I
A
E
S
I
n
t
e
r
n
at
io
n
al
Jou
r
n
al
of
A
r
t
if
ic
ia
l
I
n
t
e
ll
ig
e
n
c
e
(
I
J
-
AI
)
V
ol
.
14
, N
o.
5
,
O
c
to
be
r
2025
, pp.
3970
~
3981
I
S
S
N
:
2252
-
8938
,
D
O
I
:
10.11591/
ij
a
i.
v
14
.i
5
.pp
3970
-
3981
3970
Jou
r
n
al
h
om
e
page
:
ht
tp
:
//
ij
ai
.
ia
e
s
c
or
e
.c
om
O
p
t
i
m
i
z
i
n
g n
i
t
i
k
b
at
i
k
c
l
ass
i
f
i
c
at
i
on
t
h
r
ou
gh
c
om
p
ar
at
i
ve
an
al
ysi
s of
i
m
age
au
gm
e
n
t
at
i
on
S
u
p
r
ap
t
o
1
, M
e
il
an
y N
on
s
i
T
e
n
t
u
a
2
, A
h
m
ad
R
i
z
k
i
M
au
la
n
a
1
1
D
e
pa
r
t
m
e
nt
of
C
om
put
e
r
S
c
i
e
nc
e
a
nd E
l
e
c
t
r
oni
c
s
, F
a
c
ul
t
y of
M
a
t
he
m
a
t
i
c
s
a
nd
N
a
t
ur
a
l
S
c
i
e
nc
e
s
, U
ni
ve
r
s
i
t
a
s
G
a
dj
a
h
M
a
da
,
Y
ogya
ka
r
t
a
, I
ndone
s
i
a
2
D
e
pa
r
t
m
e
nt
of
I
nf
or
m
a
t
i
c
s
, F
a
c
ul
t
y of
S
c
i
e
nc
e
a
nd T
e
c
hnol
ogy, U
ni
ve
r
s
i
t
a
s
P
G
R
I
Y
ogya
ka
r
t
a
, Y
ogya
ka
r
t
a
, I
ndone
s
i
a
A
r
t
ic
le
I
n
f
o
A
B
S
T
R
A
C
T
A
r
ti
c
le
h
is
to
r
y
:
R
e
c
e
iv
e
d
N
ov
21
,
2024
R
e
vi
s
e
d
J
ul
4
,
2025
A
c
c
e
pt
e
d
A
ug
6
,
2025
Nitik
batik
is
one
of
the
most
intricate
and
culturally
significant
m
otifs
in
Yogyakarta'
s
batik
tradition,
characterized
by
its
complex,
geometri
c
dot
-
based
patterns.
The
unique
challenges
of
automatically
classifying
nitik
batik
motifs
stem
from
the
high
variability
within
the
class
and
the
l
imited
availabil
ity of
training
data. Th
is stu
dy
invest
igates h
ow different
ima
ge data
augmentat
ion
techniqu
es
can
enhance
the
performance
of
a
random
forest
classifi
er
for
nitik
batik
motifs
.
T
echniques
such
as
geo
metric
transf
ormations
(flip,
rotate,
and
scaling),
intensity
transformations
(c
ut
-
out,
grid
mask,
and
random
erasing),
non
-
instance
level
augmentation
(
pairing
samples),
and
unconditional
image
generation
(deep
convol
utional
generative
adversarial
network
(
DCGAN
)
)
were
used
to
expand
the
dataset
and
improve
the
model'
s
ability
to
generalize.
The
results
show
that
s
pecific
techniques,
notably
flip,
cut
-
out,
and
DCGAN,
significantly
im
proved
classifi
cation
accuracy,
with
flip
achieving
the
highest
ac
curacy
improvement
of
20.20%,
followed
by
cut
-
out
at
19.27%
and
DCG
AN
at
16.25%.
Moreover,
DCGAN
demonstrated
the
lowest
standard
de
viation
(0.78%),
indicatin
g
high
stabilit
y
and
robustness
in
classif
ication
performance
across
multiple
validation
folds.
These
findings
suggest
that
augmentat
ion
techniqu
es
effectively
improve
classifi
cation
accurac
y
and
enhance the
model'
s abili
ty to
generalize fr
om lim
ited and
complex
datasets
.
K
e
y
w
o
r
d
s
:
D
C
G
A
N
G
e
om
e
tr
ic
t
r
a
ns
f
or
m
a
ti
on
I
m
a
ge
da
ta
a
ugm
e
nt
a
ti
on
I
nt
e
ns
it
y
tr
a
ns
f
or
m
a
ti
on
N
it
ik
ba
ti
k m
ot
if
This is an
open
acce
ss artic
le unde
r the
CC BY
-
SA
license.
C
or
r
e
s
pon
di
n
g A
u
th
or
:
S
upr
a
pt
o
D
e
pa
r
tm
e
nt
of
C
om
put
e
r
S
c
ie
nc
e
a
nd E
le
c
tr
oni
c
s
, F
a
c
ul
ty
of
M
a
th
e
m
a
ti
c
s
a
nd N
a
tu
r
a
l
S
c
ie
nc
e
s
U
ni
ve
r
s
it
a
s
G
a
dj
a
h M
a
da
N
or
th
S
e
ki
p
, B
ul
a
k S
um
ur
, Y
ogya
ka
r
ta
55281, I
ndone
s
ia
E
m
a
il
:
s
pr
a
pt
o@
ugm
.a
c
.i
d
1.
I
N
T
R
O
D
U
C
T
I
O
N
B
a
ti
k
is
I
ndone
s
ia
’
s
r
ic
h
a
nd
di
ve
r
s
e
c
ul
tu
r
a
l
h
e
r
it
a
ge
,
w
it
h
m
ot
if
s
th
a
t
r
e
f
le
c
t
r
e
gi
ona
l
id
e
nt
it
ie
s
a
nd
hi
gh
a
e
s
th
e
ti
c
va
lu
e
s
[
1]
.
A
ut
om
a
ti
c
c
la
s
s
if
ic
a
ti
on
of
ba
ti
k
m
o
ti
f
s
is
vi
ta
l
in
c
ul
tu
r
a
l
pr
e
s
e
r
va
ti
on,
c
ol
le
c
ti
on
m
a
na
ge
m
e
nt
,
a
nd
d
e
ve
lo
pi
ng
im
a
ge
-
ba
s
e
d
a
ppl
ic
a
ti
on
s
r
e
la
te
d
to
th
e
c
r
e
a
ti
ve
in
du
s
tr
y
[
2]
.
O
ne
of
th
e
ol
de
s
t
ba
ti
k
m
ot
if
s
ty
pi
c
a
l
of
Y
ogya
ka
r
ta
i
s
ni
ti
k
b
a
ti
k
.
N
it
ik
ba
t
ik
m
ot
if
s
a
r
e
c
om
pl
e
x
m
ot
if
s
c
on
s
is
ti
ng
of
th
ous
a
nds
of
dot
s
a
r
r
a
nge
d
a
nd
m
e
a
s
ur
e
d
in
s
uc
h
a
w
a
y
a
s
to
f
or
m
ge
om
e
tr
ic
s
pa
c
e
s
,
a
ngl
e
s
,
a
nd
f
ie
ld
s
[
3]
.
T
he
c
la
s
s
if
ic
a
ti
on
of
ni
ti
k
ba
ti
k
m
ot
if
s
f
a
c
e
s
c
ha
ll
e
ng
e
s
,
in
c
lu
di
ng
hi
gh
in
tr
a
c
la
s
s
va
r
ia
ti
on,
pa
tt
e
r
n
c
om
pl
e
xi
ty
, a
nd l
im
it
e
d da
ta
a
va
il
a
bl
e
f
or
m
ode
l
tr
a
in
in
g
[
4]
.
I
n
c
om
put
e
r
vi
s
io
n,
i
m
a
ge
c
la
s
s
if
ic
a
ti
on
te
c
hni
que
s
ha
ve
ove
r
gr
ow
n
ow
in
g
to
a
dva
nc
e
s
in
m
a
c
hi
ne
le
a
r
ni
ng
a
lg
or
it
hm
s
a
nd
th
e
a
va
il
a
bi
li
ty
of
e
xt
e
ns
iv
e
da
ta
.
S
om
e
b
a
t
ik
c
l
a
s
s
if
i
c
a
t
io
n
m
o
de
li
n
g
h
a
s
b
e
e
n
d
on
e
us
in
g
m
a
c
hi
n
e
l
e
a
r
n
in
g a
lg
or
it
hm
s
s
u
c
h a
s
k
-
ne
a
r
e
s
t
ne
i
gh
bor
s
(
K
N
N
)
[
5
]
,
s
upp
or
t
v
e
c
to
r
m
a
c
h
in
e
s
(
S
V
M
)
[
6
]
,
Evaluation Warning : The document was created with Spire.PDF for Python.
I
nt
J
A
r
ti
f
I
nt
e
ll
I
S
S
N
:
2252
-
8938
O
pt
imi
z
in
g ni
ti
k
bat
ik
c
la
s
s
if
ic
at
io
n t
hr
ough c
om
pa
r
at
iv
e
analy
s
is
of
i
m
age
augme
nt
at
io
n
(
Supr
apt
o
)
3971
ba
c
kpr
opa
ga
ti
on
ne
ur
a
l
ne
twor
ks
(
B
N
N
)
[
7]
,
a
nd
de
c
is
io
n
tr
ee
s
[
8]
.
M
e
a
n
w
hi
le
,
c
la
s
s
if
ic
a
ti
on
us
in
g
d
e
e
p
le
a
r
ni
ng
a
lg
or
it
hm
s
f
or
ba
ti
k
m
ode
li
ng
m
a
in
ly
us
e
s
c
onv
ol
ut
io
na
l
ne
ur
a
l
ne
twor
ks
(
C
N
N
)
[
9]
.
T
he
pe
r
f
or
m
a
nc
e
of
c
la
s
s
if
ic
a
ti
on
m
od
e
ls
i
s
hi
ghl
y
de
pe
nd
e
nt
on
th
e
qua
li
ty
a
nd
qu
a
nt
it
y
of
da
ta
u
s
e
d
dur
in
g
th
e
tr
a
in
in
g
pr
oc
e
s
s
[
10]
,
in
c
lu
di
ng
ba
ti
k
c
la
s
s
if
ic
a
ti
on
m
ode
li
n
g.
I
m
a
ge
da
ta
a
ugm
e
nt
a
ti
on
is
a
n
e
f
f
e
c
ti
ve
m
e
th
od
to
im
pr
ove
m
ode
l
pe
r
f
or
m
a
nc
e
[
11]
.
I
m
a
ge
da
ta
a
ugm
e
nt
a
ti
on
m
a
ni
pul
a
te
s
th
e
or
ig
in
a
l
da
ta
to
pr
oduc
e
ne
w
va
r
ia
ti
ons
th
a
t
c
a
n
e
nr
ic
h
th
e
da
ta
s
e
t.
T
he
m
ode
l
c
a
n
le
a
r
n
be
tt
e
r
a
nd
ge
ne
r
a
li
z
e
m
o
r
e
to
da
ta
th
a
t
ha
s
ne
ve
r
be
e
n
s
e
e
n b
e
f
or
e
[
12]
.
A
ugm
e
nt
a
ti
on
te
c
hni
que
s
c
a
n
b
e
c
a
te
gor
iz
e
d
in
to
s
e
v
e
r
a
l
gr
oups
,
in
c
lu
di
ng
ge
om
e
tr
ic
tr
a
ns
f
or
m
a
ti
on,
in
te
ns
it
y
tr
a
ns
f
or
m
a
ti
on,
non
-
in
s
ta
nc
e
le
ve
l
a
ugm
e
nt
a
ti
on,
a
nd
unc
ondi
ti
one
d
im
a
ge
ge
ne
r
a
ti
on
[
13]
.
