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.
3
6
,
No
.
3
,
Dec
em
b
er
20
2
4
,
p
p
.
1
5
6
4
~1
572
I
SS
N:
2
502
-
4
7
52
,
DOI
: 1
0
.
1
1
5
9
1
/ijee
cs
.v
3
6
.
i
3
.
p
p
1
5
6
4
-
1
5
7
2
1564
J
o
ur
na
l ho
m
ep
a
g
e
:
h
ttp
:
//ij
ee
cs
.
ia
esco
r
e.
co
m
Efficien
t
de
ep lea
rning
mo
dels for
Telug
u
ha
ndwri
tt
en t
e
x
t
recog
nition
B
ud
da
ra
j
u Re
v
a
t
hi
1
, B
.
N
.
V
.
Na
ra
s
im
ha
Ra
j
u
2
,
B
o
dd
u L
.
V
.
Siv
a
Ra
m
a
K
rish
na
3
Aj
a
y
D
ilip
K
um
a
r
M
a
ra
pa
t
l
a
3
,
S.
Su
ry
a
na
ra
y
a
na
ra
j
u
4
1
D
e
p
a
r
t
me
n
t
o
f
El
e
c
t
r
o
n
i
c
s a
n
d
C
o
m
mu
n
i
c
a
t
i
o
n
En
g
i
n
e
e
r
i
n
g
,
S
R
K
R
En
g
i
n
e
e
r
i
n
g
C
o
l
l
e
g
e
,
B
h
i
ma
v
a
r
a
m
,
I
n
d
i
a
2
D
e
p
a
r
t
me
n
t
o
f
C
o
m
p
u
t
e
r
S
c
i
e
n
c
e
a
n
d
E
n
g
i
n
e
e
r
i
n
g
,
S
R
K
R
E
n
g
i
n
e
e
r
i
n
g
C
o
l
l
e
g
e
,
B
h
i
ma
v
a
r
a
m
,
I
n
d
i
a
3
D
e
p
a
r
t
me
n
t
o
f
C
o
m
p
u
t
e
r
S
c
i
e
n
c
e
a
n
d
E
n
g
i
n
e
e
r
i
n
g
,
S
R
M
U
n
i
v
e
r
s
i
t
y
-
A
P
,
M
a
n
g
a
l
a
g
i
r
i
,
I
n
d
i
a
4
D
e
p
a
r
t
me
n
t
o
f
I
n
f
o
r
mat
i
o
n
Te
c
h
n
o
l
o
g
y
,
S
R
K
R
E
n
g
i
n
e
e
r
i
n
g
C
o
l
l
e
g
e
,
B
h
i
m
a
v
a
r
a
m
,
I
n
d
i
a
Art
icle
I
nfo
AB
S
T
RAC
T
A
r
ticle
his
to
r
y:
R
ec
eiv
ed
Ap
r
3
,
2
0
2
4
R
ev
is
ed
Sep
2
,
2
0
2
4
Acc
ep
ted
Sep
7
,
2
0
2
4
Op
ti
c
a
l
c
h
a
ra
c
ter
re
c
o
g
n
it
io
n
(OCR)
tec
h
n
o
lo
g
y
is
i
n
d
is
p
e
n
sa
b
le
fo
r
c
o
n
v
e
rti
n
g
a
n
d
a
n
a
l
y
z
in
g
tex
t
fr
o
m
v
a
ri
o
u
s
s
o
u
rc
e
s
i
n
to
a
f
o
rm
a
t
th
a
t
is
e
d
it
a
b
le
a
n
d
se
a
rc
h
a
b
le.
Telu
g
u
h
a
n
d
writi
n
g
p
re
se
n
ts
n
o
tab
le
c
h
a
ll
e
n
g
e
s
d
u
e
to
th
e
re
se
m
b
lan
c
e
o
f
c
h
a
ra
c
ters
,
th
e
e
x
ten
siv
e
c
h
a
ra
c
ter
se
t,
a
n
d
th
e
n
e
e
d
to
se
g
m
e
n
t
o
v
e
rlap
p
in
g
c
h
a
ra
c
ters
.
To
se
g
m
e
n
t
th
e
o
v
e
rlap
p
in
g
c
h
a
r
a
c
ters
,
we
a
ss
e
ss
th
e
wid
th
o
f
sm
a
ll
c
h
a
ra
c
ters
with
in
a
w
o
rd
a
n
d
se
g
m
e
n
t
th
e
o
v
e
rlap
p
in
g
c
h
a
ra
c
ters
a
c
c
o
rd
in
g
ly
.
Th
is
m
e
th
o
d
is
we
ll
su
i
ted
fo
r
th
e
se
g
m
e
n
tatio
n
o
f
o
v
e
rlap
p
i
n
g
c
o
m
p
o
u
n
d
c
h
a
ra
c
ters
.
To
a
d
d
re
ss
th
e
re
c
o
g
n
it
i
o
n
o
f
sim
il
a
r
c
h
a
ra
c
ters
with
les
s
trai
n
in
g
p
e
rio
d
s
we
h
a
v
e
u
se
d
Re
sN
e
t
1
8
a
n
d
S
q
u
e
e
z
e
Ne
t
m
o
d
e
ls
wh
ich
h
a
v
e
a
c
h
iev
e
d
c
h
a
ra
c
ter
re
c
o
g
n
it
i
o
n
ra
tes
o
f
9
5
%
a
n
d
9
4
%
re
sp
e
c
ti
v
e
ly
.
K
ey
w
o
r
d
s
:
C
an
n
y
d
etec
tio
n
C
h
ar
ac
ter
s
eg
m
en
tatio
n
C
o
n
v
o
lu
tio
n
al
n
eu
r
al
n
etwo
r
k
Featu
r
e
ex
tr
ac
tio
n
R
esNet
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
:
B
u
d
d
ar
aju
R
ev
ath
i
Dep
ar
tm
en
t o
f
E
lectr
o
n
ics an
d
C
o
m
m
u
n
icatio
n
E
n
g
in
ee
r
i
n
g
,
SR
K
R
E
n
g
in
ee
r
in
g
C
o
lleg
e
B
h
im
av
ar
am
,
5
3
4
2
0
4
,
I
n
d
ia
E
m
ail: b
u
d
d
a
r
aju
.
r
ev
ath
i@
g
m
ail.
co
m
1.
I
NT
RO
D
UCT
I
O
N
Op
tical
ch
ar
ac
ter
r
ec
o
g
n
itio
n
(
OC
R
)
is
a
tech
n
o
lo
g
y
en
g
in
ee
r
ed
to
c
o
n
v
e
r
t
v
ar
io
u
s
d
o
cu
m
e
n
t
f
o
r
m
ats
in
to
te
x
t
d
ata
t
h
at
i
s
ed
itab
le
an
d
s
ea
r
ch
ab
le
[
1
]
.
T
h
is
in
v
o
lv
es
an
aly
zi
n
g
t
h
e
tex
tu
al
im
a
g
es,
r
ec
o
g
n
izin
g
th
e
ch
a
r
ac
ter
s
,
a
n
d
c
o
n
v
e
r
tin
g
t
h
em
i
n
to
a
m
ac
h
in
e
-
r
ea
d
a
b
le
f
o
r
m
at.
OC
R
tech
n
o
l
o
g
y
p
lay
s
a
p
iv
o
tal
r
o
le
in
ex
p
e
d
itin
g
th
e
d
ig
italizatio
n
jo
u
r
n
e
y
b
y
tr
an
s
f
o
r
m
in
g
p
h
y
s
ical
d
o
cu
m
en
ts
in
to
f
o
r
m
ats
th
at
ar
e
s
ea
r
ch
ab
le,
ed
itab
le,
an
d
m
ac
h
in
e
-
r
ea
d
a
b
le.
T
h
is
r
ev
ea
ls
an
ex
ce
s
s
o
f
ad
v
an
tag
es
f
o
r
b
u
s
i
n
ess
es,
in
s
t
itu
tio
n
s
,
an
d
in
d
i
v
id
u
als alik
e.
OC
R
co
m
p
r
is
es
f
iv
e
p
r
im
ar
y
s
tag
es.
I
n
itially
,
im
ag
e
ac
q
u
is
itio
n
en
tails
ca
p
tu
r
in
g
th
e
te
x
t
-
c
o
n
tain
in
g
im
ag
e,
em
p
h
asizin
g
h
ig
h
-
q
u
ality
ac
q
u
is
itio
n
f
o
r
p
r
ec
is
e
OC
R
r
e
s
u
lts
[
2
]
.
Su
b
s
eq
u
en
tly
,
p
r
ep
r
o
ce
s
s
in
g
im
p
r
o
v
es
im
ag
e
q
u
ality
th
r
o
u
g
h
tech
n
iq
u
es
lik
e
n
o
is
e
r
ed
u
ctio
n
,
co
n
tr
ast
en
h
an
ce
m
e
n
t,
an
d
b
in
ar
izatio
n
[
3
]
.
T
h
e
th
ir
d
p
h
ase,
tex
t
d
etec
tio
n
an
d
s
eg
m
en
tatio
n
,
em
p
l
o
y
s
alg
o
r
ith
m
s
to
id
en
tify
tex
t
r
eg
io
n
s
in
th
e
p
r
e
-
p
r
o
ce
s
s
ed
im
ag
e
an
d
d
is
tin
g
u
is
h
th
em
f
r
o
m
n
o
n
-
tex
t
elem
e
n
ts
.
T
h
en
,
in
d
iv
id
u
al
ch
ar
ac
te
r
s
o
r
wo
r
d
s
with
in
d
etec
ted
tex
t
r
eg
io
n
s
ar
e
is
o
lated
to
s
tr
ea
m
lin
e
r
ec
o
g
n
itio
n
f
o
r
th
e
OC
R
s
y
s
tem
[
4
]
.
I
n
th
e
f
o
u
r
th
s
tag
e,
f
ea
tu
r
e
ex
tr
ac
tio
n
ca
p
tu
r
es
p
er
tin
en
t
tex
t
ch
ar
ac
ter
is
tics
s
u
ch
as
s
h
ap
e
an
d
s
ize
f
o
r
r
ec
o
g
n
itio
n
.
L
astl
y
,
ch
ar
ac
ter
r
ec
o
g
n
itio
n
m
atch
e
s
ex
tr
ac
ted
f
ea
tu
r
es
with
p
r
ed
ef
in
ed
m
o
d
els
o
r
te
m
p
late
s
,
u
tili
zin
g
p
atter
n
r
ec
o
g
n
itio
n
o
r
m
ac
h
in
e
lear
n
i
n
g
tech
n
i
q
u
es to
id
en
tif
y
ch
ar
ac
ter
s
.
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:
2
5
0
2
-
4
7
52
E
fficien
t d
ee
p
lea
r
n
in
g
m
o
d
els fo
r
telu
g
u
h
a
n
d
w
r
itten
text
r
ec
o
g
n
itio
n
(
B
u
d
d
a
r
a
ju
R
ev
a
th
i
)
1565
OC
R
f
o
r
n
ativ
e
lan
g
u
ag
es
r
eq
u
ir
es
h
eig
h
ten
ed
atten
tio
n
to
p
r
eser
v
e
h
is
to
r
ical
d
o
cu
m
e
n
ts
,
f
ac
ilit
ate
au
to
m
atio
n
,
an
d
s
u
p
p
o
r
t
ed
u
c
atio
n
al
en
d
ea
v
o
r
s
[
5
]
.
Desp
ite
b
ein
g
a
p
r
o
m
in
e
n
t
I
n
d
ian
lan
g
u
ag
e,
T
elu
g
u
h
as
r
ec
eiv
ed
less
em
p
h
asis
th
an
o
th
er
s
d
u
e
to
its
in
tr
icate
ch
a
r
ac
ter
s
,
s
eg
m
en
tatio
n
ch
allen
g
es,
an
d
th
e
lack
o
f
s
u
itab
le
d
atasets
.
Ov
er
co
m
in
g
s
eg
m
en
tatio
n
o
b
s
tacle
s
in
T
el
u
g
u
m
an
d
ates
in
n
o
v
ativ
e
s
tr
ateg
ies,
lik
e
is
o
latin
g
th
e
v
attu
s
,
to
im
p
r
o
v
e
OC
R
p
r
ec
is
io
n
an
d
u
tili
ty
.
