,
Vo
l.
10
,
No
.
2
,
Ma
y
2
0
1
8
:
5
6
2
–
5
6
8
568
B
I
O
G
RAP
H
I
E
S O
F
AUTH
O
RS
Te
d
d
y
S
u
r
y
a
G
u
n
a
w
a
n
re
c
e
iv
e
d
h
is
BEn
g
d
e
g
re
e
in
El
e
c
tri
c
a
l
En
g
in
e
e
rin
g
w
it
h
c
u
m
lau
d
e
a
wa
rd
f
ro
m
In
stit
u
t
T
e
k
n
o
lo
g
i
Ba
n
d
u
n
g
(IT
B),
In
d
o
n
e
sia
in
1
9
9
8
.
He
o
b
tain
e
d
h
is
M
.
En
g
d
e
g
re
e
in
2
0
0
1
f
ro
m
th
e
S
c
h
o
o
l
o
f
Co
m
p
u
ter
En
g
in
e
e
rin
g
a
t
Na
n
y
a
n
g
Tec
h
n
o
lo
g
ica
l
Un
iv
e
rsity
,
S
in
g
a
p
o
re
,
a
n
d
P
h
D
d
e
g
re
e
in
2
0
0
7
f
ro
m
th
e
S
c
h
o
o
l
o
f
El
e
c
tri
c
a
l
En
g
in
e
e
rin
g
a
n
d
T
e
le
c
o
m
m
u
n
ica
ti
o
n
s,
T
h
e
Un
iv
e
r
sity
o
f
Ne
w
S
o
u
th
W
a
les
,
A
u
stra
li
a
.
His
re
s
e
a
rc
h
in
te
re
sts
a
re
in
sp
e
e
c
h
a
n
d
a
u
d
io
p
ro
c
e
ss
in
g
,
b
io
m
e
d
ica
l
sig
n
a
l
p
ro
c
e
ss
in
g
a
n
d
i
n
stru
m
e
n
tatio
n
,
im
a
g
e
a
n
d
v
id
e
o
p
ro
c
e
ss
in
g
,
a
n
d
p
a
ra
ll
e
l
c
o
m
p
u
ti
n
g
.
He
is
c
u
rre
n
tl
y
a
n
IEE
E
S
e
n
io
r
M
e
m
b
e
r
(sin
c
e
2
0
1
2
)
,
w
a
s
c
h
a
irma
n
o
f
IEE
E
I
n
stru
m
e
n
tatio
n
a
n
d
M
e
a
su
re
m
e
n
t
S
o
c
iety
–
M
a
la
y
sia
S
e
c
ti
o
n
(2
0
1
3
a
n
d
2
0
1
4
)
,
A
s
so
c
iate
P
r
o
f
e
ss
o
r
(sin
c
e
2
0
1
2
)
,
He
a
d
o
f
De
p
a
rtm
e
n
t
(2
0
1
5
-
2
0
1
6
)
a
t
De
p
a
rtm
e
n
t
o
f
El
e
c
tri
c
a
l
a
n
d
Co
m
p
u
ter
En
g
in
e
e
rin
g
,
a
n
d
He
a
d
o
f
P
ro
g
ra
m
m
e
Ac
c
re
d
it
a
t
io
n
a
n
d
Qu
a
li
ty
As
su
ra
n
c
e
f
o
r
F
a
c
u
lt
y
o
f
En
g
in
e
e
rin
g
(sin
c
e
2
0
1
7
),
I
n
tern
a
ti
o
n
a
l
Isla
m
ic
Un
iv
e
r
sity
M
a
la
y
sia
.
He
is
Ch
a
rtere
d
En
g
in
e
e
r
(IE
T
,
UK
)
a
n
d
I
n
sin
y
u
r
P
r
o
f
e
sio
n
a
l
M
a
d
y
a
(P
II,
I
n
d
o
n
e
si
a
)
sin
c
e
2
0
1
6
.
Ahm
a
d
Fa
k
h
r
u
r
Ra
z
i
M
o
h
d
N
o
o
r
h
a
s
c
o
m
p
lete
d
h
is
B.
En
g
.
(Ho
n
s)
d
e
g
re
e
in
Co
m
m
u
n
ica
ti
o
n
En
g
in
e
e
rin
g
f
ro
m
In
tern
a
ti
o
n
a
l
Is
lam
i
c
Un
iv
e
rsit
y
M
a
la
y
sia
(IIUM
)
in
2
0
1
8
.
His res
e
a
rc
h
in
tere
sts
a
re
in
sig
n
a
l
p
ro
c
e
ss
in
g
a
n
d
a
r
ti
f
icia
l
in
telli
g
e
n
c
e
a
n
d
a
ff
e
c
ti
v
e
c
o
m
p
u
ti
n
g
.
Cu
rre
n
tl
y
,
h
e
is
w
o
rk
in
g
a
s e
n
g
in
e
e
r
a
t
In
tel,
P
e
n
a
n
g
.
M
ira
K
a
r
ti
w
i
c
o
m
p
lete
d
h
e
r
st
u
d
ies
a
t
th
e
Un
iv
e
rsit
y
o
f
W
o
ll
o
n
g
o
n
g
,
A
u
stra
li
a
re
su
lt
in
g
in
t
h
e
f
o
ll
o
w
in
g
d
e
g
re
e
s
b
e
in
g
c
o
n
f
e
rr
e
d
:
Ba
c
h
e
lo
r
o
f
Co
m
m
e
r
c
e
in
Bu
sin
e
ss
In
f
o
rm
a
ti
o
n
S
y
st
e
m
s,
M
a
ste
r
in
In
f
o
r
m
a
ti
o
n
S
y
ste
m
s
in
2
0
0
1
a
n
d
h
e
r
Do
c
to
r
o
f
P
h
i
lo
so
p
h
y
in
2
0
0
9
.
S
h
e
is
c
u
rre
n
tl
y
a
n
A
s
so
c
iate
P
r
o
f
e
ss
o
r
in
De
p
a
rt
m
e
n
t
o
f
In
f
o
rm
a
ti
o
n
S
y
ste
m
s,
Ku
li
y
y
a
h
o
f
In
f
o
rm
a
ti
o
n
a
n
d
Co
m
m
u
n
ica
ti
o
n
T
e
c
h
n
o
lo
g
y
,
In
tern
a
ti
o
n
a
l
Isla
m
ic
Un
iv
e
rsit
y
M
a
la
y
sia
.
He
r
r
e
se
a
r
c
h
in
tere
sts
in
c
lu
d
e
e
lec
tro
n
ic co
m
m
e
rc
e
,
d
a
ta m
in
in
g
,
e
-
h
e
a
lt
h
a
n
d
m
o
b
il
e
a
p
p
li
c
a
ti
o
n
s d
e
v
e
lo
p
m
e
n
t.
Evaluation Warning : The document was created with Spire.PDF for Python.