G
e
om
e
tr
ic
tr
a
ns
f
or
m
a
ti
on
[
14]
a
nd
in
te
ns
it
y
tr
a
ns
f
or
m
a
ti
on
[
15]
a
r
e
of
te
n
us
e
d
in
a
ugm
e
nt
a
ti
on
te
c
hni
que
s
in
ba
ti
k
c
la
s
s
if
ic
a
ti
on
m
ode
li
ng.
B
o
th
a
ugm
e
nt
a
ti
on
te
c
hni
que
s
c
a
n
in
c
r
e
a
s
e
th
e
in
va
r
ia
nc
e
of
th
e
m
ode
l
to
c
ha
nge
s
in
pos
it
io
n.
T
he
y
c
a
n
h
e
lp
th
e
m
ode
l
de
a
l
w
it
h
li
ght
in
g
a
nd
te
xt
ur
e
va
r
ia
ti
ons
in
th
e
or
ig
in
a
l
im
a
ge
.
H
ow
e
ve
r
,
unc
ondi
ti
ona
l
im
a
ge
ge
ne
r
a
ti
on
a
ugm
e
nt
a
ti
on
te
c
hni
que
s
ha
ve
ye
t
to
be
us
e
d
f
or
da
ta
a
ddi
ti
on
in
c
la
s
s
if
ic
a
ti
on
m
ode
li
ng,
e
s
pe
c
i
a
l
ly
ni
ti
k
ba
ti
k
.
U
nc
ondi
ti
ona
l
im
a
ge
ge
ne
r
a
ti
on
te
c
hni
que
s
,
s
uc
h
a
s
de
e
p
c
onvolut
io
na
l
ge
ne
r
a
ti
ve
a
dve
r
s
a
r
i
a
l
ne
twor
k
(
D
C
G
A
N
)
,
onl
y
r
e
pr
oduc
e
ne
w
im
a
ge
s
t
o pr
oduc
e
pr
e
vi
ous
ly
unknown ba
ti
k pa
tt
e
r
ns
[
16]
, [
17]
.
T
hi
s
s
tu
dy a
im
s
to
f
il
l
th
is
ga
p
by c
onduc
ti
ng
a
c
om
p
a
r
a
ti
ve
s
t
udy
of
th
e
a
ugm
e
nt
a
ti
on
m
e
th
ods
a
nd
a
na
ly
z
in
g t
he
ir
i
m
pa
c
t
on t
he
a
c
c
ur
a
c
y
a
nd r
obus
tn
e
s
s
of
t
he
ni
ti
k
ba
ti
k m
ot
if
c
la
s
s
if
ic
a
ti
on mode
l.
T
he
us
e
of
publ
ic
da
ta
s
e
t
s
on
ni
ti
k
ba
ti
k
c
lo
th
m
ot
if
s
pr
ovi
de
s
a
n
oppor
tu
ni
ty
to
c
om
pr
e
he
ns
iv
e
ly
e
va
lu
a
te
how
th
e
s
e
a
ugm
e
nt
a
ti
on
te
c
hni
que
s
a
f
f
e
c
t
th
e
p
e
r
f
or
m
a
nc
e
of
th
e
c
la
s
s
if
i
c
a
ti
on
m
ode
l.
T
he
goa
l
is
to
und
e
r
s
ta
nd
be
tt
e
r
how
th
e
s
e
t
e
c
hni
que
s
c
a
n
e
nha
nc
e
th
e
e
f
f
e
c
ti
ve
n
e
s
s
a
nd
e
f
f
ic
ie
nc
y
of
th
e
ni
ti
k
ba
ti
k
c
la
s
s
if
ic
a
ti
on
s
ys
t
e
m
.
O
ur
c
ont
r
ib
ut
io
n i
s
s
um
m
a
r
iz
e
d a
s
f
ol
lo
w
s
:
‒
T
he
s
tu
dy
in
tr
oduc
e
s
a
nd
c
om
pa
r
e
s
th
e
e
f
f
e
c
ti
ve
n
e
s
s
of
m
ul
ti
pl
e
im
a
ge
a
ugm
e
nt
a
ti
on
te
c
hni
que
s
—
ge
om
e
tr
ic
tr
a
ns
f
or
m
a
ti
ons
,
in
te
ns
it
y
tr
a
ns
f
or
m
a
ti
ons
,
no
n
-
in
s
ta
nc
e
le
ve
l
a
ugm
e
nt
a
ti
on
,
a
nd
unc
ondi
ti
one
d
im
a
ge
ge
ne
r
a
ti
on
—
f
or
im
pr
ovi
ng
ni
ti
k
ba
ti
k
c
la
s
s
if
ic
a
ti
on
.
B
y
e
xpl
or
in
g
a
w
id
e
r
a
nge
of
a
ugm
e
nt
a
ti
on
m
e
th
ods
,
th
e
s
tu
dy
pr
ovi
de
s
a
c
om
pr
e
h
e
ns
iv
e
u
nde
r
s
ta
ndi
ng
of
how
di
f
f
e
r
e
nt
te
c
hni
que
s
im
pa
c
t
m
ode
l
a
c
c
ur
a
c
y a
nd
s
ta
bi
li
ty
.
‒
B
y
a
ppl
yi
ng
th
e
r
a
ndom
f
or
e
s
t
c
la
s
s
if
ie
r
in
c
om
bi
na
ti
on
w
it
h
a
ugm
e
nt
a
ti
on
te
c
hni
que
s
,
th
e
s
tu
dy
de
m
ons
tr
a
te
s
s
ig
ni
f
ic
a
nt
im
pr
ove
m
e
nt
s
in
c
la
s
s
if
ic
a
ti
on
a
c
c
ur
a
c
y,
e
s
pe
c
ia
ll
y
th
r
ough
te
c
hni
que
s
li
ke
f
li
p, c
ut
-
ou
t,
a
nd D
C
G
A
N
.
‒
T
he
s
tu
dy
e
m
pha
s
iz
e
s
not
onl
y
th
e
a
c
c
ur
a
c
y
but
a
ls
o
th
e
s
ta
bi
li
ty
of
th
e
m
ode
l,
a
s
m
e
a
s
ur
e
d
by
th
e
s
ta
nda
r
d
de
vi
a
ti
on
of
c
r
os
s
-
va
li
da
ti
on
a
c
c
ur
a
c
y.
T
he
f
in
di
ngs
s
how
th
a
t
c
e
r
ta
in
a
ugm
e
nt
a
ti
on
te
c
hni
que
s
,
pa
r
ti
c
ul
a
r
ly
D
C
G
A
N
,
pr
ovi
de
hi
gh
s
ta
bi
li
ty
,
w
hi
c
h
is
c
r
uc
ia
l
f
or
de
pl
oyi
ng
r
e
li
a
bl
e
c
la
s
s
if
ic
a
ti
on mode
ls
i
n pr
a
c
ti
c
e
.
2.
R
E
L
A
T
E
D
P
A
P
E
R
D
a
ta
a
ugm
e
nt
a
t
io
n
is
e
s
s
e
n
ti
a
l
i
n
ov
e
r
c
om
in
g
da
ta
s
e
t
li
m
i
t
a
ti
ons
b
y
a
r
ti
f
ic
ia
ll
y
e
x
pa
n
di
ng
t
he
tr
a
in
in
g
da
ta
,
w
h
il
e
va
r
io
us
m
ode
li
ng
te
c
h
ni
que
s
ha
ve
e
m
e
r
ge
d
to
i
m
p
r
ov
e
i
m
a
ge
c
la
s
s
if
ic
a
ti
on
pe
r
f
o
r
m
a
nc
e
.
T
hi
s
r
e
v
ie
w
e
xa
m
in
e
s
r
e
c
e
nt
r
e
s
e
a
r
c
h
'
s
r
ol
e
i
n
da
t
a
a
ug
m
e
n
ta
ti
o
n
a
n
d
b
a
t
ik
m
ode
li
ng
s
tr
a
te
g
ie
s
.
A
c
r
os
s
t
he
s
t
udi
e
s
r
e
vi
e
w
e
d
,
da
ta
a
u
gm
e
nt
a
ti
on
ge
om
e
t
r
ic
t
r
a
ns
f
o
r
m
a
t
io
ns
te
c
h
ni
que
s
w
e
r
e
f
r
e
qu
e
nt
ly
us
e
d
f
o
r
c
la
s
s
i
f
i
c
a
t
io
n
m
o
de
l
o
f
ba
ti
k
,
s
uc
h
a
s
f
l
i
ppi
ng
[
1
4]
,
[
18
]
,
[
1
9]
,
r
ot
a
t
io
n
[
2
0
]
,
s
c
a
li
ng
[
2
1]
,
[
22
]
,
s
he
a
r
in
g
[
23
]
,
a
n
d
n
oi
s
e
i
nj
e
c
t
io
n
[
24
]
to
im
pr
ove
m
ode
l
ge
ne
r
a
li
z
a
t
io
n
a
n
d
r
e
d
uc
e
ove
r
f
i
tt
in
g
.
M
or
e
a
d
va
nc
e
d
m
e
t
ho
ds
,
s
uc
h
a
s
r
a
n
do
m
e
r
a
s
in
g
da
ta
[
2
5
]
a
nd
b
r
ig
ht
ne
s
s
m
o
dul
a
t
io
n
[
15
]
,
[
26
]
,
a
r
e
im
p
le
m
e
n
te
d
t
o i
np
ut
da
ta
va
r
ia
ti
ons
.
V
a
r
io
us
c
la
s
s
if
ic
a
ti
on
m
e
th
od
s
ha
ve
be
e
n
a
ppl
ie
d
to
a
ut
om
a
te
th
e
r
e
c
ogni
ti
on
of
ba
ti
k
m
ot
if
s
,
e
a
c
h
a
ddr
e
s
s
in
g
th
e
uni
que
c
ha
ll
e
nge
s
pos
e
d
by
th
e
c
om
pl
e
x
a
nd
hi
ghl
y
de
ta
il
e
d
pa
tt
e
r
ns
in
ba
ti
k
f
a
br
ic
s
.
T
r
a
di
ti
ona
l
m
a
c
hi
ne
l
e
a
r
ni
ng t
e
c
hni
que
s
, s
uc
h a
s
S
V
M
[
27]
, K
N
N
[
28]
,
B
N
N
[
29]
,
a
nd
de
c
is
io
n t
r
e
e
[
8]
ha
ve
be
e
n
w
id
e
ly
us
e
d
f
or
b
a
ti
k
c
la
s
s
if
ic
a
ti
on
du
e
to
th
e
ir
s
im
pl
ic
it
y
a
nd
e
f
f
e
c
ti
ve
ne
s
s
in
ha
ndl
in
g
s
m
a
ll
-
s
c
a
le
da
ta
s
e
ts
.
M
or
e
r
e
c
e
nt
ly
,
d
e
e
p
le
a
r
ni
ng
a
ppr
oa
c
he
s
,
p
a
r
ti
c
ul
a
r
ly
C
N
N
,
ha
ve
ga
in
e
d
pr
om
in
e
nc
e
in
ba
ti
k
c
la
s
s
if
ic
a
ti
on
t
a
s
ks
[
2]
,
[
9]
,
[
30]
.
C
N
N
s
a
ut
om
a
te
th
e
f
e
a
tu
r
e
e
xt
r
a
c
ti
on,
le
a
r
ni
ng
in
tr
ic
a
te
p
a
tt
e
r
ns
di
r
e
c
tl
y
f
r
om
r
a
w
im
a
ge
s
,
m
a
ki
ng
th
e
m
hi
ghl
y
e
f
f
e
c
ti
ve
f
or
c
om
pl
e
x
ba
ti
k
m
ot
if
s
,
in
c
lu
di
ng
ni
ti
k
ba
ti
k
[
31]
.