Dev
elo
p
m
en
t
o
f
OC
R
f
o
r
n
at
iv
e
lan
g
u
a
g
es
s
h
o
u
ld
ac
co
u
n
t
f
o
r
th
e
lan
g
u
ag
e
s
p
ec
if
ic
f
e
atu
r
es
f
o
r
s
eg
m
en
tatio
n
an
d
r
ec
o
g
n
itio
n
.
T
o
o
v
e
r
co
m
e
s
eg
m
e
n
tatio
n
i
n
n
ativ
e
lan
g
u
ag
es
Nav
ee
n
a
a
n
d
Ar
ad
h
y
a
[
6
]
h
as
d
ev
elo
p
e
d
an
ad
v
a
n
ce
d
alg
o
r
ith
m
f
o
r
s
eg
m
en
tin
g
h
a
n
d
wr
itten
Kan
n
ad
a
s
cr
ip
ts
.
I
t
co
m
b
in
es
th
in
n
in
g
m
eth
o
d
s
,
b
r
a
n
ch
p
o
in
t
id
en
ti
f
icatio
n
,
an
d
m
i
x
tu
r
e
m
o
d
els,
u
tili
zin
g
th
e
E
M
alg
o
r
ith
m
to
lear
n
co
m
p
le
x
Gau
s
s
ian
m
ix
tu
r
es.
T
h
r
o
u
g
h
s
tr
ateg
ic
u
s
e
o
f
clu
s
ter
m
ea
n
p
o
in
ts
an
d
b
r
an
c
h
p
o
in
ts
,
Nav
e
en
a
ac
h
iev
es p
r
ec
is
e
ch
ar
ac
ter
s
eg
m
en
tatio
n
,
with
an
im
p
r
ess
iv
e
ac
cu
r
ac
y
o
f
8
5
.
5
%
d
em
o
n
s
tr
ated
in
ex
p
er
im
en
tal
tr
ials
.
B
ab
u
et
al
.
[
7
]
in
tr
o
d
u
ce
s
a
s
y
llab
le
s
eg
m
en
tatio
n
m
eth
o
d
f
o
r
p
r
i
n
ted
T
elu
g
u
tex
t.
I
t'
s
cr
u
cial
f
o
r
task
s
lik
e
wo
r
d
s
eg
m
e
n
tatio
n
,
s
p
ee
ch
s
y
n
th
esis
,
an
d
in
f
o
r
m
atio
n
r
etr
ie
v
al.
T
h
e
alg
o
r
ith
m
is
in
s
p
ir
ed
b
y
T
elu
g
u
s
cr
ip
t'
s
s
tr
u
ctu
r
e
b
u
t
f
ac
es
ch
allen
g
es
d
u
e
to
lan
g
u
ag
e
am
b
ig
u
ities
.
I
t
p
r
o
p
o
s
es
a
n
o
n
-
d
ictio
n
ar
y
ap
p
r
o
ac
h
b
ased
o
n
s
y
llab
le
wid
th
,
co
m
p
ar
in
g
it
with
th
e
asp
ec
t
r
atio
m
eth
o
d
.
E
x
p
er
im
en
tal
r
esu
lts
d
e
m
o
n
s
tr
ate
o
v
er
9
9
%
s
eg
m
en
tatio
n
ac
cu
r
ac
y
.
B
an
u
m
ath
i
an
d
C
h
an
d
r
a
[
8
]
in
t
r
o
d
u
ce
d
a
s
eg
m
en
tatio
n
s
y
s
tem
,
b
ased
o
n
th
e
Pro
jectio
n
Pro
f
ile
m
eth
o
d
,
h
a
n
d
les
g
en
er
al
a
n
d
h
is
to
r
ical
h
a
n
d
wr
itten
tex
t.
I
t
i
n
clu
d
es
p
r
e
-
p
r
o
ce
s
s
in
g
s
tep
s
to
en
h
an
ce
im
ag
e
q
u
ality
an
d
ac
h
iev
es
a
9
0
%
s
u
cc
ess
r
ate
in
l
in
e
s
eg
m
en
tatio
n
.
Ho
wev
er
,
wo
r
d
s
eg
m
en
tatio
n
,
esp
ec
ially
f
o
r
Kan
n
ad
a
s
cr
ip
t,
p
o
s
es
ch
allen
g
es
d
u
e
to
o
v
er
lap
p
in
g
s
u
b
s
cr
ip
ts
an
d
u
n
clea
r
s
p
ac
in
g
.
Desp
ite
th
ese
o
b
s
tacle
s
,
th
e
s
y
s
tem
r
ep
r
esen
ts
a
s
ig
n
if
ica
n
t
a
d
v
an
ce
m
e
n
t
in
h
an
d
wr
itten
tex
t
s
eg
m
en
tatio
n
,
h
ig
h
lig
h
tin
g
ar
ea
s
f
o
r
im
p
r
o
v
e
m
en
t in
Kan
n
a
d
a
wo
r
d
s
eg
m
e
n
tatio
n
.
T
o
d
ea
l
with
r
ec
o
g
n
itio
n
o
f
lar
g
e
n
u
m
b
er
o
f
s
im
ilar
ch
ar
ac
t
er
s
R
an
i
et
al
.
[
9
]
p
r
esen
ted
a
m
o
d
el
f
o
r
r
ec
o
g
n
izin
g
h
an
d
wr
itten
Kan
n
ad
a
ch
ar
a
cter
s
,
u
tili
zin
g
tr
a
n
s
f
er
lear
n
in
g
f
r
o
m
a
Dev
a
n
ag
ar
i
r
ec
o
g
n
itio
n
s
y
s
tem
.
T
h
e
m
o
d
el
em
p
lo
y
s
a
VGG1
9
NE
T
ar
ch
itectu
r
e,
co
m
p
r
is
in
g
f
iv
e
b
lo
c
k
s
with
co
n
v
o
lu
tio
n
an
d
m
ax
-
p
o
o
lin
g
lay
er
s
.
T
r
ain
in
g
in
v
o
l
v
es
1
2
3
,
6
5
4
d
ata
s
am
p
les,
wit
h
ex
p
er
im
en
tal
e
v
alu
atio
n
s
h
o
win
g
clo
s
e
to
9
0
%
ac
cu
r
ac
y
.
Valid
atio
n
o
v
er
1
0
ep
o
ch
s
ac
h
iev
es
7
3
.
5
1
%
ac
cu
r
ac
y
,
d
e
m
o
n
s
tr
atin
g
th
e
ef
f
ec
tiv
en
ess
o
f
tr
an
s
f
er
lear
n
in
g
f
o
r
Kan
n
ad
a
c
h
ar
ac
te
r
r
ec
o
g
n
itio
n
.
San
jeev
an
d
Sa
m
u
el
[
1
0
]
h
as
d
ev
elo
p
e
d
a
n
O
C
R
s
y
s
tem
tailo
r
ed
f
o
r
r
ec
o
g
n
izin
g
b
asic
ch
ar
ac
t
er
s
in
p
r
in
ted
Ka
n
n
ad
a
tex
t,
ac
co
m
m
o
d
atin
g
v
a
r
io
u
s
f
o
n
t
s
izes
an
d
ty
p
es.
B
y
u
tili
zin
g
Hu
'
s
in
v
ar
ian
t
m
o
m
en
ts
an
d
Z
e
r
n
ik
e
m
o
m
e
n
ts
f
o
r
f
ea
tu
r
e
ex
t
r
ac
tio
n
a
n
d
n
e
u
r
al
class
if
ier
s
f
o
r
ch
ar
ac
ter
class
if
icatio
n
,
th
e
s
y
s
tem
ac
h
iev
es
a
r
em
ar
k
ab
le
r
ec
o
g
n
itio
n
r
ate
o
f
9
6
.
8
%.
T
h
is
m
eth
o
d
o
l
o
g
y
s
h
o
ws
p
r
o
m
is
e
f
o
r
ex
ten
d
in
g
r
ec
o
g
n
itio
n
to
o
th
er
So
u
t
h
I
n
d
i
an
lan
g
u
ag
es,
p
ar
ticu
lar
ly
T
el
u
g
u
,
d
e
m
o
n
s
tr
atin
g
ef
f
icac
y
in
ac
cu
r
ately
r
ec
o
g
n
i
zin
g
p
r
in
ted
Kan
n
ad
a
ch
a
r
ac
ter
s
an
d
in
d
icatin
g
p
o
ten
tial
ap
p
licab
ilit
y
to
o
th
e
r
la
n
g
u
ag
es
in
th
e
r
eg
io
n
.
B
ash
a
et
al
.
[
1
1
]
i
n
tr
o
d
u
ce
d
a
g
r
o
u
n
d
b
r
ea
k
in
g
m
eth
o
d
th
at
h
ar
n
e
s
s
es
Dee
p
C
NNs
to
ac
cu
r
ately
id
e
n
tify
h
an
d
wr
itten
T
elu
g
u
alp
h
ab
ets
d
ep
icted
in
i
m
ag
es.
T
h
r
o
u
g
h
ex
ten
s
i
v
e
ex
p
er
im
en
tatio
n
,
B
ash
a’
s
m
o
d
el
ac
h
iev
es
r
em
a
r
k
ab
le
ad
v
an
ce
m
e
n
ts
,
ac
h
iev
i
n
g
an
ac
c
u
r
ac
y
r
ate
r
a
n
g
in
g
b
etwe
en
8
0
%
to
9
5
%
in
ac
cu
r
ately
r
ec
o
g
n
izin
g
in
d
i
v
id
u
al
alp
h
a
b
ets.
Seg
m
en
tatio
n
s
o
lu
tio
n
s
f
o
r
n
ativ
e
lan
g
u
ag
e
r
ec
o
g
n
itio
n
n
e
ed
to
ad
d
r
ess
th
e
s
p
ec
if
ic
tr
ai
ts
o
f
ea
ch
lan
g
u
ag
e.
Ho
wev
e
r
,
a
d
v
an
ce
m
en
ts
h
av
e
b
ee
n
lim
ited
d
u
e
to
t
h
e
d
if
f
icu
lties
in
s
eg
m
en
tin
g
o
v
er
lap
p
in
g
h
an
d
wr
itten
ch
ar
ac
ter
s
.
T
o
tac
k
le
th
is
is
s
u
e,
im
p
r
o
v
em
en
ts
h
av
e
b
ee
n
m
ad
e
to
th
e
v
er
tical
p
r
o
jectio
n
p
r
o
f
iles
alg
o
r
ith
m
,
r
esu
ltin
g
in
m
o
r
e
p
r
ec
is
e
s
eg
m
en
tatio
n
o
f
o
v
er
lap
p
i
n
g
ch
ar
ac
te
r
s
.
P
r
ev
io
u
s
ly
,
r
esear
ch
er
s
p
r
ed
o
m
i
n
an
tly
u
s
ed
m
ac
h
in
e
l
ea
r
n
in
g
o
r
d
ee
p
lear
n
in
g
m
o
d
els
f
o
r
c
h
ar
ac
ter
r
ec
o
g
n
itio
n
[
1
2
]
-
[
1
4
]
.
Ho
wev
e
r
,
m
ac
h
in
e
lear
n
in
g
m
o
d
els
o
f
t
en
s
tr
u
g
g
led
t
o
ac
h
iev
e
h
ig
h
ac
cu
r
ac
y
,
a
n
d
d
ee
p
lear
n
i
n
g
m
o
d
els
g
en
er
all
y
r
eq
u
ir
ed
m
an
y
p
ar
am
eter
s
,
lea
d
in
g
to
s
ig
n
i
f
ican
t c
o
m
p
u
tatio
n
al
ch
allen
g
es.
2.
O
CR
T
E
X
T
D
E
T
E
C
T
I
O
N
2
.
1
.
T
ex
t
det
ec
t
io
n
f
o
r
t
he
t
e
s
t
pa
g
e
At
f
ir
s
t,
th
e
OC
R
s
y
s
tem
tak
es
th
e
s
ca
n
n
e
d
p
a
g
e
with
th
e
tex
t
to
b
e
r
ec
o
g
n
ized
as
in
p
u
t.
Fo
llo
win
g
th
is
,
th
e
s
y
s
tem
co
n
v
er
ts
th
e
i
m
ag
e
to
g
r
ay
s
ca
le.