T
he
a
bi
li
ty
of
C
N
N
s
to
c
a
pt
ur
e
m
ul
ti
-
le
ve
l
f
e
a
tu
r
e
s
,
f
r
om
e
dge
s
to
te
xt
ur
e
s
,
ha
s
s
ig
ni
f
ic
a
nt
ly
im
pr
ove
d
th
e
a
c
c
ur
a
c
y
a
nd
e
f
f
ic
ie
nc
y
of
ba
ti
k
m
o
ti
f
c
la
s
s
if
ic
a
ti
on.
H
ow
e
ve
r
,
C
N
N
s
of
te
n
r
e
qui
r
e
la
r
ge
a
m
ount
s
of
la
be
le
d
da
ta
, w
hi
c
h c
a
n be
a
l
im
it
a
ti
on f
or
ba
ti
k
da
ta
s
e
ts
.
I
n a
ddi
ti
on t
o C
N
N
s
,
r
a
ndom
f
or
e
s
t
ha
ve
be
e
n e
m
pl
oye
d f
or
S
u
r
a
ka
r
ta
ba
ti
k f
a
br
ic
c
la
s
s
if
ic
a
ti
on due
to
it
s
r
obus
tn
e
s
s
a
nd
a
bi
li
ty
to
ha
ndl
e
s
m
a
ll
,
im
ba
la
nc
e
d
d
a
ta
s
e
ts
[
32]
.
R
a
ndom
f
or
e
s
t
c
r
e
a
te
s
a
n
e
ns
e
m
bl
e
of
de
c
is
io
n
tr
e
e
s
,
e
a
c
h
tr
a
in
e
d
on
di
f
f
e
r
e
nt
pa
r
ts
of
th
e
da
ta
, a
ll
ow
in
g
th
e
m
ode
l
to
c
a
pt
ur
e
va
r
io
us
f
e
a
tu
r
e
s
f
r
om
ba
ti
k
m
ot
if
s
a
nd
pr
ovi
di
ng
s
ta
bl
e
pr
e
di
c
ti
ons
e
v
e
n
w
it
h
li
m
it
e
d
da
ta
.
T
he
e
xi
s
ti
ng
li
te
r
a
tu
r
e
s
s
how
th
a
t
im
a
ge
da
ta
a
ugm
e
nt
a
ti
on
s
ig
ni
f
ic
a
nt
ly
im
pa
c
t
s
th
e
pe
r
f
or
m
a
nc
e
of
c
l
a
s
s
if
ic
a
ti
on
m
ode
ls
,
e
s
pe
c
ia
ll
y
in
th
e
c
ont
e
xt
of
Evaluation Warning : The document was created with Spire.PDF for Python.
I
S
S
N
:
2252
-
8938
I
nt
J
A
r
ti
f
I
nt
e
ll
, V
ol
.
14
, N
o.
5
,
O
c
to
be
r
20
25
:
3970
-
3981
3972
li
m
it
e
d
or
c
om
pl
e
x
da
ta
s
e
ts
. T
hi
s
s
tu
dy
a
im
s
to
c
om
pa
r
e
v
a
r
io
us
a
ugm
e
nt
a
ti
on
te
c
hni
que
s
in
c
la
s
s
if
yi
ng
ni
ti
k
ba
ti
k
m
ot
if
s
a
nd
p
r
ovi
de
ne
w
c
ont
r
ib
u
ti
ons
to
de
ve
lo
pi
ng
m
or
e
e
f
f
e
c
ti
ve
c
la
s
s
if
ic
a
ti
on
m
e
th
ods
f
or
c
ul
tu
r
e
-
ba
s
e
d a
ppl
ic
a
ti
ons
.
3.
M
E
T
H
O
D
T
he
r
e
s
e
a
r
c
h
us
e
s
a
c
om
pa
r
a
ti
ve
a
ppr
oa
c
h
.
S
e
ve
r
a
l
a
ugm
e
nt
a
ti
on
te
c
hni
que
s
a
r
e
a
ppl
ie
d
to
th
e
ni
t
ik
ba
ti
k m
ot
if
da
ta
s
e
t,
a
nd t
he
r
e
s
ul
ts
a
r
e
c
om
pa
r
e
d ba
s
e
d on c
la
s
s
if
ic
a
ti
on pe
r
f
or
m
a
nc
e
m
e
tr
ic
s
. F
ig
ur
e
1
s
how
s
th
e
s
ta
ge
s
of
t
he
r
e
s
e
a
r
c
h m
e
th
odol
ogy tha
t
w
e
r
e
c
ondu
c
te
d i
n t
he
r
e
s
e
a
r
c
h.
F
ig
ur
e
1. T
he
c
onduc
te
d r
e
s
e
a
r
c
h m
e
th
odol
ogy
3.1. Dat
a c
ol
le
c
t
io
n
T
he
publ
ic
d
a
ta
s
e
t
c
ont
a
in
s
im
a
ge
s
of
ni
ti
k
ba
ti
k
m
ot
if
s
[
33]
.
T
hi
s
da
t
a
s
e
t
c
ons
is
t
s
of
240
im
a
ge
s
,
c
ons
is
ti
ng
of
60
ni
ti
k
ba
ti
k
m
ot
if
s
(
e
qua
l
num
be
r
of
im
a
ge
s
pe
r
c
a
te
gor
y)
.
E
a
c
h
im
a
ge
is
512×
512
pi
xe
ls
i
n
s
iz
e
. F
ig
ur
e
2 s
how
s
a
s
a
m
pl
e
of
ni
ti
k ba
ti
k m
ot
if
s
in
t
he
da
ta
s
e
t.
F
ig
ur
e
2. A
s
a
m
pl
e
of
ni
ti
k ba
ti
k
3.2. Au
gm
e
n
t
at
io
n
t
e
c
h
n
iq
u
e
I
n
ge
ne
r
a
l,
c
la
s
s
if
ic
a
ti
on
m
ode
l
in
c
lu
di
ng
th
e
c
la
s
s
if
ic
a
ti
on
of
ni
ti
k
ba
ti
k
,
a
ppl
yi
ng
va
r
io
us
da
ta
a
ugm
e
nt
a
ti
on
te
c
hni
que
s
is
e
s
s
e
nt
ia
l
f
or
e
nha
nc
in
g
m
ode
l’
s
pe
r
f
or
m
a
nc
e
.
T
he
s
e
m
e
th
od
s
in
tr
oduc
e
va
r
ia
bi
li
ty
to
th
e
da
ta
s
e
t,
im
pr
ovi
ng
th
e
m
ode
l’
s
ge
n
e
r
a
li
z
a
ti
on
c
a
pa
bi
li
ti
e
s
.
T
hi
s
a
ppr
oa
c
h
e
v
a
lu
a
te
s
w
hi
c
h
Evaluation Warning : The document was created with Spire.PDF for Python.
I
nt
J
A
r
ti
f
I
nt
e
ll
I
S
S
N
:
2252
-
8938
O
pt
imi
z
in
g ni
ti
k
bat
ik
c
la
s
s
if
ic
at
io
n t
hr
ough c
om
pa
r
at
iv
e
analy
s
is
of
i
m
age
augme
nt
at
io
n
(
Supr
apt
o
)
3973
a
ugm
e
nt
a
ti
on
te
c
hni
que
pr
ovi
de
s
th
e
m
os
t
e
f
f
e
c
ti
ve
r
e
s
ul
ts
f
or
a
c
c
ur
a
te
ly
c
la
s
s
if
yi
ng
ni
ti
k
ba
ti
k
m
ot
if
s
w
hi
le
pr
e
s
e
r
vi
ng t
he
e
s
s
e
nt
ia
l
pa
tt
e
r
ns
.
T
he
a
ugm
e
nt
a
ti
on
te
c
hni
que
s
w
e
r
e
s
e
le
c
te
d
ba
s
e
d
on
th
e
ir
a
bi
li
ty
to
a
ddr
e
s
s
th
e
c
ha
ll
e
nge
s
of
ni
ti
k
ba
ti
k
c
la
s
s
if
ic
a
ti
on,
in
c
lu
di
ng
hi
gh
in
tr
a
c
la
s
s
va
r
ia
bi
li
ty
,
c
om
pl
e
x
dot
-
ba
s
e
d
pa
tt
e
r
ns
,
a
nd
li
m
it
e
d
da
ta
s
e
t
s
iz
e
(
240
im
a
ge
s
)
.
G
e
om
e
tr
ic
tr
a
n
s
f
or
m
a
ti
on
s
(
f
li
ppi
ng
,
r
ot
a
ti
on, s
c
a
li
ng,
s
he
a
r
in
g,
a
nd
tr
a
ns
l
a
ti
on)
w
e
r
e
c
ho
s
e
n t
o
e
nha
n
c
e
m
od
e
l
r
obu
s
tn
e
s
s
t
o
po
s
it
io
na
l
a
nd
or
ie
nt
a
ti
o
na
l
va
r
ia
t
io
ns
,
a
s
th
e
s
e
a
r
e
c
om
m
on
in
r
e
a
l
-
w
or
l
d
b
a
ti
k
im
a
ge
s
[
14]
, [
18]
, [
1
9]
. I
nt
e
n
s
it
y t
r
a
ns
f
or
m
a
ti
on
s
(
c
ut
-
out
, gr
id
m
a
s
k
, hi
de
a
nd
s
e
e
k,
a
n
d
r
a
ndo
m
e
r
a
s
in
g)
w
e
r
e
s
e
le
c
te
d
to
s
im
ul
a
te
oc
c
lu
s
io
ns
a
nd
li
ght
in
g
va
r
ia
ti
ons
,
im
pr
ovi
ng
ge
ne
r
a
li
z
a
ti
on
to
im
pe
r
f
e
c
t
im
a
ge
s
[
15]
,
[
25
]
,
[
26]
.
C
ut
-
ou
t
a
nd
r
a
ndom
e
r
a
s
in
g
,
in
pa
r
ti
c
ul
a
r
,
in
t
r
oduc
e
lo
c
a
l
di
s
r
upt
io
ns
,
m
im
ic
ki
ng
r
e
a
l
-
w
or
ld
im
pe
r
f
e
c
ti
ons
w
hi
le
pr
e
s
e
r
vi
ng
ove
r
a
ll
m
ot
if
s
tr
uc
tu
r
e
.
P
a
ir
in
g
s
a
m
pl
e
s
w
a
s
in
c
lu
de
d
to
e
xp
lo
r
e
non
-
in
s
ta
nc
e
-
le
ve
l
a
ugm
e
nt
a
t
io
n,
c
om
bi
ni
ng i
m
a
ge
s
t
o
c
r
e
a
t
e
di
v
e
r
s
e
hybr
i
d
pa
tt
e
r
n
s
,
a
s
d
e
m
on
s
tr
a
te
d i
n
[
34]
. D
C
G
A
N
w
a
s
c
hos
e
n
to
ge
ne
r
a
te
s
ynt
he
ti
c
ni
ti
k
ba
ti
k
im
a
ge
s
,
a
ddr
e
s
s
in
g
da
ta
s
e
t
li
m
it
a
ti
ons
by
pr
oduc
in
g
hi
gh
-
qua
li
t
y
s
a
m
pl
e
s
th
a
t
c
a
pt
ur
e
in
tr
ic
a
te
te
xt
ur
e
s
[
16]
,
[
17]
.
T
he
s
e
m
e
th
ods
w
e
r
e
s
e
l
e
c
te
d
to
ba
la
nc
e
di
ve
r
s
it
y,
r
obus
tn
e
s
s
,
a
nd
pr
e
s
e
r
va
ti
on
of
ni
ti
k
ba
ti
k’
s
c
ul
tu
r
a
l
a
nd
vi
s
ua
l
c
ha
r
a
c
te
r
is
ti
c
s
.
T
a
bl
e
1
s
um
m
a
r
iz
e
s
th
e
a
ugm
e
nt
a
ti
on
te
c
hni
que
s
a
nd
th
e
ir
pa
r
a
m
e
te
r
s
,
w
it
h
a
de
ta
il
e
d
de
s
c
r
ip
ti
on
of
th
e
D
C
G
A
N
a
r
c
hi
te
c
tu
r
e
pr
ovi
de
d.
A
de
ta
il
e
d
de
s
c
r
ip
ti
on
of
th
e
D
C
G
A
N
a
r
c
hi
te
c
tu
r
e
is
pr
ovi
de
d
in
F
ig
ur
e
3,
il
lu
s
tr
a
ti
ng
th
e
la
ye
r
c
onf
ig
ur
a
ti
ons
, di
m
e
ns
io
ns
, a
nd da
ta
f
lo
w
f
or
bot
h t
he
ge
ne
r
a
to
r
a
nd dis
c
r
im
in
a
to
r
.
T
a
bl
e
1. A
ugm
e
nt
a
ti
on t
e
c
hni
qu
e
s
a
nd p
a
r
a
m
e
te
r
s
A
ugm
e
nt
a
t
i
on
T
e
c
hni
que
D
e
s
c
r
i
pt
i
on a
nd pa
r
a
m
e
t
e
r
s
G
e
om
e
t
r
i
c
t
r
a
ns
f
or
m
a
t
i
ons
[
35]
R
ot
a
t
i
on
R
a
ndom
r
ot
a
t
i
on w
i
t
hi
n [
-
30
°
, 30
°
]
, s
t
e
p s
i
z
e
of
10°
.
F
l
i
ppi
ng
H
or
i
z
ont
a
l
a
nd ve
r
t
i
c
a
l
f
l
i
ppi
ng w
i
t
h a
pr
oba
bi
l
i
t
y o
f
0.5.
S
c
a
l
i
ng
R
a
ndom
s
c
a
l
i
ng be
t
w
e
e
n 0.8
×
a
nd 1.2
×
of
or
i
gi
na
l
i
m
a
ge
s
i
z
e
.
S
he
a
r
i
ng
S
he
a
r
a
ngl
e
r
a
nge
of
[
-
15°
, 15°
]
.
T
r
a
ns
l
a
t
i
on
R
a
ndom
s
hi
f
t
s
a
l
ong X
a
nd Y
a
xe
s
w
i
t
hi
n [
-
10%
, 10%
]
of
i
m
a
ge
di
m
e
ns
i
ons
.
I
nt
e
ns
i
t
y
t
r
a
ns
f
or
m
a
t
i
ons
[
36]
C
ut
-
out
R
a
ndom
r
e
m
ova
l
of
1
–
3 s
qua
r
e
pa
t
c
he
s
(
50
×
50 pi
xe
l
s
)
pe
r
i
m
a
ge
.
G
r
i
d m
a
s
k
G
r
i
d of
10
×
10 pi
xe
l
bl
oc
ks
, 50%
pr
oba
bi
l
i
t
y of
m
a
s
ki
ng e
a
c
h bl
oc
k.
H
i
de
a
nd s
e
e
k
R
a
ndom
l
y hi
de
16
×
16 pi
xe
l
pa
t
c
he
s
,
c
ove
r
i
ng 20%
of
t
he
i
m
a
ge
.
R
a
ndom
e
r
a
s
i
ng da
t
a
E
r
a
s
e
r
e
c
t
a
ngul
a
r
r
e
gi
ons
(
10%
–
20%
of
i
m
a
ge
a
r
e
a
)
f
i
l
l
e
d w
i
t
h r
a
ndom
noi
s
e
.
N
on
i
ns
t
a
nc
e
l
e
v
e
l
a
ugm
e
nt
a
t
i
on
[
34]
P
a
i
r
i
ng s
a
m
pl
e
s
C
om
bi
ne
t
w
o i
m
a
ge
s
w
i
t
h a
bl
e
ndi
ng r
a
t
i
o of
0.5 (
e
qua
l
c
ont
r
i
but
i
on)
.
U
nc
ondi
t
i
ona
l
i
m
a
ge
ge
ne
r
a
t
i
on
[
37]
D
C
G
A
N
T
he
ge
ne
r
a
t
or
t
a
ke
s
a
100
-
di
m
e
ns
i
ona
l
noi
s
e
ve
c
t
or
a
nd
ups
a
m
pl
e
s
i
t
t
hr
ough
a
s
e
r
i
e
s
of
t
r
a
ns
po
s
e
d
c
onvol
ut
i
ona
l
l
a
ye
r
s
t
o
pr
oduc
e
512
×
512
gr
a
ys
c
a
l
e
i
m
a
ge
s
,
m
a
t
c
hi
ng
t
he
pr
e
pr
oc
e
s
s
e
d
ni
t
i
k
ba
t
i
k
da
t
a
s
e
t
.
T
he
di
s
c
r
i
m
i
na
t
or
e
va
l
ua
t
e
s
w
h
e
t
he
r
i
nput
i
m
a
ge
s
a
r
e
r
e
a
l
(
f
r
om
t
he
240
-
i
m
a
ge
da
t
a
s
e
t
)
or
s
ynt
he
t
i
c
.
T
he
m
ode
l
w
a
s
t
r
a
i
ne
d
f
or
200
e
poc
hs
w
i
t
h
a
ba
t
c
h
s
i
z
e
of
32,
us
i
ng
t
he
A
da
m
opt
i
m
i
z
e
r
(
l
e
a
r
ni
ng
r
a
t
e
0.0002,
β
1
=0.5,
β
2
=0.999)
a
nd
bi
na
r
y c
r
os
s
-
e
nt
r
opy l
os
s
.
F
ig
ur
e
3. D
C
G
A
N
a
r
c
hi
te
c
tu
r
e
f
or
ge
ne
r
a
ti
ng s
ynt
he
ti
c
ni
ti
k ba
ti
k i
m
a
ge
s
T
h
e
D
C
G
A
N
w
a
s
i
m
p
le
m
e
nt
e
d us
in
g
P
y
T
or
c
h,
f
ol
l
ow
in
g
f
r
a
m
e
w
or
k
pr
o
po
s
e
d by R
a
df
or
d
e
t
al
.
[
3
8]
.
T
r
a
in
in
g
w
a
s
c
ondu
c
te
d
on
th
e
or
ig
in
a
l
240
ni
ti
k
ba
ti
k
im
a
ge
s
,
ge
ne
r
a
ti
ng
720
s
ynt
h
e
ti
c
im
a
ge
s
.
T
he
s
e
im
a
ge
s
w
e
r
e
vi
s
ua
ll
y
in
s
p
e
c
te
d
to
e
ns
ur
e
th
e
y
pr
e
s
e
r
ve
d
th
e
c
ha
r
a
c
te
r
is
ti
c
dot
-
ba
s
e
d
ge
om
e
tr
ic
pa
tt
e
r
n
s
of
ni
ti
k
ba
ti
k
m
ot
if
s
.
T
he
a
r
c
hi
te
c
tu
r
e
w
a
s
tu
ne
d
to
c
a
pt
ur
e
th
e
in
tr
ic
a
te
te
xt
ur
e
s
of
ni
t
ik
ba
ti
k
,
w
it
h
f
il
te
r
s
iz
e
s
a
nd
la
ye
r
de
pt
hs
a
dj
us
te
d
to
ha
ndl
e
th
e
512×
512
r
e
s
ol
ut
io
n.
A
l
l
a
ugm
e
nt
a
ti
on
te
c
hni
que
s
w
e
r
e
im
pl
e
m
e
nt
e
d
Evaluation Warning : The document was created with Spire.PDF for Python.
I
S
S
N
:
2252
-
8938
I
nt
J
A
r
ti
f
I
nt
e
ll
, V
ol
.
14
, N
o.
5
,
O
c
to
be
r
20
25
:
3970
-
3981
3974
us
in
g
s
ta
nda
r
d
li
br
a
r
ie
s
.
P
a
r
a
m
e
te
r
s
w
e
r
e
s
e
le
c
te
d
ba
s
e
d
on
e
s
ta
bl
is
he
d
pr
a
c
ti
c
e
s
a
nd
tu
ne
d
to
ba
la
nc
e
di
ve
r
s
it
y
a
nd
pr
e
s
e
r
va
ti
on
of
ni
ti
k
ba
ti
k
p
a
tt
e
r
ns
,
e
n
s
ur
in
g
th
e
a
ugm
e
nt
e
d
da
ta
s
e
t
e
nh
a
nc
e
s
m
ode
l
ge
ne
r
a
li
z
a
ti
on w
hi
le
m
a
in
ta
in
in
g c
ul
tu
r
a
l
a
nd vis
ua
l
in
te
gr
it
y.
3.3. P
r
e
-
p
r
oc
e
s
s
in
g
T
he
pr
e
-
pr
oc
e
s
s
in
g
u
s
e
d
in
our
r
e
s
e
a
r
c
h
c
onve
r
ts
th
e
im
a
g
e
in
to
gr
a
ys
c
a
le
.
T
hi
s
i
s
due
to
th
e
r
e
la
ti
ve
ly
li
m
it
e
d
c
ol
or
va
r
ia
ti
ons
in
ni
ti
k
ba
ti
k
f
a
br
ic
s
.
B
y c
on
ve
r
ti
ng
th
e
im
a
ge
to
gr
a
ys
c
a
le
,
th
e
f
oc
us
s
hi
f
ts
f
r
om
c
ol
or
in
f
or
m
a
ti
on,
w
hi
c
h
is
not
c
r
it
ic
a
l
in
th
is
c
a
s
e
,
to
th
e
in
tr
ic
a
te
m
ot
if
s
a
nd
te
xt
ur
e
s
th
a
t
d
e
f
in
e
ni
ti
k
ba
ti
k
.
T
hi
s
s
im
pl
if
ic
a
ti
on
r
e
duc
e
s
da
ta
c
om
pl
e
xi
ty
a
nd
e
nha
nc
e
s
th
e
m
ode
l’
s
a
bi
li
ty
to
e
xt
r
a
c
t
m
e
a
ni
ngf
ul
f
e
a
tu
r
e
s
r
e
la
te
d
to
th
e
pa
tt
e
r
ns
a
nd
s
tr
uc
tu
r
e
s
in
th
e
ba
ti
k
[
39]
,
le
a
di
ng
to
m
or
e
e
f
f
ic
ie
nt
a
nd
a
c
c
ur
a
te
c
la
s
s
if
ic
a
ti
on r
e
s
ul
t
s
.
3.4. F
e
at
u
r
e
e
xt
r
ac
t
io
n
W
e
us
e
th
e
bi
na
r
iz
e
d
s
ta
ti
s
ti
c
a
l
im
a
ge
f
e
a
tu
r
e
s
(
B
S
I
F
)
f
or
f
e
a
tu
r
e
e
xt
r
a
c
ti
on
be
c
a
us
e
it
is
a
pr
a
c
ti
c
a
l
f
e
a
tu
r
e
e
xt
r
a
c
ti
on
m
e
th
od
to
c
a
pt
ur
e
ba
ti
k'
s
c
om
pl
e
x
a
nd
uni
que
te
xt
ur
e
s
a
nd
m
ot
if
s
.
B
S
I
F
us
e
s
f
il
te
r
s
le
a
r
ne
d
f
r
om
na
tu
r
a
l
im
a
ge
s
to
e
xt
r
a
c
t
lo
c
a
l
s
ta
ti
s
ti
c
a
l
f
e
a
tu
r
e
s
.