Af
ter
war
d
,
th
e
b
ilater
al
f
ilter
,
k
n
o
wn
as
a
non
-
lin
ea
r
f
ilter
,
s
m
o
o
th
s
an
im
ag
e
w
h
ile
p
r
ese
r
v
in
g
its
ed
g
es.
I
f
is
s
ca
n
n
ed
im
ag
e
an
d
a
is
th
e
p
ix
el
l
o
ca
tio
n
,
th
e
n
f
ilter
ed
o
u
tp
u
t
o
f
b
ilater
al
f
ilter
(
)
(
)
is
g
iv
en
b
y
(
1
)
.
(
)
(
)
=
1
(
)
∑
(
)
∙
(
‖
−
‖
)
∙
(
‖
(
)
−
(
)
‖
(
1
)
wh
er
e
R
is
th
e
s
p
atial
n
eig
h
b
o
r
h
o
o
d
o
f
a
a
n
d
is
th
e
s
p
atial
Gau
s
s
ian
k
er
n
el
with
s
tan
d
ar
d
d
ev
iatio
n
,
is
th
e
r
an
g
e
Gau
s
s
ian
k
er
n
el
with
s
tan
d
ar
d
d
e
v
iatio
n
.
(
)
is
th
e
n
o
r
m
aliza
tio
n
f
ac
to
r
.
Evaluation Warning : The document was created with Spire.PDF for Python.
I
SS
N
:
2
5
0
2
-
4
7
52
I
n
d
o
n
esian
J
E
lec
E
n
g
&
C
o
m
p
Sci
,
Vo
l.
3
6
,
No
.
3
,
Dec
em
b
er
20
2
4
:
1
5
6
4
-
1
5
7
2
1566
So
b
el
ed
g
e
d
etec
tio
n
aid
s
in
ex
tr
ac
tin
g
ed
g
es
f
r
o
m
an
im
a
g
e,
f
ac
ilit
atin
g
th
e
r
ec
o
g
n
itio
n
o
f
o
b
ject
b
o
u
n
d
ar
ies
s
u
ch
as
wo
r
d
s
[
1
5
]
.
Ap
p
l
y
in
g
a
t
h
r
esh
o
ld
to
th
e
g
r
a
d
ien
t
m
a
g
n
itu
d
e
im
ag
e
M
(
a
,
b
)
,
p
r
o
d
u
ce
s
a
b
in
ar
y
e
d
g
e
m
a
p
,
s
im
p
lify
in
g
d
etec
tio
n
g
iv
e
n
b
y
(
2
)
.
=
{
1
,
(
,
)
>
0
,
ℎ
(
2
)
wh
er
e
is
b
in
ar
ized
im
ag
e
an
d
T
is
th
e
th
r
esh
o
ld
lev
el.
Nex
t,
co
n
n
ec
ted
co
m
p
o
n
en
t a
n
aly
s
is
is
u
tili
ze
d
o
n
th
e
b
in
a
r
y
ed
g
e
m
ap
to
s
e
p
ar
ate
in
d
iv
id
u
al
wo
r
d
s
,
g
r
o
u
p
in
g
co
n
n
ec
ted
p
i
x
els
to
g
eth
er
.
On
ce
th
ese
co
m
p
o
n
en
ts
ar
e
o
b
tain
e
d
,
i
n
ter
s
ec
tio
n
an
d
u
n
io
n
o
p
er
a
tio
n
s
r
ef
in
e
wo
r
d
d
etec
tio
n
b
y
im
p
r
o
v
in
g
t
h
e
d
elin
ea
tio
n
o
f
wo
r
d
b
o
u
n
d
ar
ie
s
.
2
.
2
.
Seg
m
ent
a
t
i
o
n a
nd
f
ea
t
u
re
ex
t
ra
ct
i
o
n
C
h
ar
ac
ter
s
eg
m
en
tatio
n
is
i
n
d
is
p
en
s
ab
le
f
o
r
th
e
ac
c
u
r
ac
y
an
d
e
f
f
icac
y
o
f
OC
R
[
1
6
]
,
[
1
7
]
.
I
t
g
u
ar
an
tees
th
e
is
o
latio
n
o
f
i
n
d
iv
id
u
al
ch
ar
ac
te
r
s
,
ad
ep
tly
m
an
ag
es
h
an
d
wr
itten
te
x
t,
en
h
an
ce
s
r
ec
o
g
n
itio
n
ac
cu
r
ac
y
,
ac
c
o
m
m
o
d
ates
d
iv
e
r
s
e
lan
g
u
ag
es
an
d
s
cr
ip
ts
,
b
o
o
s
ts
p
r
o
ce
s
s
in
g
ef
f
icien
cy
,
a
n
d
aid
s
in
d
o
c
u
m
en
t
co
m
p
r
eh
e
n
s
io
n
.
Pr
o
p
er
s
eg
m
e
n
tatio
n
h
o
ld
s
a
p
i
v
o
tal
p
o
s
itio
n
in
ac
h
iev
in
g
elev
ated
r
ec
o
g
n
itio
n
r
ates
with
in
OC
R
s
y
s
tem
s
.
T
h
e
p
r
o
p
o
s
ed
ch
ar
ac
ter
s
eg
m
en
tatio
n
alg
o
r
i
th
m
s
tep
s
ar
e
g
iv
en
b
elo
w.
Fig
u
r
e
1
s
h
o
ws
th
e
ch
ar
ac
ter
s
eg
m
en
tatio
n
p
r
o
jec
tio
n
p
r
o
f
ile
g
r
a
p
h
s
an
d
th
e
s
eg
m
en
ts
.
Fig
u
r
e
1
(
a)
s
h
o
ws
th
e
p
r
o
jectio
n
p
r
o
f
ile
g
r
ap
h
f
o
r
a
n
in
p
u
t
wo
r
d
w
h
e
r
ea
s
Fig
u
r
e
s
1
(
b
)
a
n
d
1
(
c)
s
h
o
ws
th
e
s
eg
m
en
ted
p
ar
t
o
b
tai
n
ed
af
ter
a
p
p
ly
in
g
s
tan
d
ar
d
alg
o
r
ith
m
an
d
th
e
p
r
o
p
o
s
ed
o
n
e
.
Step
1
:
B
eg
in
with
a
n
in
p
u
t
i
m
ag
e
(
,
)
an
d
ca
lcu
late
th
e
co
lu
m
n
-
wis
e
s
u
m
o
f
p
ix
el
in
te
n
s
ities
[
1
8
]
g
iv
en
b
y
th
e
(
3
)
:
=
∑
=
1
(
3
)
Step
2
: D
ef
in
e
s
eg
m
en
ts
wh
er
e
th
e
s
am
e
v
alu
e
r
e
p
ea
ts
co
n
s
ec
u
tiv
ely
ac
r
o
s
s
co
lu
m
n
s
u
s
in
g
(
4
)
:
=
{
(
,
ℎ
)
|
=
(
(
+
1
)
}
(
4
)
wh
er
e
h
is
wid
th
o
f
th
e
p
a
r
t a
f
ter
s
eg
m
en
tatio
n
.
Step
3
: D
eter
m
in
e
th
e
m
in
im
u
m
s
eg
m
en
t w
id
th
f
r
o
m
t
h
e
id
e
n
tifie
d
s
eg
m
en
ts
g
iv
en
in
(
5
)
:
=
min
(
ℎ
)
(
5
)
S
t
e
p
4
:
C
o
m
p
a
r
e
t
h
e
m
i
n
i
m
u
m
s
e
g
m
e
n
t
wi
d
t
h
w
i
t
h
t
h
e
w
i
d
t
h
s
o
f
p
r
e
v
i
o
u
s
l
y
s
e
g
m
e
n
t
e
d
b
l
o
c
k
s
u
s
i
n
g
(
6
)
:
ℎ
>
2
×
(
6
)
Fo
r
s
eg
m
en
t
wid
th
th
at
ex
ce
ed
s
d
o
u
b
le
th
e
m
in
im
u
m
wid
t
h
,
s
p
lit
th
e
im
ag
e
at
th
e
co
lu
m
n
with
th
e
lo
west
p
ix
el
in
ten
s
ity
f
r
o
m
th
e
m
i
d
d
l
e,
co
n
s
id
er
in
g
1
0
co
lu
m
n
s
o
n
eith
er
s
id
e.
C
NNs
ar
e
p
r
ef
er
r
ed
f
o
r
e
x
tr
a
ctin
g
im
ag
e
f
ea
t
u
r
es
d
u
e
to
th
eir
ab
ilit
y
to
lear
n
h
ier
ar
c
h
ica
l
f
ea
tu
r
es,
d
is
p
lay
tr
an
s
latio
n
in
v
a
r
ian
ce
,
ef
f
icien
tly
s
h
ar
e
p
ar
am
eter
s
,
em
p
lo
y
l
o
ca
l
r
ec
ep
tiv
e
f
ield
s
,
in
teg
r
ate
p
o
o
lin
g
lay
er
s
,
m
ain
tain
r
o
b
u
s
tn
ess
to
v
ar
iatio
n
s
,
an
d
u
tili
ze
p
r
et
r
ain
ed
m
o
d
els
[
1
9
]
,
[
2
0
]
.
T
e
lu
g
u
lan
g
u
a
g
e
h
as
in
tr
icate
d
s
tr
u
ctu
r
es
with
m
an
y
n
u
m
b
er
s
o
f
ch
a
r
ac
ter
s
with
s
im
ilar
s
tr
u
ctu
r
e.
As
th
e
C
N
N
ar
ch
itectu
r
es
ar
e
ca
p
ab
le
to
ex
tr
a
ct
th
e
f
ea
tu
r
es,
we
h
av
e
u
s
ed
lig
h
t w
eig
h
ted
ar
ch
itectu
r
es to
r
ec
o
g
n
ize
T
el
u
g
u
ch
ar
ac
ter
s
.
T
h
e
co
n
v
o
l
u
tio
n
al
la
y
er
p
r
o
c
e
s
s
es
in
p
u
t
im
ag
es
b
y
co
n
v
o
lv
i
n
g
th
em
with
lear
n
ab
le
f
ilter
s
to
ex
tr
ac
t
f
ea
tu
r
es.
E
ac
h
f
ilter
s
ca
n
s
ac
r
o
s
s
th
e
in
p
u
t
im
ag
e,
p
r
o
d
u
cin
g
f
ea
tu
r
e
m
ap
s
th
r
o
u
g
h
c
o
n
v
o
lu
tio
n
o
p
e
r
atio
n
s
,
wh
ich
ar
e
th
en
p
ass
ed
th
r
o
u
g
h
an
ac
tiv
atio
n
f
u
n
ctio
n
.
Giv
en
an
i
n
p
u
t
im
a
g
e
X
with
d
im
en
s
io
n
s
h
×w
×c
,
wh
er
e
h
r
ep
r
esen
ts
th
e
h
eig
h
t,
w
d
en
o
tes
th
e
wid
th
,
an
d
c
s
t
an
d
s
f
o
r
th
e
n
u
m
b
er
o
f
c
h
an
n
els.
I
f
A
(
a
,
b
,
z
)
is
p
ix
el
in
ten
s
ity
o
f
i
n
p
u
t im
a
g
e
at
(
a
,
b
)
in
ch
a
n
n
el
z
a
n
d
(
,
)
is
th
e
f
ea
tu
r
e
m
a
p
at
(
a
,
b
)
g
iv
en
b
y
(
7
)
:
(
,
)
=
(
∑
∑
∑
(
+
,
+
,
)
.
(
,
,
)
+
)
=
1
=
1
=
1
(
7
)
wh
er
e
(
,
,
)
r
ep
r
esen
ts
th
e
weig
h
t
o
f
f
ilter
a
n
d
is
th
e
ac
tiv
atio
n
f
u
n
ctio
n
.
W
e
h
av
e
u
s
ed
R
ec
tifie
d
L
in
ea
r
u
n
it to
ac
tiv
ate
th
e
C
NN
m
o
d
el.