T
hi
s
m
a
ke
s
it
ve
r
y
u
s
e
f
ul
f
or
a
na
ly
z
in
g
c
om
pl
e
x pa
tt
e
r
ns
s
uc
h
a
s
t
hos
e
in
ni
ti
k ba
ti
k
.
I
m
a
ge
s
w
e
r
e
c
onve
r
te
d
to
gr
a
ys
c
a
l
e
to
f
oc
us
on
te
xt
ur
e
pa
tt
e
r
ns
,
a
s
ni
ti
k
ba
ti
k
m
ot
if
s
a
r
e
de
f
in
e
d
by
dot
-
ba
s
e
d
s
tr
uc
tu
r
e
s
r
a
th
e
r
th
a
n
c
ol
or
.
G
r
a
ys
c
a
le
im
a
ge
s
w
e
r
e
nor
m
a
li
z
e
d
to
[
0,
1]
f
or
c
ons
is
te
nt
f
e
a
tu
r
e
e
xt
r
a
c
ti
on.
B
S
I
F
f
il
te
r
s
,
in
s
pi
r
e
d
by
in
de
pe
nde
nt
c
om
pone
nt
a
na
ly
s
is
(
I
C
A
)
[
40]
,
w
e
r
e
a
ppl
ie
d
to
e
xt
r
a
c
t
te
xt
ur
e
f
e
a
tu
r
e
s
. T
he
s
e
pr
e
-
tr
a
in
e
d f
il
te
r
s
, de
r
iv
e
d f
r
om
na
tu
r
a
l
im
a
ge
s
, us
e
8
-
bi
t,
5
×
5 pa
tc
he
s
t
o c
a
pt
ur
e
l
oc
a
l
te
xt
ur
e
pa
tt
e
r
ns
,
id
e
a
l
f
or
e
nc
odi
ng
th
e
e
dg
e
-
li
ke
a
nd
ge
om
e
tr
ic
f
e
a
tu
r
e
s
of
ni
ti
k
b
a
ti
k
.
V
is
ua
l
in
s
pe
c
ti
on
c
onf
ir
m
e
d
th
a
t
B
S
I
F
pr
oc
e
s
s
in
g
pr
e
s
e
r
ve
d
c
r
it
ic
a
l
dot
-
ba
s
e
d
pa
tt
e
r
ns
,
a
nd
th
e
hi
gh
c
la
s
s
if
ic
a
ti
on
a
c
c
ur
a
c
y
va
li
da
te
s
th
e
da
ta
s
e
t’
s
s
ui
ta
bi
li
ty
.
T
he
da
t
a
s
e
t’
s
ba
la
n
c
e
d
s
t
r
uc
tu
r
e
(
16
im
a
ge
s
pe
r
c
a
te
gor
y)
a
nd
hi
gh
r
e
s
ol
ut
io
n
e
ns
ur
e
s
uf
f
ic
ie
nt
di
ve
r
s
it
y
a
nd
d
e
ta
il
,
m
a
ki
ng
it
a
p
pr
opr
ia
te
f
or
te
xt
ur
e
-
ba
s
e
d
f
e
a
tu
r
e
e
xt
r
a
c
ti
on
w
it
hout
l
os
in
g e
s
s
e
nt
ia
l
m
ot
if
c
ha
r
a
c
te
r
is
ti
c
s
.
T
he
f
or
m
ul
a
f
or
B
S
I
F
c
a
n be
de
s
c
r
ib
e
d a
s
f
ol
lo
w
s
:
i)
A
l
oc
a
l
ne
ig
hbor
hood pa
tc
h r
e
pr
e
s
e
nt
s
e
a
c
h pi
xe
l
in
a
gr
a
y
s
c
a
le
i
m
a
ge
.
ii)
A
s
e
t
of
l
e
a
r
ne
d f
il
te
r
s
1
,
2
,
⋯
,
is
a
ppl
ie
d t
o t
he
s
e
pa
tc
he
s
.
iii)
T
he
out
put
of
e
a
c
h f
il
te
r
i
s
bi
na
r
iz
e
d us
in
g (
1)
:
=
{
1
if
ℎ
≥
0
0
o
t
he
r
wis
e
(
1)
iv
)
T
he
bi
na
r
y r
e
s
pons
e
s
1
,
2
,
⋯
,
f
or
m
a
bi
na
r
y s
tr
in
g f
or
e
a
c
h pi
xe
l,
w
hi
c
h i
s
t
he
n us
e
d a
s
t
he
f
e
a
tu
r
e
de
s
c
r
ip
to
r
.
T
he
s
e
bi
na
r
y
c
ode
s
e
f
f
ic
ie
nt
ly
c
a
pt
ur
e
lo
c
a
l
te
xt
ur
e
s
,
w
hi
c
h
is
pa
r
ti
c
ul
a
r
ly
us
e
f
ul
in
r
e
c
ogni
z
in
g
th
e
c
om
pl
e
x
m
ot
if
s
i
n
ni
ti
k
ba
ti
k
c
la
s
s
if
ic
a
ti
on.
3.5. Clas
s
if
i
c
at
io
n
m
od
e
l
d
e
ve
lo
p
m
e
n
t
R
a
nd
om
f
or
e
s
t
i
s
a
n
e
f
f
e
c
t
iv
e
c
l
a
s
s
if
i
c
a
ti
on
m
od
e
l
f
or
ni
t
ik
b
a
ti
k
b
e
c
a
us
e
it
ha
ndl
e
s
c
om
pl
e
x,
hi
gh
-
di
m
e
ns
io
na
l
da
ta
s
u
c
h
a
s
t
e
x
tu
r
e
p
a
tt
e
r
n
s
.
T
he
m
od
e
l
w
o
r
ks
b
y
bu
il
d
in
g
m
ul
t
ip
l
e
d
e
c
i
s
io
n
tr
e
e
s
du
r
in
g
tr
a
i
ni
n
g,
e
a
c
h
a
na
l
y
z
in
g
di
f
f
e
r
e
n
t
f
e
a
t
ur
e
s
of
th
e
b
a
t
ik
m
ot
if
.
T
h
is
m
o
de
l
bu
il
d
s
m
ul
t
ip
l
e
d
e
c
i
s
io
n
tr
e
e
s
a
n
d
c
om
bi
n
e
s
th
e
ir
pr
e
di
c
ti
on
s
f
or
th
e
c
l
a
s
s
if
i
c
a
ti
o
n
of
n
it
i
k b
a
t
ik
.
T
h
e
c
r
it
ic
a
l
f
or
m
ul
a
us
e
d
i
n r
a
nd
om
f
or
e
s
t
is
[
4
1]
:
i)
G
in
i
im
pu
r
it
y or
e
nt
r
opy
f
or
c
la
s
s
if
ic
a
ti
on t
r
e
e
s
c
a
n be
s
e
e
n a
t
(
2)
a
nd (
3)
:
(
)
=
1
−
∑
2
=
1
(
2)
(
)
=
−
∑
2
(
)
=
1
(
3)
W
he
r
e
is
t
he
pr
opor
ti
on of
c
la
s
s
in
t
he
d
a
ta
s
e
t
.
ii)
P
r
e
di
c
ti
on:
t
he
f
in
a
l
pr
e
di
c
ti
on f
o
r
c
la
s
s
if
ic
a
ti
on i
s
t
he
m
a
jo
r
it
y vote
f
r
om
a
ll
t
r
e
e
s
.
R
a
ndom
f
or
e
s
t
is
w
e
ll
-
s
ui
te
d
to
ha
ndl
e
noi
s
e
a
nd
va
r
ia
nc
e
in
t
he
da
ta
.
I
t
is
a
gr
e
a
t
c
hoi
c
e
f
or
ni
ti
k
ba
ti
k,
w
he
r
e
c
la
s
s
if
ic
a
ti
on
de
pe
nd
s
on
de
ta
il
e
d
t
e
xt
ur
e
in
f
or
m
a
ti
on
r
a
th
e
r
th
a
n
pr
im
a
r
y
pi
xe
l
-
ba
s
e
d
da
ta
.
T
o
va
li
da
te
th
e
ni
ti
k
ba
ti
k
c
la
s
s
if
ic
a
ti
on
m
ode
l,
w
e
e
m
pl
oy
-
f
ol
d
c
r
os
s
-
va
li
da
ti
on
,
e
m
pha
s
iz
in
g
th
e
im
por
ta
nc
e
of
da
ta
di
ve
r
s
it
y a
nd r
e
li
a
bi
li
ty
. T
hi
s
a
ppr
oa
c
h di
vi
de
s
t
he
d
a
ta
s
e
t
in
to
=
4
f
ol
ds
of
e
qua
l
s
iz
e
. T
he
m
ode
l
i
s
tr
a
in
e
d on (
−
1
)
f
ol
ds
a
nd t
e
s
te
d on the
r
e
m
a
in
in
g
f
ol
ds
. T
hi
s
pr
oc
e
s
s
r
e
pe
a
ts
, e
ns
ur
in
g t
ha
t
e
ve
r
y s
a
m
pl
e
is
us
e
d
f
or
tr
a
in
in
g
a
nd
va
li
da
ti
on.
B
y
a
ve
r
a
gi
ng
th
e
r
e
s
ul
ts
,
th
is
m
e
th
od
r
e
duc
e
s
bi
a
s
a
nd
va
r
ia
nc
e
,
th
us
pr
ovi
di
ng a
r
obus
t
e
va
lu
a
ti
on of
t
he
m
ode
l'
s
pe
r
f
or
m
a
nc
e
on dif
f
e
r
e
nt
s
ubs
e
ts
of
da
ta
.
Evaluation Warning : The document was created with Spire.PDF for Python.
I
nt
J
A
r
ti
f
I
nt
e
ll
I
S
S
N
:
2252
-
8938
O
pt
imi
z
in
g ni
ti
k
bat
ik
c
la
s
s
if
ic
at
io
n t
hr
ough c
om
pa
r
at
iv
e
analy
s
is
of
i
m
age
augme
nt
at
io
n
(
Supr
apt
o
)
3975
3.6. E
val
u
at
io
n
O
nc
e
th
e
c
la
s
s
if
ic
a
ti
on
m
ode
l
h
a
s
be
e
n
de
ve
lo
pe
d,
it
w
a
s
e
v
a
lu
a
te
d
f
or
a
s
s
e
s
s
in
g
it
s
a
c
c
ur
a
c
y
to
de
te
r
m
in
e
how
w
e
ll
th
e
m
ode
l
c
a
n
c
la
s
s
if
y
di
f
f
e
r
e
nt
ba
ti
k
m
ot
if
s
.
A
n
a
ddi
ti
ona
l
i
m
por
ta
nt
m
e
t
r
ic
is
th
e
s
ta
nda
r
d
de
vi
a
ti
on
of
th
e
a
c
c
ur
a
c
y
a
c
r
os
s
m
ul
ti
pl
e
c
r
os
s
-
va
li
da
ti
on
f
ol
ds
,
a
s
it
m
e
a
s
ur
e
s
th
e
s
ta
bi
li
ty
of
th
e
m
ode
l.
A
lo
w
e
r
s
ta
nd
a
r
d
de
vi
a
ti
on
in
di
c
a
te
s
c
on
s
is
te
nt
p
e
r
f
or
m
a
nc
e
a
c
r
os
s
di
f
f
e
r
e
nt
da
ta
s
ubs
e
ts
,
im
pl
yi
ng
th
a
t
th
e
m
ode
l
ge
ne
r
a
li
z
e
s
w
e
ll
.
T
hi
s
c
om
bi
na
ti
on
of
m
e
tr
ic
s
e
ns
ur
e
s
th
a
t
th
e
c
la
s
s
if
ic
a
ti
on
m
ode
l
is
a
c
c
ur
a
te
a
nd r
e
li
a
bl
e
f
or
r
e
a
l
-
w
or
ld
a
ppl
ic
a
ti
on.
4.
R
E
S
U
L
T
S
A
N
D
D
I
S
C
U
S
S
I
O
N
A
f
te
r
a
ppl
yi
ng
da
ta
a
ugm
e
nt
a
ti
on
te
c
hni
que
s
on
ni
ti
k
ba
ti
k
i
m
a
ge
s
us
in
g
ge
om
e
tr
ic
tr
a
ns
f
or
m
a
ti
on,
in
te
ns
it
y
tr
a
ns
f
or
m
a
ti
on,
non
-
in
s
ta
nc
e
le
ve
l,
a
nd
unc
ondi
ti
ona
l
im
a
ge
ge
ne
r
a
ti
on.
F
ig
ur
e
4
il
lu
s
tr
a
te
s
th
e
pr
e
pr
oc
e
s
s
in
g
out
c
om
e
s
f
or
r
e
pr
e
s
e
nt
a
ti
ve
ni
ti
k
ba
ti
k
im
a
g
e
s
.
F
ig
ur
e
4(
a
)
s
how
s
a
n
or
ig
in
a
l
R
G
B
im
a
ge
of
th
e
‘
S
e
ka
r
c
e
ngke
h’
m
ot
if
,
F
ig
ur
e
s
4(
b)
to
4(
f
)
di
s
pl
a
ys
it
s
ge
om
e
tr
ic
tr
a
ns
f
or
m
a
ti
on
a
ugm
e
nt
a
ti
on
,
F
ig
ur
e
s
4(
g)
to
4(
j)
pr
e
s
e
nt
th
e
in
te
ns
it
y
tr
a
ns
f
or
m
a
ti
on
,
F
ig
ur
e
4(
k)
de
pi
c
t
th
e
non
-
in
s
ta
nc
e
l
e
ve
l
a
ugm
e
nt
a
ti
on
,
a
nd F
ig
ur
e
4(
l)
s
how
s
unc
ondi
ti
ona
l
im
a
ge
ge
ne
r
a
ti
on a
ugm
e
nt
a
ti
on.
T
he
r
e
s
ul
t
s
of
f
li
ppi
ng
in
F
ig
ur
e
4(
b)
a
nd
r
ot
a
ti
ng
in
F
ig
ur
e
4(
c
)
a
ugm
e
nt
a
ti
ons
r
e
s
e
m
bl
e
th
e
or
ig
in
a
l.
B
ot
h
te
c
hni
que
s
a
ppe
a
r
to
r
e
ta
in
th
e
ove
r
a
ll
pa
tt
e
r
n
o
f
th
e
m
ot
i
f
,
onl
y
c
ha
ngi
ng
th
e
or
ie
nt
a
ti
on
a
n
d
di
r
e
c
ti
on.
T
he
s
e
tr
a
ns
f
or
m
a
ti
ons
m
a
in
ta
in
th
e
in
te
gr
it
y
of
th
e
or
ig
in
a
l
ni
ti
k
ba
ti
k
de
s
ig
n.
T
he
m
ot
if
s
th
a
t
c
ha
nge
th
e
m
os
t
f
r
om
th
e
or
ig
in
a
l
a
r
e
th
e
c
ut
-
out
in
F
ig
ur
e
4(
g)
a
nd
r
a
ndom
e
r
a
s
in
g
da
ta
in
F
ig
ur
e
4(
j)
a
ugm
e
nt
a
ti
ons
.
T
he
s
e
a
ugm
e
nt
a
ti
ons
dr
a
s
ti
c
a
ll
y
m
odi
f
y
th
e
pa
tt
e
r
n
by
r
e
m
ovi
ng
pa
r
ts
of
th
e
m
ot
i
f
,
th
us
c
ha
ngi
ng
it
s
a
ppe
a
r
a
nc
e
s
ig
ni
f
ic
a
nt
ly
.
P
a
ir
in
g
s
a
m
pl
e
s
in
F
ig
ur
e
4(
k)
a
ls
o
dr
a
m
a
ti
c
a
ll
y
a
lt
e
r
s
th
e
im
a
ge
by
c
om
bi
ni
ng dif
f
e
r
e
nt
m
ot
if
s
, c
r
e
a
ti
ng a
hybr
id
pa
tt
e
r
n.
(
a
)
(
b)
(
c
)
(
d)
(
e
)
(f)
(
g)
(
h)
(
i)
(
j)
(
k)
(
l)
F
ig
ur
e
4.
S
a
m
pl
e
r
e
s
ul
t
of
da
ta
a
ugm
e
nt
a
ti
on on the
s
e
ka
r
c
e
ng
ke
h m
ot
if
:
(
a
)
n
o a
ugm
e
nt
a
ti
on, (
b)
f
li
p,
(
c
)
r
ot
a
te
, (
d)
s
c
a
li
ng, (
e
)
s
he
a
r
in
g, (
f
)
t
r
a
ns
la
ti
on, (
g)
c
ut
-
out
, (
h)
gr
id
m
a
s
k, (
i)
hi
de
a
nd s
e
e
k,
(
j)
r
a
ndom e
r
a
s
in
g da
ta
, (
k)
pa
ir
in
g s
a
m
pl
e
s
, a
nd (
l)
D
C
G
A
N
B
S
I
F
e
xt
r
a
c
ts
te
xt
ur
e
f
e
a
tu
r
e
s
by
a
ppl
yi
ng
le
a
r
ne
d
f
il
te
r
s
t
o
lo
c
a
l
im
a
ge
pa
tc
he
s
.
W
e
us
e
th
e
pa
r
a
m
e
te
r
s
f
il
te
r
s
iz
e
15×
15
a
nd
f
il
te
r
bi
t
10×
10
to
s
pe
c
if
y
th
e
te
xt
ur
e
e
xt
r
a
c
ti
on
in
de
ta
il
.
T
h
e
e
xt
r
a
c
te
d
f
e
a
tu
r
e
s
a
r
e
s
to
r
e
d i
n a
f
ea
tu
r
e
a
r
r
a
y a
nd s
a
ve
d t
o a
C
S
V
f
il
e
. F
ig
ur
e
5 s
how
s
t
he
bi
na
r
y f
e
a
tu
r
e
di
s
tr
ib
ut
io
n t
o
he
lp
a
s
s
e
s
s
th
e
s
pr
e
a
d
a
nd
in
te
ns
it
y
of
th
e
f
e
a
tu
r
e
s
in
th
e
f
o
r
m
of
a
hi
s
to
gr
a
m
.
F
ig
ur
e
5
(
a
)
s
how
s
th
e
bi
na
r
y
f
e
a
tu
r
e
di
s
tr
ib
ut
io
n
f
or
o
r
ig
in
a
l
im
a
ge
,
F
ig
u
r
e
s
5(
b)
to
5(
f
)
di
s
pl
a
y
s
how
s
th
e
bi
na
r
y
f
e
a
tu
r
e
di
s
tr
ib
ut
io
n
f
or
ge
om
e
tr
ic
t
r
a
ns
f
or
m
a
ti
on
a
ugm
e
nt
a
ti
on, F
ig
ur
e
s
5(
g)
t
o 5(
j
)
de
pi
c
t
th
e
bi
na
r
y f
e
a
tu
r
e
di
s
tr
ib
ut
io
n f
or
in
te
ns
it
y
tr
a
ns
f
or
m
a
ti
on
a
ugm
e
nt
a
ti
on
,
F
ig
ur
e
5(
k)
pr
e
s
e
nt
th
e
bi
na
r
y
f
e
a
tu
r
e
di
s
tr
ib
ut
io
n
f
or
non
-
in
s
ta
nc
e
le
ve
l,
a
nd
F
ig
ur
e
5(
l)
s
how
t
he
bi
na
r
y f
e
a
tu
r
e
di
s
tr
ib
ut
io
n f
o
r
unc
ondi
ti
on
a
l
im
a
ge
ge
ne
r
a
ti
on
.
F
ig
ur
e
5
pr
e
s
e
nt
s
th
e
pl
ot
a
s
a
s
ta
c
ke
d
hi
s
to
gr
a
m
r
e
pr
e
s
e
nt
in
g
th
e
di
s
tr
ib
ut
io
n
of
pi
xe
l
in
te
ns
it
y
va
lu
e
s
or
f
e
a
tu
r
e
va
lu
e
s
e
xt
r
a
c
te
d
f
r
om
s
e
ve
r
a
l
im
a
ge
s
.
O
ve
r
la
id
c
ol
or
s
in
di
c
a
t
e
di
f
f
e
r
e
nt
ba
tc
he
s
of
pr
oc
e
s
s
e
d
im
a
ge
s
.
T
he
be
ll
-
s
ha
pe
d
pa
tt
e
r
n
in
di
c
a
te
s
th
a
t
m
o
s
t
va
lu
e
s
c
lu
s
te
r
a
r
ound
th
e
c
e
nt
e
r
,
r
e
f
le
c
ti
ng
a
ba
la
nc
e
d
in
te
ns
it
y
di
s
tr
ib
ut
io
n
or
c
onc
e
nt
r
a
ti
on
of
f
e
a
tu
r
e
s
a
r
ound
a
pa
r
ti
c
ul
a
r
c
e
nt
r
a
l
va
lu
e
.
T
he
x
-
a
xi
s
r
e
pr
e
s
e
nt
s
th
e
r
a
nge
of
in
te
ns
it
y
va
lu
e
s
th
a
t
r
e
s
ul
t
f
r
om
th
e
c
onvolut
io
n
of
th
e
im
a
ge
w
it
h
B
S
I
F
f
il
te
r
s
(
f
r
om
-
4
to
4)
,
in
di
c
a
ti
ng
how
w
e
ll
th
e
te
xt
u
r
e
in
th
e
i
m
a
ge
a
li
gns
w
it
h
th
e
le
a
r
ne
d
f
il
te
r
s
.
T
he
va
lu
e
s
gi
ve
in
s
ig
ht
s
in
to
th
e
s
tr
e
ngt
h
of
th
e
te
xt
ur
e
pa
tt
e
r
ns
a
nd
th
e
i
r
a
li
gnm
e
nt
w
it
h
th
e
f
il
te
r
s
.
T
he
y
-
a
xi
s
on
th
e
hi
s
to
gr
a
m
s
r
e
pr
e
s
e
nt
s
t
he
f
r
e
que
nc
y or
c
ount
of
pi
xe
l
in
te
ns
it
y
va
lu
e
s
or
f
e
a
tu
r
e
r
e
s
pons
e
s
e
xt
r
a
c
te
d u
s
in
g t
he
Evaluation Warning : The document was created with Spire.PDF for Python.
I
S
S
N
:
2252
-
8938
I
nt
J
A
r
ti
f
I
nt
e
ll
, V
ol
.
14
, N
o.
5
,
O
c
to
be
r
20
25
:
3970
-
3981
3976
B
S
I
F
m
e
th
od.
T
he
di
f
f
e
r
e
nc
e
s
in
y
-
a
xi
s
v
a
lu
e
s
a
c
r
os
s
th
e
h
is
to
gr
a
m
s
oc
c
ur
b
e
c
a
u
s
e
e
a
c
h
a
ugm
e
nt
a
ti
on
te
c
hni
que
a
lt
e
r
s
t
he
di
s
tr
ib
ut
io
n a
nd numbe
r
of
f
e
a
tu
r
e
s
or
pi
xe
l
va
lu
e
s
t
he
B
S
I
F
f
il
te
r
c
a
pt
ur
e
s
.