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:
2
5
0
2
-
4
7
52
E
fficien
t d
ee
p
lea
r
n
in
g
m
o
d
els fo
r
telu
g
u
h
a
n
d
w
r
itten
text
r
ec
o
g
n
itio
n
(
B
u
d
d
a
r
a
ju
R
ev
a
th
i
)
1567
(
a)
(
b
)
(
c)
Fig
u
r
e
1
.
C
h
ar
ac
ter
s
eg
m
en
tat
io
n
: (
a)
v
er
tical
p
r
o
jectio
n
p
r
o
f
ile
g
r
ap
h
f
o
r
a
wo
r
d
,
(
b
)
p
r
o
jectio
n
-
b
ased
s
eg
m
en
tatio
n
o
u
t
p
u
t
,
an
d
(
c)
o
u
tp
u
t o
f
p
r
o
p
o
s
ed
s
eg
m
e
n
tatio
n
M
ax
-
p
o
o
lin
g
la
y
er
r
ed
u
ce
s
th
e
s
p
atial
d
im
en
s
io
n
s
o
f
f
ea
tu
r
e
m
ap
s
b
y
s
elec
tin
g
th
e
m
ax
i
m
u
m
v
alu
e
with
in
s
p
ec
if
ied
r
eg
io
n
s
.
T
h
i
s
m
eth
o
d
ca
p
tu
r
es
im
p
o
r
ta
n
t
f
ea
tu
r
es
wh
ile
d
o
wn
s
am
p
lin
g
th
e
d
ata,
th
er
eb
y
d
ec
r
ea
s
in
g
its
r
eso
lu
tio
n
.
I
f
P
is
th
e
p
o
o
l
s
ize,
th
en
f
o
r
f
ea
tu
r
e
m
ap
(
,
)
,
th
e
m
ax
im
u
m
v
alu
e
in
th
e
p
o
o
lin
g
r
eg
io
n
is
g
iv
en
b
y
(
8
)
.
(
)
,
=
,
(
(
.
+
,
.
+
)
(
8
)
T
h
e
Ad
am
o
p
tim
izer
is
u
s
ed
t
o
ad
ju
s
t
th
e
n
etwo
r
k
'
s
weig
h
ts
b
ased
o
n
th
e
lo
s
s
f
u
n
ctio
n
g
r
ad
ien
t.
I
t
co
m
b
in
es
asp
ec
ts
o
f
th
e
Ad
aGr
ad
an
d
R
MSPr
o
p
o
p
tim
ize
r
s
to
ad
ap
tiv
ely
u
p
d
ate
lear
n
in
g
r
ates
f
o
r
ea
ch
p
ar
am
eter
d
u
r
in
g
tr
ain
in
g
.
L
ater
So
f
tMa
x
tr
an
s
f
o
r
m
s
th
e
o
u
tp
u
t
o
f
t
h
e
f
in
al
lay
e
r
in
to
a
p
r
o
b
ab
ilit
y
d
is
tr
ib
u
tio
n
ac
r
o
s
s
m
u
ltip
le
class
es.
I
t
ex
p
o
n
en
tiates
th
e
lo
g
its
o
f
ea
ch
class
an
d
th
e
n
n
o
r
m
alize
s
th
em
,
p
r
o
d
u
cin
g
p
r
o
b
ab
ilit
ies.
T
h
is
n
o
r
m
aliza
tio
n
en
ab
les
class
if
icatio
n
d
ec
is
io
n
s
b
ased
o
n
th
e
class
with
th
e
h
ig
h
est p
r
o
b
ab
ilit
y
.
Af
ter
ch
ar
ac
ter
s
eg
m
en
tatio
n
,
th
e
s
eg
m
en
ted
ch
ar
ac
ter
s
o
b
ta
in
ed
ar
e
ex
tr
ac
ted
f
o
r
f
ea
tu
r
es
.
A
C
NN
m
o
d
el
is
em
p
lo
y
ed
f
o
r
f
ea
tu
r
e
ex
tr
ac
tio
n
,
tr
ain
ed
u
s
in
g
th
e
I
E
E
E
d
atap
o
r
t
c
h
ar
ac
ter
d
ata
s
et.
W
e
h
av
e
u
s
ed
Sq
u
ee
ze
Net
an
d
R
esNet
1
8
m
o
d
els
f
o
r
f
ea
tu
r
e
ex
tr
ac
ti
o
n
.
S
u
b
s
eq
u
en
tly
,
th
e
f
ea
tu
r
es
e
x
tr
ac
ted
f
r
o
m
th
e
test
s
eg
m
en
ts
ar
e
c
o
m
p
a
r
ed
with
t
h
o
s
e
o
b
tain
ed
d
u
r
in
g
tr
ain
in
g
.
T
h
e
class
with
th
e
h
ig
h
est
m
atch
in
g
p
r
o
b
ab
ilit
y
is
th
en
ass
ig
n
ed
as th
e
o
u
tp
u
t.
Pre
v
io
u
s
r
esear
ch
r
ec
o
m
m
en
d
ed
em
p
lo
y
in
g
m
ac
h
in
e
lea
r
n
in
g
m
o
d
els
an
d
v
er
y
d
ee
p
lear
n
in
g
m
o
d
els
f
o
r
T
el
u
g
u
te
x
t
d
etec
t
io
n
.
Ho
wev
er
,
m
ac
h
in
e
lea
r
n
i
n
g
m
o
d
els
y
ield
ed
lo
wer
r
ec
o
g
n
itio
n
r
ates,
wh
ile
d
ee
p
lear
n
in
g
m
o
d
els
ac
h
iev
ed
h
ig
h
e
r
r
ec
o
g
n
itio
n
r
ates
at
th
e
co
s
t
o
f
in
c
r
ea
s
ed
tr
ain
a
b
le
p
ar
am
eter
s
an
d
lo
n
g
er
tr
ain
in
g
p
er
i
o
d
s
[
1
3
]
,
[
1
4
]
,
[
2
1
]
.
T
o
a
d
d
r
ess
th
ese
ch
allen
g
es,
we
ad
o
p
ted
R
esNet
-
1
8
an
d
Sq
u
ee
ze
Net
m
o
d
els
with
v
ar
iatio
n
s
in
f
ilter
s
izes.
T
h
ese
m
o
d
els
o
f
f
e
r
r
e
d
u
ce
d
p
ar
am
eter
c
o
n
s
u
m
p
tio
n
wh
ile
m
ain
tain
in
g
s
ig
n
if
ican
t
ac
cu
r
ac
y
r
ates.
T
h
e
s
elec
tio
n
b
etwe
en
th
ese
m
o
d
els
d
ep
en
d
s
o
n
s
p
ec
if
ic
r
eq
u
ir
em
e
n
ts
an
d
ap
p
licatio
n
co
n
tex
ts
.
T
h
e
m
o
d
el
o
f
R
esNet
1
8
is
s
h
o
wn
in
Fig
u
r
e
2
an
d
Sq
u
ee
ze
Net
is
s
h
o
wn
in
Fig
u
r
e
3
.
R
esNet
-
1
8
h
o
ld
s
s
ig
n
if
ican
ce
f
o
r
its
ab
ilit
y
to
f
in
d
eq
u
ilib
r
iu
m
b
etwe
en
d
e
p
th
,
co
m
p
l
ex
ity
,
an
d
p
er
f
o
r
m
an
ce
,
r
e
n
d
er
in
g
it
a
f
lex
ib
le
an
d
ef
f
icien
t
s
o
lu
tio
n
f
o
r
d
iv
er
s
e
co
m
p
u
ter
v
is
io
n
ap
p
licatio
n
s
[
2
2
]
.
I
ts
p
io
n
ee
r
in
g
u
s
e
o
f
r
esid
u
al
co
n
n
ec
tio
n
s
h
as
r
esh
ap
e
d
th
e
la
n
d
s
ca
p
e
o
f
d
ee
p
lear
n
in
g
a
r
ch
itectu
r
es,
lay
in
g
th
e
f
o
u
n
d
atio
n
f
o
r
t
h
e
em
er
g
en
ce
o
f
in
c
r
ea
s
in
g
ly
s
o
p
h
is
ticated
m
o
d
els.
W
h
ile
Sq
u
ee
ze
Net
s
tan
d
s
o
u
t
as
a
d
ee
p
lear
n
in
g
m
o
d
el
tailo
r
ed
f
o
r
h
eig
h
ten
ed
ac
cu
r
ac
y
wh
ile
co
n
s
er
v
in
g
co
m
p
u
tatio
n
al
r
eso
u
r
ce
s
[
2
3
]
.
I
ts
d
is
tin
g
u
is
h
in
g
f
ea
tu
r
e
is
th
e
in
co
r
p
o
r
atio
n
o
f
1
x
1
c
o
n
v
o
lu
tio
n
al
lay
er
s
,
k
n
o
wn
as
n
etwo
r
k
-
in
-
n
etwo
r
k
,
wh
ich
ef
f
icien
tly
r
ed
u
ce
s
p
ar
am
eter
co
u
n
t
wh
ile
p
r
eser
v
in
g
r
ep
r
esen
tatio
n
al
s
tr
en
g
th
.
T
h
is
in
n
o
v
ativ
e
ap
p
r
o
ac
h
em
p
o
wer
s
Sq
u
ee
ze
Net
to
m
ain
tain
h
ig
h
ac
cu
r
ac
y
lev
els
d
esp
ite
its
n
o
tab
ly
s
m
aller
m
o
d
el
s
ize
wh
en
co
m
p
ar
ed
with
estab
lis
h
ed
d
ee
p
ar
ch
itectu
r
es
lik
e
VGG.
B
o
th
R
esNet1
8
an
d
Sq
u
ee
ze
Net
ar
e
u
n
iq
u
e
i
n
ar
ch
itectu
r
e
b
y
p
r
o
v
id
in
g
o
p
t
im
al
p
er
f
o
r
m
an
ce
in
im
a
g
e
r
ec
o
g
n
itio
n
task
s
b
y
u
tili
zin
g
f
ewe
r
p
ar
am
eter
s
co
m
p
ar
ed
to
VGG.
Evaluation Warning : The document was created with Spire.PDF for Python.
I
SS
N
:
2
5
0
2
-
4
7
52
I
n
d
o
n
esian
J
E
lec
E
n
g
&
C
o
m
p
Sci
,
Vo
l.
3
6
,
No
.
3
,
Dec
em
b
er
20
2
4
:
1
5
6
4
-
1
5
7
2
1568
B
o
th
m
o
d
els
wer
e
tr
ain
ed
u
s
in
g
th
e
I
E
E
E
Data
p
o
r
t
d
ataset
an
d
ass
e
s
s
ed
f
o
r
th
eir
ch
ar
ac
ter
r
ec
o
g
n
itio
n
ca
p
a
b
ilit
ies.
Su
b
s
eq
u
en
tly
,
a
test
p
ag
e
co
n
tain
in
g
h
an
d
wr
itten
T
elu
g
u
wo
r
d
s
was
in
p
u
tted
in
to
th
e
OC
R
s
y
s
tem
.
T
h
e
OC
R
id
en
ti
f
ied
th
e
wo
r
d
s
an
d
s
eg
m
e
n
ted
th
em
in
to
ch
ar
ac
ter
s
,
wh
ich
wer
e
th
en
p
ass
ed
to
th
e
m
o
d
els
f
o
r
r
ec
o
g
n
itio
n
.
B
ased
o
n
th
e
o
u
tp
u
t,
th
e
s
y
s
tem
'
s
wo
r
d
r
ec
o
g
n
itio
n
r
ates
w
er
e
ev
alu
ated
.
T
h
e
m
o
d
els we
r
e
test
ed
with
1
0
0
0
wo
r
d
s
,
an
d
t
h
eir
p
er
f
o
r
m
a
n
ce
was a
s
s
e
s
s
ed
ac
co
r
d
in
g
ly
.
Fig
u
r
e
2
.
Mo
d
el
o
f
R
esNet
1
8
Fig
u
r
e
3
.
Mo
d
el
o
f
Sq
u
ee
ze
N
et
3.
RE
SU
L
T
S
E
ac
h
im
ag
e
was
co
n
v
e
r
ted
to
g
r
ay
s
ca
le
an
d
r
esized
to
d
im
e
n
s
io
n
s
o
f
6
4
x
6
4
p
ix
els.