(
a
)
(
b)
(
c
)
(
d
)
(
e
)
(f)
(
g)
(
h)
(
i)
(
j)
(
k)
(
l)
F
ig
ur
e
5. T
he
bi
na
r
y f
e
a
tu
r
e
di
s
tr
ib
ut
io
n us
in
g B
S
I
F
:
(
a
)
no a
ug
m
e
nt
a
ti
on, (
b)
f
li
p, (
c
)
r
ot
a
te
, (
d)
s
c
a
li
ng,
(
e
)
s
he
a
r
in
g, (
f
)
t
r
a
ns
la
ti
on, (
g)
c
ut
-
out
, (
h)
g
r
id
m
a
s
k, (
i)
hi
de
a
nd s
e
e
k, (
j)
r
a
ndom e
r
a
s
in
g da
ta
, (
k)
pa
ir
in
g
s
a
m
pl
e
s
, a
nd (
l)
D
C
G
A
N
E
a
c
h
a
ugm
e
nt
a
ti
on
m
e
th
od
im
pa
c
ts
th
e
f
e
a
tu
r
e
e
xt
r
a
c
ti
on
p
r
oc
e
s
s
di
f
f
e
r
e
nt
ly
.
T
he
B
S
I
F
f
e
a
tu
r
e
s
e
xt
r
a
c
te
d
a
f
te
r
f
li
ppi
ng
in
F
ig
ur
e
5(
b)
,
th
e
im
a
ge
m
a
y
a
ppe
a
r
q
ui
te
s
im
il
a
r
to
th
e
or
ig
in
a
l
(
no
a
ugm
e
nt
a
ti
on)
,
a
s
f
li
ppi
ng
doe
s
not
s
ig
ni
f
ic
a
nt
ly
a
lt
e
r
th
e
lo
c
a
l
te
xt
ur
e
but
c
ha
nge
s
th
e
or
ie
nt
a
ti
on.
T
hu
s
,
th
e
e
xt
r
a
c
t
e
d
te
xt
ur
e
f
e
a
tu
r
e
s
r
e
m
a
in
c
ons
i
s
te
nt
.
R
ot
a
t
io
n
,
s
c
a
li
ng,
a
nd s
he
a
r
i
ng
in
F
ig
ur
e
s
5(
c
)
to
5(
e
)
in
tr
oduc
e
not
ic
e
a
bl
e
c
ha
nge
s
to
th
e
f
e
a
tu
r
e
s
due
to
a
lt
e
r
a
ti
ons
in
or
ie
nt
a
ti
on
a
nd
s
i
z
e
.
C
ut
-
out
,
gr
id
m
a
s
k,
a
nd
hi
de
a
nd
s
e
e
k
in
F
ig
ur
e
s
5(
g)
to
5(
i)
le
a
d
to
s
pa
r
s
e
or
s
e
gm
e
nt
e
d
f
e
a
tu
r
e
di
s
tr
ib
u
ti
ons
due
to
o
c
c
lu
s
io
n.
D
C
G
A
N
in
F
ig
ur
e
5(
l)
Evaluation Warning : The document was created with Spire.PDF for Python.
I
nt
J
A
r
ti
f
I
nt
e
ll
I
S
S
N
:
2252
-
8938
O
pt
imi
z
in
g ni
ti
k
bat
ik
c
la
s
s
if
ic
at
io
n t
hr
ough c
om
pa
r
at
iv
e
analy
s
is
of
i
m
age
augme
nt
at
io
n
(
Supr
apt
o
)
3977
s
how
s
th
e
im
pa
c
t
of
s
ynt
he
ti
c
da
ta
ge
ne
r
a
ti
on
on
f
e
a
tu
r
e
r
e
pr
e
s
e
nt
a
ti
on.
E
a
c
h
a
ugm
e
nt
a
ti
on
m
e
th
od
a
f
f
e
c
ts
th
e
s
ta
bi
li
ty
a
nd
c
ons
is
t
e
nc
y
of
th
e
e
xt
r
a
c
te
d
B
S
I
F
f
e
a
tu
r
e
s
,
w
hi
c
h
di
r
e
c
tl
y
in
f
lu
e
nc
e
s
th
e
c
la
s
s
if
ic
a
ti
on pe
r
f
or
m
a
nc
e
.
A
f
te
r
e
xt
r
a
c
ti
ng f
e
a
tu
r
e
s
f
r
om
e
a
c
h
a
ugm
e
nt
e
d
ni
ti
k ba
ti
k
im
a
g
e
us
in
g B
S
I
F
, w
e
e
v
a
lu
a
te
d t
h
e
i
m
pa
c
t
of
va
r
i
ous
a
ugm
e
nt
a
ti
on
t
e
c
hni
qu
e
s
on
th
e
a
c
c
ur
a
c
y
a
nd
s
ta
bi
li
ty
of
a
r
a
ndom
f
or
e
s
t
-
b
a
s
e
d
c
la
s
s
if
ic
a
ti
on
m
ode
l.
T
a
bl
e
2
c
om
p
a
r
e
s
im
a
g
e
a
ugm
e
nt
a
ti
o
n
t
e
c
hni
que
s
a
pp
li
e
d
to
th
e
ni
ti
k
b
a
ti
k
c
la
s
s
if
ic
a
ti
on
m
od
e
l
.
I
t
f
oc
us
es
on
a
c
c
ur
a
c
y a
nd
s
ta
bi
li
ty
(
s
ta
n
da
r
d d
e
vi
a
ti
on)
a
c
r
o
s
s
f
ou
r
c
r
os
s
-
va
l
id
a
ti
on f
ol
d
s
.
T
a
bl
e
2. C
om
pa
r
is
on of
i
m
a
ge
a
ugm
e
nt
a
ti
on t
e
c
hni
que
s
a
ppl
ie
d t
o
ni
ti
k ba
ti
k
c
la
s
s
if
ic
a
ti
on
A
ugm
e
nt
a
t
i
on
C
r
os
s
va
l
i
da
t
i
on (
%
)
A
ve
r
a
ge
a
c
c
ur
a
c
y (
%
)
S
t
a
nda
r
d
de
vi
a
t
i
on (
%
)
1
2
3
4
N
o a
ugm
e
nt
a
t
i
on
80
.00
70
.00
85
.00
75
.00
77.5
0
5
.
59
G
e
om
e
t
r
i
c
t
r
a
ns
f
or
m
a
t
i
on
F
l
i
p
9.5
0
99
.
16
99
.
58
99
.
58
97
.
70
3
.
01
R
ot
a
t
e
41
.
25
43
.
75
42
.
91
41
.
66
42
.
39
0
.
99
S
c
a
l
i
ng
69.16
57
.
91
62
.
91
65
.
41
63
.
85
4
.
08
S
ha
r
i
ng
65
.00
35
.00
41
.
66
42
.
5
0
46
.
04
11
.
32
T
r
a
ns
l
a
t
i
on
74
.
58
61
.
66
70
.
41
71
.
66
69
.
58
4
.
81
I
nt
e
ns
i
t
y t
r
a
ns
f
or
m
a
t
i
on
C
ut
-
out
99
.
58
95
.
83
95
.
41
96
.
25
96
.
77
1
.
65
G
r
i
d m
a
s
k
92
.
5
0
70
.
41
71
.
25
66
.
25
75
.
10
10
.
22
H
i
de
a
nd s
e
e
k
37
.
91
14
.
58
14
.
16
13
.
75
20
.
10
10
.
28
R
a
ndom
e
r
a
s
i
ng da
t
a
94
.
16
89
.
58
89
.
16
83
.
75
89
.
16
3
.
69
N
on
-
i
ns
t
a
nc
e
l
e
ve
l
P
a
i
r
i
ng s
a
m
pl
e
s
3
.
33
11
.
66
13
.
75
9
.
16
9
.
47
3
.
90
U
nc
ondi
t
i
ona
l
i
m
a
ge
ge
ne
r
a
t
i
on
D
C
G
A
N
93
.
75
92
.
5
0
94
.
58
94
.
16
93
.
75
0
.
78
T
he
e
va
lu
a
ti
on
of
im
a
ge
a
ugm
e
nt
a
ti
on
te
c
hni
que
s
on
th
e
ni
ti
k
ba
ti
k
960
da
ta
s
e
t
r
e
ve
a
le
d
s
ig
ni
f
ic
a
nt
va
r
ia
ti
ons
in
c
la
s
s
if
ic
a
ti
on
pe
r
f
or
m
a
nc
e
us
in
g
a
r
a
ndom
f
o
r
e
s
t
m
ode
l
,
a
s
s
how
n
in
T
a
bl
e
2.
W
it
hout
a
ugm
e
nt
a
ti
on,
th
e
m
ode
l
a
c
hi
e
v
e
d
a
n
a
c
c
ur
a
c
y
of
77.50%
w
i
th
a
s
ta
nda
r
d
de
vi
a
ti
on
of
5.59%
,
in
di
c
a
ti
ng
r
e
a
s
ona
bl
e
pe
r
f
or
m
a
nc
e
but
hi
ghe
r
va
r
ia
nc
e
a
c
r
os
s
f
ol
ds
.
I
n
c
ont
r
a
s
t,
a
ugm
e
nt
a
ti
on
m
e
th
ods
li
ke
f
li
ppi
ng
(
97.70%
a
c
c
ur
a
c
y,
3.01%
s
ta
nda
r
d
de
vi
a
ti
on)
,
c
ut
-
out
(
96.
77%
a
c
c
ur
a
c
y,
1.65%
s
ta
nda
r
d
de
vi
a
ti
on)
,
a
nd
D
C
G
A
N
(
93.75%
a
c
c
ur
a
c
y,
0.78%
s
ta
nda
r
d
de
vi
a
ti
on)
s
ubs
ta
nt
ia
ll
y
im
pr
ove
d
bot
h
a
c
c
ur
a
c
y
a
nd
s
ta
bi
li
ty
.
T
he
s
e
r
e
s
ul
ts
und
e
r
s
c
or
e
th
e
im
por
ta
nc
e
of
a
ugm
e
nt
a
ti
on
in
e
nha
nc
in
g
m
ode
l
pe
r
f
or
m
a
nc
e
on
th
e
li
m
it
e
d a
nd
c
om
pl
e
x ni
ti
k ba
ti
k
da
ta
s
e
t.
T
he
s
upe
r
io
r
pe
r
f
or
m
a
nc
e
of
f
li
ppi
ng,
c
ut
-
ou
t
,
a
nd
D
C
G
A
N
c
a
n
be
a
tt
r
ib
ut
e
d
to
th
e
ir
a
bi
li
ty
to
ba
la
nc
e
d
a
ta
s
e
t
di
ve
r
s
it
y
w
it
h
th
e
pr
e
s
e
r
va
ti
on
of
ni
ti
k
ba
ti
k’
s
i
nt
r
ic
a
te
dot
-
ba
s
e
d
p
a
tt
e
r
ns
.
F
li
ppi
ng
m
a
in
ta
in
s
ge
om
e
tr
ic
m
ot
if
in
te
gr
it
y
by
a
lt
e
r
in
g
or
ie
nt
a
ti
on
w
it
hout
di
s
r
upt
in
g
te
xt
ur
e
,
e
na
bl
in
g
c
ons
i
s
te
nt
f
e
a
tu
r
e
e
xt
r
a
c
ti
on
vi
a
B
S
I
F
.
C
ut
-
out
in
tr
oduc
e
s
c
ont
r
ol
le
d
oc
c
lu
s
io
n
s
, s
im
ul
a
ti
ng
r
e
a
l
-
w
or
ld
im
pe
r
f
e
c
ti
ons
li
ke
f
a
br
ic
w
e
a
r
w
hi
le
r
e
ta
in
in
g
s
uf
f
ic
ie
nt
pa
tt
e
r
n
in
f
o
r
m
a
ti
on
f
or
r
obu
s
t
c
la
s
s
if
ic
a
ti
on.