T
h
e
d
ataset
wa
s
o
b
tain
ed
f
r
o
m
I
E
E
E
Data
p
o
r
t [
2
4
]
,
a
n
d
tr
ain
i
n
g
in
v
o
lv
ed
b
o
th
Sh
u
f
f
leNe
t
an
d
R
esNet
1
8
m
o
d
els
o
v
er
2
5
,
0
0
0
s
tep
s
.
T
r
ain
in
g
d
ata
c
o
m
p
r
is
ed
1
1
,
6
0
2
-
ch
ar
ac
ter
im
ag
es,
wi
th
an
ad
d
itio
n
al
2
,
5
6
5
-
ch
ar
ac
t
er
im
ag
es
r
eser
v
ed
f
o
r
v
alid
atio
n
p
u
r
p
o
s
es.
Fig
u
r
e
4
s
h
o
ws
th
e
p
lo
ts
o
f
R
esNet
1
8
.
Fig
u
r
e
4
(
a)
d
is
p
lay
s
th
e
c
h
ar
ac
ter
r
ec
o
g
n
itio
n
tr
ain
an
d
v
alid
atio
n
ac
c
u
r
ac
y
p
lo
ts
f
o
r
R
esNet1
8
.
Me
an
wh
ile,
Fig
u
r
e
4
(
b
)
illu
s
tr
ates
th
e
v
alid
atio
n
lo
s
s
p
lo
t
o
f
R
esNet
1
8
.
Fig
u
r
e
5
s
h
o
ws
th
e
p
lo
ts
o
f
Sq
u
u
ez
eNe
t.
Fig
u
r
e
5
(
a)
d
ep
icts
th
e
ch
ar
ac
ter
r
ec
o
g
n
itio
n
tr
ain
in
g
an
d
v
alid
atio
n
ac
cu
r
ac
y
p
lo
t
s
f
o
r
Sq
u
ee
ze
Net,
wh
ile
Fig
u
r
e
5
(
b
)
s
h
o
wca
s
es
th
e
v
ali
d
atio
n
lo
s
s
p
lo
t
o
f
Sq
u
ee
ze
Net.
R
esNet1
8
h
as
r
ea
ch
ed
a
c
h
ar
a
cter
ac
cu
r
ac
y
o
f
9
5
%
an
d
a
lo
s
s
o
f
6
%
b
ased
o
n
th
e
p
l
o
ts
.
Me
an
wh
ile,
Sq
u
ee
ze
Net
ac
h
iev
es
a
ch
ar
ac
ter
r
ec
o
g
n
itio
n
r
ate
o
f
9
4
%
with
a
lo
s
s
o
f
8
%,
as
in
d
icate
d
b
y
th
e
p
lo
t
d
ata.
B
o
th
R
esNet1
8
an
d
Sq
u
ee
ze
Net
d
em
o
n
s
tr
ate
n
o
tewo
r
t
h
y
p
er
f
o
r
m
an
ce
in
r
ec
o
g
n
izin
g
Han
d
wr
itten
T
elu
g
u
ch
ar
ac
ter
s
,
s
h
o
wca
s
in
g
s
u
b
s
ta
n
tial c
h
ar
ac
ter
r
ec
o
g
n
itio
n
r
ate
s
.
T
ab
le
1
p
r
esen
ts
a
c
o
m
p
a
r
is
o
n
b
etwe
en
Sq
u
ee
ze
Net
an
d
R
esNet1
8
m
o
d
els.
R
esNet1
8
,
r
en
o
wn
ed
f
o
r
its
ar
ch
itectu
r
e,
r
eq
u
i
r
es
1
,
3
6
8
,
6
4
0
p
a
r
am
eter
s
,
wh
er
ea
s
Sq
u
ee
ze
Net
d
em
an
d
s
o
n
ly
8
6
1
,
3
4
4
p
a
r
am
eter
s
.
Desp
ite
h
av
in
g
f
ewe
r
tr
ain
a
b
le
p
ar
am
eter
s
,
Sq
u
ee
ze
Net
a
ch
iev
es
ac
cu
r
ac
y
r
ates
clo
s
er
to
R
esNet1
8
.
T
h
e
v
alid
atio
n
,
test
,
an
d
wo
r
d
ac
c
u
r
ac
y
r
ates
f
o
r
R
esNet1
8
ar
e
9
5
%,
9
0
.
5
8
%,
an
d
8
4
%,
r
esp
ec
tiv
ely
.
On
th
e
o
th
er
h
an
d
,
Sq
u
ee
ze
Net
ac
h
iev
es
a
v
alid
atio
n
ac
c
u
r
ac
y
o
f
9
4
%,
a
test
ac
cu
r
ac
y
o
f
8
9
.
4
3
8
%,
a
n
d
a
wo
r
d
ac
cu
r
ac
y
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:
2
5
0
2
-
4
7
52
E
fficien
t d
ee
p
lea
r
n
in
g
m
o
d
els fo
r
telu
g
u
h
a
n
d
w
r
itten
text
r
ec
o
g
n
itio
n
(
B
u
d
d
a
r
a
ju
R
ev
a
th
i
)
1569
o
f
8
3
.
5
%.
B
o
th
m
o
d
els
d
em
o
n
s
tr
ate
s
tr
o
n
g
p
er
f
o
r
m
an
ce
o
n
a
T
elu
g
u
ch
ar
ac
te
r
d
ataset.
T
h
e
ch
o
ice
b
etwe
e
n
th
em
d
ep
e
n
d
s
o
n
t
h
e
s
p
ec
if
ic
p
r
o
b
lem
r
eq
u
ir
e
m
en
ts
an
d
d
esire
d
ac
cu
r
ac
y
r
ates.
(
a)
(
b
)
Fig
u
r
e
4
.
Plo
ts
o
f
R
esNet1
8
(
a
)
ac
cu
r
ac
y
p
lo
t a
n
d
(
b
)
lo
s
s
p
l
o
t
(
a)
(
b
)
Fig
u
r
e
5
.
Plo
ts
o
f
Sq
u
ee
ze
Net
(
a)
ac
cu
r
ac
y
p
lo
t a
n
d
(
b
)
lo
s
s
p
lo
t
T
o
co
m
p
u
te
th
e
wo
r
d
r
ec
o
g
n
i
tio
n
r
ates,
we
u
tili
ze
d
5
0
0
test
p
ag
es
au
th
o
r
e
d
b
y
v
ar
io
u
s
in
d
iv
id
u
als,
co
n
tain
in
g
n
ea
r
ly
1
0
0
0
T
el
u
g
u
wo
r
d
s
.
Af
ter
d
etec
tin
g
p
ag
es
an
d
wo
r
d
s
,
th
ese
wo
r
d
s
u
n
d
e
r
wen
t
s
eg
m
en
tatio
n
,
f
o
llo
wed
b
y
ch
ar
ac
ter
r
ec
o
g
n
itio
n
u
s
in
g
th
e
Sq
u
ee
ze
Net
an
d
R
esNet1
8
m
o
d
els.
T
h
e
ac
h
iev
ed
wo
r
d
r
ec
o
g
n
itio
n
r
ates a
r
e
d
et
ailed
in
T
ab
le
1
.
C
o
m
p
ar
is
o
n
o
f
Sq
u
ee
ze
Net
a
n
d
R
esNet1
8
is
s
h
o
w
n
in
Fig
u
r
e
6
.
Fig
u
r
e
6
(
a)
g
i
v
es
th
e
lo
s
s
p
lo
t
co
m
p
ar
is
o
n
o
f
R
esNet1
8
an
d
Sq
u
ee
ze
Net.
Fig
u
r
e
6
(
b
)
s
h
o
ws
th
e
v
alid
atio
n
,
test
,
an
d
wo
r
d
ac
cu
r
ac
y
r
ate
co
m
p
ar
is
o
n
o
f
b
o
th
th
e
m
o
d
el
s
.
R
esNet1
8
h
as
s
lig
h
tly
en
h
an
ce
d
p
er
f
o
r
m
a
n
ce
th
an
Sq
u
ee
ze
Net.
B
u
t
in
ter
m
s
o
f
tr
ain
in
g
tim
e
Sq
u
ee
ze
Net
r
e
q
u
ir
es less
tr
ain
in
g
tim
e
an
d
t
r
ain
ab
le
p
ar
am
eter
s
c
o
m
p
ar
e
d
t
o
R
esNet1
8
.
T
ab
le
2
d
is
p
lay
s
a
p
er
f
o
r
m
a
n
c
e
co
m
p
a
r
is
o
n
b
etwe
en
o
u
r
m
o
d
el
an
d
ex
is
tin
g
m
o
d
els
[
1
3
]
,
[
1
4
]
,
[
2
5
]
.
Dev
elo
p
in
g
a
n
OC
R
f
o
r
T
e
lu
g
u
,
a
c
o
m
p
lex
lan
g
u
a
g
e
with
n
u
m
er
o
u
s
co
m
p
o
u
n
d
c
h
ar
ac
ter
s
,
p
o
s
es
a
s
ig
n
if
ican
t
ch
allen
g
e,
lead
i
n
g
to
lim
ited
r
esear
ch
in
th
is
ar
ea
.
W
h
ile
th
e
VGG
m
o
d
el
ac
h
iev
ed
n
o
tab
le
r
ec
o
g
n
itio
n
r
ates,
its
d
ee
p
a
r
c
h
itectu
r
e
r
eq
u
ir
es
a
s
u
b
s
tan
tia
l
n
u
m
b
e
r
o
f
tr
ain
ab
le
p
ar
am
et
er
s
.
I
n
co
n
tr
ast,
o
u
r
lig
h
tweig
h
t
m
o
d
els,
with
f
ew
er
p
ar
am
eter
s
,
ex
h
ib
it
h
ig
h
r
ec
o
g
n
itio
n
r
ates
an
d
p
er
f
o
r
m
ef
f
icien
tly
with
a
r
ed
u
ce
d
p
ar
am
eter
c
o
u
n
t.
T
ab
le
1
.
C
o
m
p
a
r
is
o
n
o
f
R
esN
et1
8
an
d
Sq
u
ee
ze
Net
M
o
d
e
l
P
a
r
a
me
t
e
r
s
V
a
l
i
d
a
t
i
o
n
c
h
a
r
a
c
t
e
r
a
c
c
u
r
a
c
y
Te
st
c
h
a
r
a
c
t
e
r
a
c
c
u
r
a
c
y
W
o
r
d
a
c
c
u
r
a
c
y
R
e
sN
e
t
1
8
1
3
,
6
8
,
6
4
0
9
5
%
9
0
.
5
8
0
8
4
%
S
q
u
e
e
z
e
N
e
t
8
,
6
1
,
3
4
4
9
4
%
8
9
.
4
3
8
8
3
%
Evaluation Warning : The document was created with Spire.PDF for Python.
I
SS
N
:
2
5
0
2
-
4
7
52
I
n
d
o
n
esian
J
E
lec
E
n
g
&
C
o
m
p
Sci
,
Vo
l.
3
6
,
No
.
3
,
Dec
em
b
er
20
2
4
:
1
5
6
4
-
1
5
7
2
1570
(
a)
(
b
)
Fig
u
r
e
6
.
C
o
m
p
a
r
is
o
n
o
f
R
esNet1
8
an
d
Sq
u
ee
ze
Net
(
a)
l
o
s
s
r
ate
(
b
)
ac
cu
r
ac
y
T
ab
le
2
.
C
o
m
p
a
r
is
o
n
o
f
R
esN
et1
8
an
d
Sq
u
ee
ze
Net
N
a
me
o
f
t
h
e
a
u
t
h
o
r
Ty
p
e
o
f
t
e
x
t
Te
c
h
n
i
q
u
e
A
c
c
u
r
a
c
y
P
a
n
y
a
m
N
a
r
a
h
a
r
i
S
a
s
t
r
y
H
a
n
d
w
r
i
t
t
e
n
T
e
l
u
g
u
c
h
a
r
a
c
t
e
r
s
N
e
a
r
e
st
n
e
i
g
h
b
o
u
r
c
l
a
ss
i
f
i
e
r
7
8
%
S
a
r
i
k
a
.
N
H
a
n
d
w
r
i
t
t
e
n
Te
l
u
g
u
C
h
a
r
a
c
t
e
r
s
VGG
-
16
9
2
%
M
i
n
e
s
h
M
a
t
h
e
w
S
c
e
n
e
T
e
l
u
g
u
Te
x
t
d
e
t
e
c
t
i
o
n
H
y
b
r
i
d
C
N
N
-
R
N
N
(
c
h
a
r
a
c
t
e
r
-
l
e
v
e
l
)
8
6
.
2
%
H
y
b
r
i
d
C
N
N
-
R
N
N
(
w
o
r
d
-
l
e
v
e
l
)
5
7
.
2
%
P
r
o
p
o
se
d
H
a
n
d
w
r
i
t
t
e
n
Te
l
u
g
u
Te
x
t
R
e
sN
e
t
1
8
(
C
h
a
r
a
c
t
e
r
-
l
e
v
e
l
)
(
w
o
r
d
-
l
e
v
e
l
)
9
5
%
8
4
%
S
q
u
e
e
z
e
N
e
t
(
c
h
a
r
a
c
t
e
r
-
l
e
v
e
l
)
(
w
o
r
d
-
l
e
v
e
l
)
9
4
%
8
3
%
4.
CO
NCLU
SI
O
N
I
d
en
tify
in
g
h
an
d
wr
itten
T
elu
g
u
tex
t
p
o
s
es
s
ig
n
if
ican
t
ch
alle
n
g
es
d
u
e
to
th
e
v
ar
iety
o
f
wr
it
in
g
s
ty
les,
o
v
er
lap
p
i
n
g
c
h
ar
ac
ter
s
,
a
n
d
t
h
e
n
ee
d
to
d
if
f
e
r
en
tiate
b
etwe
en
s
im
ilar
s
tr
u
ctu
r
es
with
in
a
lar
g
e
ch
a
r
ac
ter
s
et.
T
h
e
s
eg
m
en
tatio
n
alg
o
r
ith
m
p
r
o
p
o
s
ed
in
th
is
s
tu
d
y
h
as
d
em
o
n
s
tr
ated
n
o
tab
le
s
u
cc
ess
in
ac
cu
r
ately
s
ep
ar
atin
g
o
v
er
lap
p
i
n
g
ch
ar
ac
ter
s
,
th
er
e
b
y
en
h
an
cin
g
wo
r
d
r
ec
o
g
n
itio
n
r
ates.
Am
o
n
g
th
e
m
o
d
els
ev
alu
ated
,
R
esNet1
8
an
d
Sq
u
ee
ze
Net
h
av
e
e
m
er
g
e
d
as
p
ar
ticu
lar
ly
ef
f
ec
tiv
e
f
o
r
th
is
task
.
R
es
Net1
8
,
with
its
d
ee
p
ar
ch
itectu
r
e
,
ac
h
iev
es
a
ch
ar
ac
ter
r
ec
o
g
n
iti
o
n
r
ate
(
C
R
R
)
o
f
9
5
%
an
d
a
wo
r
d
r
ec
o
g
n
itio
n
r
ate
(
W
R
R
)
o
f
8
4
%.
I
n
co
n
tr
ast,
Sq
u
ee
ze
Net,
d
esp
ite
h
av
in
g
f
ewe
r
p
ar
am
eter
s
,
ac
h
iev
es
a
C
R
R
o
f
9
4
%
an
d
a
W
R
R
o
f
8
3
%.
B
o
th
m
o
d
els
ca
p
tu
r
e
th
e
in
t
r
icate
f
ea
tu
r
es
o
f
T
elu
g
u
tex
t
ef
f
ec
tiv
ely
.
M
o
r
eo
v
e
r
,
th
eir
s
h
o
r
ter
tr
ain
in
g
tim
es
co
m
p
ar
ed
t
o
d
ee
p
er
m
o
d
els lik
e
VGG
m
ak
e
th
em
h
ig
h
ly
s
u
itab
le
f
o
r
th
e
d
ig
italizatio
n
o
f
T
elu
g
u
tex
t.
REFEREN
CES
[
1
]
J.
M
e
mo
n
,
M
.
S
a
mi
,
R
.
A
.
K
h
a
n
,
a
n
d
M
.
U
d
d
i
n
,
“
H
a
n
d
w
r
i
t
t
e
n
o
p
t
i
c
a
l
c
h
a
r
a
c
t
e
r
r
e
c
o
g
n
i
t
i
o
n
(
O
C
R
)
a
c
o
m
p
r
e
h
e
n
s
i
v
e
sy
s
t
e
m
a
t
i
c
l
i
t
e
r
a
t
u
r
e
r
e
v
i
e
w
(
S
LR
)
,
”
i
n
I
EEE
Ac
c
e
ss
,
v
o
l
.
8
,
p
p
.
1
4
2
6
4
2
-
1
4
2
6
6
8
,
2
0
2
0
,
d
o
i
:
1
0
.
1
1
0
9
/
A
C
C
ESS
.
[
2
]
M
.
K
u
mar,
S
.
R
.
Ji
n
d
a
l
,
M
.
K
.
J
i
n
d
a
l
,
a
n
d
G
.
S
.
Le
h
a
l
,
“
I
mp
r
o
v
e
d
r
e
c
o
g
n
i
t
i
o
n
r
e
s
u
l
t
s
o
f
me
d
i
e
v
a
l
h
a
n
d
w
r
i
t
t
e
n
G
u
r
mu
k
h
i
man
u
scr
i
p
t
s
u
s
i
n
g
b
o
o
st
i
n
g
a
n
d
b
a
g
g
i
n
g
me
t
h
o
d
o
l
o
g
i
e
s
,
”
N
e
u
r
a
l
Pro
c
e
ssi
n
g
L
e
t
t
e
rs,
v
o
l
.
5
0
,
p
p
.
4
3
-
5
6
,
S
e
p
.
2
0
1
8
,
d
o
i
:
1
0
.
1
0
0
7
/
s
1
1
0
6
3
-
0
1
8
-
9
9
1
3
-
6.
[
3
]
C
.
W
o
l
f
,
J
.
M
.
J
o
l
i
o
n
,
a
n
d
F
.
C
h
a
s
sai
n
g
,
“
T
e
x
t
l
o
c
a
l
i
z
a
t
i
o
n
,
e
n
h
a
n
c
e
me
n
t
a
n
d
b
i
n
a
r
i
z
a
t
i
o
n
i
n
mu
l
t
i
me
d
i
a
d
o
c
u
me
n
t
s
,
”
2
0
0
2
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
rn
Re
c
o
g
n
i
t
i
o
n
,
v
o
l
.
2
,
p
p
.
1
0
3
7
-
1
0
4
0
,
2
0
0
2
,
d
o
i
:
1
0
.
1
1
0
9
/
I
C
P
R
.
2
0
0
2
.
1
0
4
8
4
8
2
.
[
4
]
P
.
S
a
h
a
r
e
a
n
d
S
.
B
.
D
h
o
k
,
“
M
u
l
t
i
l
i
n
g
u
a
l
c
h
a
r
a
c
t
e
r
se
g
me
n
t
a
t
i
o
n
a
n
d
r
e
c
o
g
n
i
t
i
o
n
s
c
h
e
mes
f
o
r
I
n
d
i
a
n
d
o
c
u
me
n
t
i
mag
e
s,
”
in
I
EE
E
Ac
c
e
ss
,
v
o
l
.
6
,
p
p
.
1
0
6
0
3
-
1
0
6
1
7
,
2
0
1
8
,
d
o
i
:
1
0
.
1
1
0
9
/
A
C
C
ESS
.
2
0
1
8
.
2
7
9
5
1
0
4
.
[
5
]
U
.
B
h
a
t
t
a
c
h
a
r
y
a
a
n
d
B
.
B
.
C
h
a
u
d
h
u
r
i
,
“
H
a
n
d
w
r
i
t
t
e
n
n
u
mer
a
l
d
a
t
a
b
a
ses
o
f
i
n
d
i
a
n
scri
p
t
s
a
n
d
m
u
l
t
i
s
t
a
g
e
r
e
c
o
g
n
i
t
i
o
n
o
f
mi
x
e
d
n
u
m
e
r
a
l
s,
”
i
n
I
E
EE
T
ra
n
s
a
c
t
i
o
n
s
o
n
Pa
t
t
e
r
n
An
a
l
y
si
s
a
n
d
M
a
c
h
i
n
e
I
n
t
e
l
l
i
g
e
n
c
e
,
v
o
l
.
3
1
,
n
o
.
3
,
p
p
.
4
4
4
-
4
5
7
,
M
a
r
c
h
2
0
0
9
,
d
o
i
:
1
0
.
1
1
0
9
/
TPA
M
I
.
2
0
0
8
.
8
8
.
[
6
]
C
.
N
a
v
e
e
n
a
a
n
d
V
.
N
.
M
.
A
r
a
d
h
y
a
,
“
H
a
n
d
w
r
i
t
t
e
n
c
h
a
r
a
c
t
e
r
se
g
me
n
t
a
t
i
o
n
f
o
r
K
a
n
n
a
d
a
s
c
r
i
p
t
s,
”
2
0
1
2
W
o
r
l
d
C
o
n
g
ress
o
n
I
n
f
o
rm
a
t
i
o
n
a
n
d
C
o
m
m
u
n
i
c
a
t
i
o
n
T
e
c
h
n
o
l
o
g
i
e
s
,
p
p
.
1
4
4
-
1
4
9
,
2
0
1
2
,
d
o
i
:
1
0
.
1
1
0
9
/
W
I
C
T.
2
0
1
2
.
6
4
0
9
0
6
5
.
[
7
]
T.
R
.
B
a
b
u
,
N
.
V
.
R
a
o
,
L.
P
.
R
e
d
d
y
,
T.
S
.
K
.
P
r
a
b
h
u
,
a
n
d
B
.
R
.
B
a
b
u
,
“
S
y
l
l
a
b
l
e
s
e
g
me
n
t
a
t
i
o
n
o
f
T
e
l
u
g
u
d
o
c
u
m
e
n
t
i
ma
g
e
s,
”
2
0
1
0
2
n
d
I
n
t
e
rn
a
t
i
o
n
a
l
C
o
n
f
e
re
n
c
e
o
n
S
i
g
n
a
l
Pro
c
e
ss
i
n
g
S
y
st
e
m
s
,
2
0
1
0
,
p
p
.
7
8
-
8
2
,
d
o
i
:
1
0
.
1
1
0
9
/
I
C
S
P
S
.
2
0
1
0
.
5
5
5
5
2
1
7
.
[
8
]
K
.
L
.
B
a
n
u
m
a
t
h
i
a
n
d
A
.
P
.
J
.
C
h
a
n
d
r
a
,
“
L
i
n
e
a
n
d
w
o
r
d
s
e
g
me
n
t
a
t
i
o
n
o
f
K
a
n
n
a
d
a
h
a
n
d
w
r
i
t
t
e
n
t
e
x
t
d
o
c
u
me
n
t
s
u
si
n
g
p
r
o
j
e
c
t
i
o
n
p
r
o
f
i
l
e
t
e
c
h
n
i
q
u
e
,
”
2
0
1
6
I
n
t
e
rn
a
t
i
o
n
a
l
C
o
n
f
e
r
e
n
c
e
o
n
El
e
c
t
ri
c
a
l
,
El
e
c
t
r
o
n
i
c
s
,
C
o
m
m
u
n
i
c
a
t
i
o
n
,
C
o
m
p
u
t
e
r
a
n
d
O
p
t
i
m
i
z
a
t
i
o
n
T
e
c
h
n
i
q
u
e
s
(
I
C
EE
C
C
O
T
)
,
M
y
s
u
r
u
,
I
n
d
i
a
,
2
0
1
6
,
p
p
.
1
9
6
-
2
0
1
,
d
o
i
:
1
0
.
1
1
0
9
/
I
C
EE
C
C
O
T
.
2
0
1
6
.
7
9
5
5
2
1
4
.
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:
2
5
0
2
-
4
7
52
E
fficien
t d
ee
p
lea
r
n
in
g
m
o
d
els fo
r
telu
g
u
h
a
n
d
w
r
itten
text
r
ec
o
g
n
itio
n
(
B
u
d
d
a
r
a
ju
R
ev
a
th
i
)
1571
[
9
]
N
.
S
.
R
a
n
i
,
A
.
C
.
S
u
b
r
a
m
a
n
i
,
A
.
K
u
mar,
a
n
d
B
.
R
.
P
u
sh
p
a
,
“
D
e
e
p
l
e
a
r
n
i
n
g
n
e
t
w
o
r
k
a
r
c
h
i
t
e
c
t
u
r
e
b
a
se
d
K
a
n
n
a
d
a
h
a
n
d
w
r
i
t
t
e
n
c
h
a
r
a
c
t
e
r
r
e
c
o
g
n
i
t
i
o
n,
”
2
0
2
0
S
e
c
o
n
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
I
n
v
e
n
t
i
v
e
Re
s
e
a
rc
h
i
n
C
o
m
p
u
t
i
n
g
A
p
p
l
i
c
a
t
i
o
n
s
(
I
C
I
R
C
A)
,
p
p
.
2
1
3
-
2
2
0
,
2
0
2
0
,
d
o
i
:
1
0
.
1
1
0
9
/
I
C
I
R
C
A
4
8
9
0
5
.
2
0
2
0
.
9
1
8
3
1
6
0
.
[
1
0
]
R.
S
.
K
u
n
t
e
a
n
d
R
.
D
.
S
.
S
a
m
u
e
l
,
“
A
si
m
p
l
e
a
n
d
e
f
f
i
c
i
e
n
t
o
p
t
i
c
a
l
c
h
a
r
a
c
t
e
r
r
e
c
o
g
n
i
t
i
o
n
sy
s
t
e
m
f
o
r
b
a
s
i
c
sy
mb
o
l
s
i
n
p
r
i
n
t
e
d
K
a
n
n
a
d
a
t
e
x
t
,
”
S
a
d
h
a
n
a
,
v
o
l
.
3
2
,
p
p
.
5
2
1
–
5
3
3
,
2
0
0
7
,
d
o
i
:
1
0
.
1
0
0
7
/
s
1
2
0
4
6
-
0
0
7
-
0
0
3
9
-
1
.
[
1
1
]
S
.
J
.
B
a
s
h
a
,
D
.
V
e
e
r
a
i
a
h
,
G
.
P
a
v
a
n
i
,
S
.
T
.
A
f
r
e
e
n
,
P
.
R
a
j
e
s
h
,
a
n
d
M
.
S
.
S
a
s
a
n
k
,
“
A
n
o
v
e
l
a
p
p
r
o
a
c
h
f
o
r
o
p
t
i
c
a
l
c
h
a
r
a
c
t
e
r
r
e
c
o
g
n
i
t
i
o
n
(
O
C
R
)
o
f
h
a
n
d
w
r
i
t
t
e
n
T
e
l
u
g
u
a
l
p
h
a
b
e
t
s
u
s
i
n
g
c
o
n
v
o
l
u
t
i
o
n
a
l
n
e
u
r
a
l
n
e
t
w
o
r
k
s
,
”
2
0
2
1
S
e
c
o
n
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
E
l
e
c
t
r
o
n
i
c
s
a
n
d
S
u
s
t
a
i
n
a
b
l
e
C
o
m
m
u
n
i
c
a
t
i
o
n
S
y
s
t
e
m
s
(
I
C
E
S
C
)
,
p
p
.
1
4
9
4
-
1
5
0
0
,
2
0
2
1
,
d
o
i
:
1
0
.
1
1
0
9
/
I
C
E
S
C
5
1
4
2
2
.
2
0
2
1
.
9
5
3
2
6
5
8
.
[
1
2
]
N
.
P
r
a
mee
l
a
,
P
.
A
n
j
u
s
h
a
,
a
n
d
R
.
K
a
r
t
h
i
k
,
“
Off
-
l
i
n
e
T
e
l
u
g
u
h
a
n
d
w
r
i
t
t
e
n
c
h
a
r
a
c
t
e
r
s
r
e
c
o
g
n
i
t
i
o
n
u
s
i
n
g
o
p
t
i
c
a
l
c
h
a
r
a
c
t
e
r
r
e
c
o
g
n
i
t
i
o
n
,
”
2
0
1
7
I
n
t
e
r
n
a
t
i
o
n
a
l
C
o
n
f
e
re
n
c
e
o
f
E
l
e
c
t
r
o
n
i
c
s
,
C
o
m
m
u
n
i
c
a
t
i
o
n
a
n
d
Ae
r
o
s
p
a
c
e
T
e
c
h
n
o
l
o
g
y
(
I
C
E
C
A)
,
p
p
.
2
2
3
-
2
2
6
,
2
0
1
7
,
d
o
i
:
1
0
.
1
1
0
9
/
I
C
EC
A
.
2
0
1
7
.
8
2
1
2
8
0
1
.
[
1
3
]
P
.
N
.
S
a
s
t
r
y
,
T
.
R
.
V
.
L
a
k
s
h
m
i
,
N
.
V
.
K
.
Rao
,
T.
V
.
R
a
j
i
n
i
k
a
n
t
h
,
a
n
d
A
.
W
a
h
a
b
,
“
T
e
l
u
g
u
h
a
n
d
w
r
i
t
t
e
n
c
h
a
r
a
c
t
e
r
r
e
c
o
g
n
i
t
i
o
n
u
s
i
n
g
z
o
n
i
n
g
f
e
a
t
u
r
e
s
,
”
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
I
T
C
o
n
v
e
r
g
e
n
c
e
a
n
d
S
e
c
u
r
i
t
y
(
I
C
I
T
C
S
)
,
p
p
.
1
-
4,
2
0
1
4
,
d
o
i
:
1
0
.
1
1
0
9
/
I
C
I
T
C
S
.
2
0
1
4
.
7
0
2
1
8
1
7
.
[
1
4
]
N
.
S
a
r
i
k
a
,
N
.
S
i
r
i
sa
l
a
,
a
n
d
M
.
S
.
V
e
l
p
u
r
u
,
“
C
N
N
b
a
s
e
d
o
p
t
i
c
a
l
c
h
a
r
a
c
t
e
r
r
e
c
o
g
n
i
t
i
o
n
a
n
d
a
p
p
l
i
c
a
t
i
o
n
s,
”
2
0
2
1
6
t
h
I
n
t
e
r
n
a
t
i
o
n
a
l
C
o
n
f
e
re
n
c
e
o
n
I
n
v
e
n
t
i
v
e
C
o
m
p
u
t
a
t
i
o
n
T
e
c
h
n
o
l
o
g
i
e
s (I
C
I
C
T
)
,
p
p
.
6
6
6
-
6
7
2
,
2
0
2
1
,
d
o
i
:
1
0
.
1
1
0
9
/
I
C
I
C
T5
0
8
1
6
.
2
0
2
1
.
9
3
5
8
7
3
5
.
[
1
5
]
N
.
K
a
n
o
p
o
u
l
o
s
,
N
.
V
a
s
a
n
t
h
a
v
a
d
a
,
a
n
d
R
.
L
.
B
a
k
e
r
,
“
D
e
si
g
n
o
f
a
n
i
m
a
g
e
e
d
g
e
d
e
t
e
c
t
i
o
n
f
i
l
t
e
r
u
s
i
n
g
t
h
e
S
o
b
e
l
o
p
e
r
a
t
o
r
,
”
in
I
EEE
J
o
u
rn
a
l
o
f
S
o
l
i
d
-
S
t
a
t
e
C
i
r
c
u
i
t
s
,
v
o
l
.
2
3
,
n
o
.
2
,
p
p
.
3
5
8
-
3
6
7
,
A
p
r
i
l
1
9
8
8
,
d
o
i
:
1
0
.
1
1
0
9
/
4
.
9
9
6
.
[
1
6
]
A
.
C
h
o
u
d
h
a
r
y
,
R
.
R
i
s
h
i
,
a
n
d
S
.
A
h
l
a
w
a
t
,
“
A
n
e
w
c
h
a
r
a
c
t
e
r
s
e
g
m
e
n
t
a
t
i
o
n
a
p
p
r
o
a
c
h
f
o
r
o
f
f
-
l
i
n
e
c
u
r
si
v
e
h
a
n
d
w
r
i
t
t
e
n
w
o
r
ds
,
”
Pro
c
e
d
i
a
C
o
m
p
u
t
e
r
S
c
i
e
n
c
e
,
v
o
l
.
1
7
,
p
p
.
8
8
–
95
,
2
0
1
3
,
d
o
i
:
1
0
.
1
0
1
6
/
j
.
p
r
o
c
s.
2
0
1
3
.
0
5
.
0
1
3
.
[
1
7
]
R
.
G
.
C
a
s
e
y
a
n
d
E
.
Le
c
o
l
i
n
e
t
,
“
A
su
r
v
e
y
o
f
m
e
t
h
o
d
s
a
n
d
st
r
a
t
e
g
i
e
s
i
n
c
h
a
r
a
c
t
e
r
se
g
me
n
t
a
t
i
o
n
,
”
I
EE
E
T
ra
n
s
a
c
t
i
o
n
s
o
n
P
a
t
t
e
r
n
An
a
l
y
si
s
a
n
d
M
a
c
h
i
n
e
I
n
t
e
l
l
i
g
e
n
c
e
,
v
o
l
.
1
8
,
n
o
.
7
,
p
p
.
6
9
0
–
7
0
6
,
J
u
l
.
1
9
9
6
,
d
o
i
:
1
0
.
1
1
0
9
/
3
4
.
5
0
6
7
9
2
.
[
1
8
]
Y
.
H
.
Li
u
,
“
F
e
a
t
u
r
e
e
x
t
r
a
c
t
i
o
n
a
n
d
i
m
a
g
e
r
e
c
o
g
n
i
t
i
o
n
w
i
t
h
c
o
n
v
o
l
u
t
i
o
n
a
l
n
e
u
r
a
l
n
e
t
w
o
r
k
s
,”
J
o
u
r
n
a
l
o
f
Ph
y
si
c
s:
C
o
n
f
e
r
e
n
c
e
S
e
r
i
e
s
,
1
0
8
7
,
0
6
2
0
3
2
,
d
o
i
:
1
0
.
1
0
8
8
/
1
7
4
2
-
6
5
9
6
/
1
0
8
7
/
6
/
0
6
2
0
3
2
.
[
1
9
]
H
.
R
.
M
a
ma
t
h
a
a
n
d
K
.
S
r
i
k
a
n
t
a
m
u
r
t
h
y
,
“
M
o
r
p
h
o
l
o
g
i
c
a
l
o
p
e
r
a
t
i
o
n
s
a
n
d
p
r
o
j
e
c
t
i
o
n
p
r
o
f
i
l
e
s
b
a
s
e
d
se
g
me
n
t
a
t
i
o
n
o
f
h
a
n
d
w
r
i
t
t
e
n
k
a
n
n
a
d
a
d
o
c
u
m
e
n
t
,
”
I
n
t
e
rn
a
t
i
o
n
a
l
J
o
u
rn
a
l
o
f
Ap
p
l
i
e
d
I
n
f
o
rm
a
t
i
o
n
S
y
st
e
m
s
,
I
J
AI
S
J
o
u
r
n
a
l
,
v
o
l
.
4
,
n
o
.
5
,
2
0
1
2
,
d
o
i
:
1
0
.
5
1
2
0
/
i
j
a
i
s1
2
-
4
5
0
7
0
4
.
[
2
0
]
M
.
Jo
g
i
n
,
M
o
h
a
n
a
,
M
.
S
.
M
a
d
h
u
l
i
k
a
,
G
.
D
.
D
i
v
y
a
,
R
.
K
.
M
e
g
h
a
n
a
,
a
n
d
S
.
A
p
o
o
r
v
a
,
“
F
e
a
t
u
r
e
e
x
t
r
a
c
t
i
o
n
u
s
i
n
g
c
o
n
v
o
l
u
t
i
o
n
n
e
u
r
a
l
n
e
t
w
o
r
k
s
(
C
N
N
)
a
n
d
d
e
e
p
l
e
a
r
n
i
n
g
,
”
2
0
1
8
3
rd
I
E
EE
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
R
e
c
e
n
t
T
re
n
d
s
i
n
E
l
e
c
t
r
o
n
i
c
s
,
I
n
f
o
rm
a
t
i
o
n
a
n
d
C
o
m
m
u
n
i
c
a
t
i
o
n
T
e
c
h
n
o
l
o
g
y
(
RTEI
C
T
)
,
p
p
.
2
3
1
9
-
2
3
2
3
,
2
0
1
8
,
d
o
i
:
1
0
.
1
1
0
9
/
R
TEI
C
T4
2
9
0
1
.
2
0
1
8
.
9
0
1
2
5
0
7
.
[
2
1
]
N
.
S
.
R
a
n
i
,
S
.
K
.
V
e
r
m
a
,
a
n
d
A
.
J
o
se
p
h
,
“
A
z
o
n
e
b
a
se
d
a
p
p
r
o
a
c
h
f
o
r
c
l
a
ss
i
f
i
c
a
t
i
o
n
a
n
d
r
e
c
o
g
n
i
t
i
o
n
o
f
T
e
l
u
g
u
h
a
n
d
w
r
i
t
t
e
n
c
h
a
r
a
c
t
e
rs
,
”
I
n
t
e
r
n
a
t
i
o
n
a
l
J
o
u
rn
a
l
o
f
El
e
c
t
r
i
c
a
l
a
n
d
C
o
m
p
u
t
e
r
E
n
g
i
n
e
e
r
i
n
g
(
I
J
EC
E)
,
v
o
l
.
6
,
n
o
.
4
,
p
p
.
1
6
4
7
-
1
6
5
3
,
2
0
1
6
,
d
o
i
:
1
0
.
1
1
5
9
1
/
i
j
e
c
e
.
v
6
i
4
.
1
0
5
5
3
.
[
2
2
]
K
.
H
e
,
X
.
Z
h
a
n
g
,
S
.
R
e
n
,
a
n
d
J
.
S
u
n
,
“
D
e
e
p
r
e
s
i
d
u
a
l
l
e
a
r
n
i
n
g
f
o
r
i
m
a
g
e
r
e
c
o
g
n
i
t
i
o
n
,
”
2
0
1
6
I
EEE
C
o
n
f
e
re
n
c
e
o
n
C
o
m
p
u
t
e
r V
i
s
i
o
n
a
n
d
Pa
t
t
e
rn
Re
c
o
g
n
i
t
i
o
n
(
C
V
PR)
,
p
p
.
7
7
0
-
7
7
8
,
2
0
1
6
,
d
o
i
:
1
0
.
1
1
0
9
/
C
V
P
R
.
2
0
1
6
.
9
0
.
[
2
3
]
F
.
I
a
n
d
o
l
a
,
M
.
M
o
s
k
e
w
i
c
z
,
K
.
A
sh
r
a
f
,
S.
H
a
n
,
W
.
D
a
l
l
y
,
a
n
d
K
.
K
e
u
t
z
e
r
,
“
S
q
u
e
e
z
e
N
e
t
:
A
l
e
x
N
e
t
-
l
e
v
e
l
a
c
c
u
r
a
c
y
w
i
t
h
5
0
x
f
e
w
e
r
p
a
r
a
m
e
t
e
r
s a
n
d
t
e
x
t
l
e
ss
1
M
B
m
o
d
e
l
si
z
e
,
”
a
rXi
v
:
1
6
0
2
.
0
7
3
6
0
,
2
0
1
6
.
[
2
4
]
M
.
S
.
V
e
l
p
u
r
u
,
G
.
T
e
j
a
sr
e
e
,
a
n
d
M
.
R
.
K
u
m
a
r
,
“
T
e
l
u
g
u
h
a
n
d
w
r
i
t
t
e
n
c
h
a
r
a
c
t
e
r
d
a
t
a
s
e
t
,
”
I
EEE
D
a
t
a
p
o
r
t
,
2
0
2
0
,
d
o
i
:
1
0
.
2
1
2
2
7
/
M
w
6
a
-
D
6
6
2
.
[
2
5
]
M
.
M
a
t
h
e
w
,
M
.
Ja
i
n
,
a
n
d
C.
V
.
Jaw
a
h
a
r
,
“
B
e
n
c
h
m
a
r
k
i
n
g
sc
e
n
e
t
e
x
t
r
e
c
o
g
n
i
t
i
o
n
i
n
d
e
v
a
n
a
g
a
r
i
,
T
e
l
u
g
u
a
n
d
M
a
l
a
y
a
l
a
m
,
”
1
4
t
h
I
APR
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
D
o
c
u
m
e
n
t
A
n
a
l
y
s
i
s
a
n
d
R
e
c
o
g
n
i
t
i
o
n
(
I
C
D
AR),
v
o
l
.
9
,
N
o
v
e
m
b
e
r
2
0
1
7
,
p
p
.
9
-
1
5
,
d
o
i
:
1
0
.
1
1
0
9
/
I
C
D
A
R
.
2
0
1
7
.
3
6
4
.
B
I
O
G
RAP
H
I
E
S O
F
AUTH
O
RS
Bu
d
d
a
r
a
ju
Re
v
a
t
h
i
c
u
rre
n
tl
y
wo
rk
i
n
g
a
s
a
n
As
sista
n
t
P
r
o
fe
ss
o
r
a
t
S
.
R.
K.
R
En
g
i
n
e
e
rin
g
c
o
ll
e
g
e
,
h
o
ld
s
a
n
M
.
Tec
h
De
g
re
e
fro
m
S
R
K
R
En
g
i
n
e
e
rin
g
Co
l
leg
e
.
He
r
a
c
a
d
e
m
ic
p
u
rsu
it
s
c
e
n
ter
a
ro
u
n
d
wire
les
s
c
o
m
m
u
n
ica
ti
o
n
,
n
a
tu
ra
l
lan
g
u
a
g
e
p
r
o
c
e
ss
in
g
,
a
n
d
d
a
ta sc
ie
n
c
e
.
S
h
e
c
a
n
b
e
c
o
n
tac
te
d
a
t
e
m
a
il
:
b
u
d
d
a
ra
ju
.
re
v
a
t
h
i@g
m
a
il
.
c
o
m
.
B
.
N
.
V
.
Na
r
a
sim
h
a
Ra
j
u
c
u
rre
n
tl
y
wo
rk
in
g
a
s
a
n
As
sista
n
t
P
r
o
fe
ss
o
r
a
t
S
.
R.
K.R
E
n
g
i
n
e
e
rin
g
c
o
ll
e
g
e
,
h
o
ld
s
a
n
M
.
Tec
h
De
g
re
e
fro
m
S
R
K
R
E
n
g
in
e
e
rin
g
Co
ll
e
g
e
.
He
r
a
c
a
d
e
m
ic
p
u
rsu
it
s
c
e
n
ter
a
ro
u
n
d
n
a
tu
ra
l
la
n
g
u
a
g
e
p
ro
c
e
ss
in
g
,
m
a
c
h
in
e
tran
sla
ti
o
n
,
a
rti
ficia
l
in
telli
g
e
n
c
e
a
n
d
d
e
e
p
lea
rn
i
n
g
.
He
c
a
n
b
e
c
o
n
tac
ted
a
t
e
m
a
il
:
b
u
d
d
a
ra
ju
.
n
a
ra
sim
h
a
ra
ju
@g
m
a
il
.
c
o
m
.
Evaluation Warning : The document was created with Spire.PDF for Python.
I
SS
N
:
2
5
0
2
-
4
7
52
I
n
d
o
n
esian
J
E
lec
E
n
g
&
C
o
m
p
Sci
,
Vo
l.
3
6
,
No
.
3
,
Dec
em
b
er
20
2
4
:
1
5
6
4
-
1
5
7
2
1572
Bo
d
d
u
L
V
S
i
v
a
Ra
m
a
K
r
ish
n
a
is
wo
r
k
i
n
g
a
t
S
R
M
U
n
iv
e
r
sity
-
AP,
S
i
n
c
e
2
0
2
3
S
e
p
tem
b
e
r
a
s
As
sista
n
t
P
ro
fe
ss
o
r
in
t
h
e
De
p
a
rtme
n
t
o
f
Co
m
p
u
ter
S
c
ien
c
e
a
n
d
En
g
i
n
e
e
rin
g
.
He
is
w
o
r
k
e
d
a
t
S
a
g
i
Ra
m
a
Krish
n
a
m
Ra
ju
En
g
i
n
e
e
rin
g
Co
ll
e
g
e
fro
m
2
0
1
6
M
a
rc
h
1
st
t
o
2
0
2
3
Au
g
u
st
a
s
an
As
sista
n
t
P
ro
fe
ss
o
r
in
th
e
De
p
a
rt
m
e
n
t
o
f
C
o
m
p
u
ter
S
c
ien
c
e
a
n
d
En
g
i
n
e
e
rin
g
.
Re
se
a
rc
h
a
re
a
is
d
a
ta
m
in
in
g
,
m
a
c
h
in
e
lea
rn
in
g
,
d
e
e
p
lea
rn
in
g
tran
sla
ti
o
n
.
h
e
is
a
li
fe
m
e
m
b
e
r
o
f
c
o
m
p
u
ter
so
c
iety
o
f
In
d
ia
(CS
I),
a
n
d
i
n
stit
u
te
o
f
e
n
g
in
e
e
rs
(IE
).
H
e
c
a
n
b
e
c
o
n
tac
ted
a
t
e
m
a
il
:
siv
a
ra
m
a
k
rish
n
a
.
b
@s
rm
a
p
.
e
d
u
.
i
n
a
n
d
k
rish
n
a
2
9
2
8
@g
m
a
il
.
c
o
m
.
Aja
y
Dili
p
K
u
m
a
r
M
a
r
a
p
a
tl
a
is
a
n
A
ss
istan
t
p
ro
fe
ss
o
r
in
t
h
e
De
p
a
rtme
n
t
o
f
Co
m
p
u
ter
S
c
ien
c
e
a
n
d
En
g
in
e
e
ri
n
g
a
t
S
RM
U
n
iv
e
rsit
y
.
His
re
se
a
rc
h
a
re
a
s
a
re
In
tern
e
t
o
f
Th
in
g
s,
a
n
d
M
a
c
h
in
e
Lea
rn
i
n
g
.
He
h
a
s
1
2
Ye
a
rs
o
f
tea
c
h
in
g
e
x
p
e
rien
c
e
.
Cu
rre
n
tl
y
h
e
is
p
u
rsu
i
n
g
P
h
.
D.
fro
m
P
o
n
d
ich
e
rr
y
Ce
n
tral
U
n
iv
e
rsit
y
,
P
o
n
d
ich
e
rr
y
.
H
e
c
a
n
b
e
c
o
n
tac
ted
a
t
e
m
a
il
:
d
il
e
e
p
m
a
ra
p
a
tl
a
@g
m
a
il
.
c
o
m
.
S
.
S
u
r
y
a
n
a
r
a
y
a
n
a
r
a
j
u
is
p
r
e
se
n
tl
y
wo
r
k
i
n
g
a
s
a
n
a
ss
istan
t
p
ro
fe
ss
o
r
in
th
e
De
p
a
rtme
n
t
o
f
In
f
o
rm
a
ti
o
n
Tec
h
n
o
lo
g
y
,
S
RKR
En
g
in
e
e
ri
n
g
C
o
ll
e
g
e
(A),
c
h
i
n
a
a
m
iram
,
Bh
ima
v
a
ra
m
,
An
d
h
ra
P
ra
d
e
sh
,
I
n
d
ia
.
He
c
o
m
p
lete
d
h
is
B.
Tec
h
i
n
2
0
0
8
,
M
.
Tec
h
in
2
0
1
0
.
He
h
a
s
p
u
b
li
sh
e
d
7
j
o
u
r
n
a
l
p
a
p
e
rs
a
n
d
2
c
o
n
fe
re
n
c
e
p
a
p
e
rs.
His
re
se
a
rc
h
in
tere
sts
in
c
lu
d
e
d
a
tam
in
in
g
,
a
rti
ficia
l
i
n
telli
g
e
n
c
e
,
m
a
c
h
in
e
lea
rn
in
g
,
b
i
g
d
a
ta,
a
n
d
i
n
te
rn
e
t
o
f
th
i
n
g
s.
H
e
c
a
n
b
e
c
o
n
tac
ted
a
t
e
m
a
il
:
s
u
ry
a
n
a
ra
y
a
n
a
ra
ju
.
s@
g
m
a
il
.
c
o
m
.
Evaluation Warning : The document was created with Spire.PDF for Python.