D
C
G
A
N
,
le
ve
r
a
gi
ng
C
N
N
a
r
c
hi
te
c
tu
r
e
,
ge
ne
r
a
te
s
hi
gh
-
qua
li
ty
s
ynt
he
ti
c
im
a
ge
s
th
a
t
c
lo
s
e
ly
m
im
ic
ni
ti
k
m
ot
if
s
,
a
s
c
onf
ir
m
e
d
by
vi
s
ua
l
in
s
pe
c
ti
on,
pr
ovi
di
ng
di
ve
r
s
e
tr
a
in
in
g
s
a
m
pl
e
s
w
it
h
m
in
im
a
l
va
r
ia
nc
e
(
lo
w
e
s
t
s
ta
nda
r
d
de
vi
a
ti
on
of
0.78%
)
.
C
onve
r
s
e
ly
,
m
e
th
ods
li
ke
pa
ir
in
g
s
a
m
pl
e
s
(
9.47%
a
c
c
ur
a
c
y)
,
hi
de
a
nd
s
e
e
k
(
20.10%
a
c
c
ur
a
c
y)
,
a
nd
gr
id
m
a
s
k
(
75.10%
a
c
c
ur
a
c
y)
unde
r
pe
r
f
or
m
e
d
due
to
e
xc
e
s
s
iv
e
di
s
r
upt
io
n
of
te
xt
u
r
e
c
ohe
r
e
nc
e
.
P
a
ir
in
g
s
a
m
pl
e
s
c
r
e
a
te
s
hybr
id
p
a
tt
e
r
ns
by
c
om
bi
ni
ng
unr
e
la
te
d
m
ot
if
s
,
c
onf
us
i
ng
th
e
c
la
s
s
if
ie
r
,
w
hi
le
hi
de
a
nd
s
e
e
k
a
nd
gr
id
m
a
s
k
in
tr
oduc
e
o
c
c
lu
s
io
n
s
th
a
t
ob
s
c
ur
e
c
r
it
ic
a
l
dot
pa
tt
e
r
n
s
,
hi
ghl
ig
ht
in
g
th
e
ne
e
d
f
or
a
ugm
e
nt
a
ti
on
m
e
th
ods
th
a
t
pr
e
s
e
r
ve
ni
ti
k ba
ti
k’
s
uni
que
te
xt
ur
e
s
.
R
a
ndom
f
or
e
s
t
w
a
s
s
e
l
e
c
te
d
a
s
th
e
c
l
a
s
s
if
ie
r
due
to
it
s
a
li
g
nm
e
nt
w
it
h
th
e
s
tu
dy’
s
obj
e
c
ti
ve
of
e
va
lu
a
ti
ng
a
ugm
e
nt
a
ti
on
e
f
f
e
c
ts
on
a
s
m
a
ll
,
hi
gh
-
di
m
e
ns
io
n
a
l
da
ta
s
e
t.
I
ts
r
obus
tn
e
s
s
to
B
S
I
F
-
e
xt
r
a
c
te
d
te
xt
ur
e
f
e
a
tu
r
e
s
a
nd e
f
f
e
c
ti
ve
ne
s
s
on
th
e
ni
ti
k ba
ti
k
960
da
ta
s
e
t
(
960 im
a
ge
s
a
c
r
os
s
60 mot
if
c
a
te
gor
ie
s
)
m
a
ke
it
s
ui
ta
bl
e
f
or
c
a
pt
ur
in
g
in
t
r
ic
a
te
dot
-
ba
s
e
d
pa
tt
e
r
ns
,
a
s
e
vi
de
n
c
e
d
by
th
e
hi
gh
a
c
c
ur
a
c
ie
s
w
it
h
f
li
ppi
ng
a
nd
c
ut
-
ou
t.
T
he
in
te
gr
a
ti
on
of
C
N
N
a
r
c
hi
te
c
tu
r
e
w
it
hi
n
D
C
G
A
N
,
w
hi
c
h
a
c
hi
e
ve
d
s
tr
ong
a
c
c
ur
a
c
y
(
93.75%
)
a
nd
th
e
hi
ghe
s
t
s
ta
bi
li
ty
,
f
ur
th
e
r
s
uppor
ts
r
a
ndom
f
or
e
s
t’
s
r
e
pr
e
s
e
nt
a
ti
ve
ne
s
s
.
B
y
le
ve
r
a
gi
ng
C
N
N
-
ge
ne
r
a
te
d
s
ynt
he
ti
c
im
a
ge
s
,
r
a
ndom
f
or
e
s
t
be
ne
f
it
s
f
r
om
e
nha
nc
e
d
da
ta
s
e
t
di
ve
r
s
it
y
w
it
hout
r
e
qui
r
in
g
a
C
N
N
-
ba
s
e
d
c
la
s
s
if
ie
r
,
m
a
in
ta
in
in
g
th
e
s
tu
dy’
s
f
oc
us
on
a
ugm
e
nt
a
ti
on
i
m
pa
c
ts
.
C
om
pa
r
in
g
r
a
ndom
f
or
e
s
t
to
ot
he
r
c
la
s
s
if
ie
r
s
li
ke
C
N
N
s
w
oul
d
ne
c
e
s
s
it
a
te
e
xt
e
ns
iv
e
e
xp
e
r
im
e
nt
s
a
nd
s
hi
f
t
a
tt
e
nt
io
n
a
w
a
y
f
r
om
a
ugm
e
nt
a
ti
on
e
f
f
e
c
ts
, w
hi
c
h i
s
be
yond thi
s
s
tu
dy’
s
s
c
ope
.
D
e
s
pi
te
i
ts
s
tr
e
ng
th
s
,
r
a
ndo
m
f
or
e
s
t
ha
s
l
im
i
ta
t
io
ns
.
I
ts
r
e
li
a
nc
e
on
h
a
ndc
r
a
f
te
d
B
S
I
F
f
e
a
tu
r
e
s
m
a
y
no
t
c
a
pt
ur
e
m
ul
ti
-
le
ve
l
hi
e
r
a
r
c
hi
c
a
l
pa
t
te
r
ns
a
s
e
f
f
e
c
t
iv
e
ly
a
s
de
e
p
le
a
r
n
in
g
m
ode
ls
li
ke
C
N
N
s
,
w
hi
c
h
le
a
r
n
f
e
a
tu
r
e
s
d
ir
e
c
t
ly
f
r
om
r
a
w
i
m
a
g
e
s
.
A
d
di
ti
ona
ll
y,
it
s
pe
r
f
o
r
m
a
nc
e
is
s
e
ns
i
ti
ve
to
a
ug
m
e
n
ta
t
io
n
qua
l
it
y,
a
s
s
e
e
n
i
n
t
he
l
ow
a
c
c
ur
a
c
ie
s
of
pa
i
r
in
g
s
a
m
p
le
s
a
nd
h
id
e
a
nd
s
e
e
k,
w
hi
c
h
d
is
r
up
te
d
m
ot
i
f
c
ohe
r
e
n
c
e
.
N
e
ve
r
th
e
le
s
s
,
r
a
nd
om
f
o
r
e
s
t
’
s
s
t
r
o
ng
pe
r
f
o
r
m
a
n
c
e
,
in
t
e
r
pr
e
ta
bi
li
ty
,
a
n
d
a
li
gn
m
e
n
t
w
it
h
t
he
Evaluation Warning : The document was created with Spire.PDF for Python.
I
S
S
N
:
2252
-
8938
I
nt
J
A
r
ti
f
I
nt
e
ll
, V
ol
.
14
, N
o.
5
,
O
c
to
be
r
20
25
:
3970
-
3981
3978
s
tu
d
y’
s
g
oa
ls
m
a
ke
it
a
n
a
pp
r
op
r
ia
t
e
c
ho
ic
e
f
o
r
a
s
s
e
s
s
in
g
t
he
im
pa
c
t
of
a
ug
m
e
n
ta
t
io
n
te
c
hni
que
s
on
ni
t
ik
ba
ti
k
c
la
s
s
i
f
ic
a
t
io
n.
W
e
ta
ke
th
e
c
o
n
f
us
i
on
m
a
t
r
i
x
on
th
r
e
e
c
la
s
s
i
f
i
c
a
t
io
n
m
ode
ls
th
a
t
ha
ve
hi
gh
pe
r
f
o
r
m
a
nc
e
.
T
he
m
ode
l
w
a
s
bu
il
t
w
it
h
da
ta
us
in
g
f
l
ip
pi
ng,
c
u
t
-
ou
t
,
a
n
d
D
C
G
A
N
a
u
gm
e
nt
a
ti
on
.
C
la
s
s
i
f
ic
a
t
io
n
r
e
s
u
lt
s
a
r
e
p
r
e
s
e
n
te
d
in
m
a
tr
ix
f
o
r
m
a
t,
w
he
r
e
r
o
w
s
r
e
p
r
e
s
e
nt
a
c
tu
a
l
c
la
s
s
e
s
,
a
nd
c
o
lu
m
ns
r
e
pr
e
s
e
n
t
pr
e
di
c
te
d
c
la
s
s
e
s
.
D
ia
gon
a
l
e
le
m
e
n
ts
i
ndi
c
a
te
c
o
r
r
e
c
t
pr
e
d
ic
t
io
n
s
,
w
h
il
e
of
f
-
d
ia
g
ona
l
e
le
m
e
nt
s
in
di
c
a
te
m
is
c
la
s
s
i
f
i
c
a
t
io
n.
F
i
gu
r
e
6
vi
s
ua
l
iz
e
s
th
e
c
la
s
s
i
f
ic
a
t
io
n
pe
r
f
o
r
m
a
nc
e
m
e
t
r
i
c
s
.
F
ig
u
r
e
6
(
a
)
d
is
p
la
ys
a
c
on
f
us
io
n
m
a
t
r
i
x
f
o
r
th
e
f
li
pp
in
g
-
a
ug
m
e
n
te
d
da
ta
s
e
t
.
F
i
gu
r
e
6(
b)
p
r
e
s
e
n
ts
a
c
o
nf
us
i
on
m
a
tr
ix
f
or
t
he
c
ut
out
-
a
ug
m
e
n
te
d
da
t
a
s
e
t
.
F
ig
ur
e
6
(
c
)
s
h
ow
s
a
c
o
nf
us
i
on
m
a
t
r
i
x
f
or
t
he
D
C
G
A
N
-
a
u
gm
e
nt
e
d
da
ta
s
e
t
.
T
he
s
e
r
e
s
u
lt
s
de
m
ons
t
r
a
te
t
ha
t
a
u
gm
e
nt
a
ti
on
m
e
th
o
ds
p
r
e
s
e
r
vi
n
g
te
xt
ur
e
c
o
he
r
e
nc
e
s
i
gn
if
ic
a
nt
ly
e
nha
nc
e
c
la
s
s
i
f
ic
a
t
io
n p
e
r
f
o
r
m
a
nc
e
.
(
a
)
(
b)
(
c
)
F
ig
ur
e
6. C
onf
us
io
n m
a
tr
ic
e
s
of
t
he
c
la
s
s
if
ic
a
ti
on mode
l
u
s
in
g
di
f
f
e
r
e
nt
a
ugm
e
nt
a
ti
on t
e
c
hni
que
s
:
(
a
)
f
li
ppi
ng, (
b
)
c
ut
-
out
, a
nd (
c
)
D
C
G
A
N
.
F
ig
ur
e
6
s
how
s
th
a
t
e
a
c
h
a
ugm
e
nt
a
ti
on
m
e
th
od
ha
s
a
s
li
g
ht
ly
di
f
f
e
r
e
nt
e
f
f
e
c
t
on
th
e
m
ode
l'
s
pe
r
f
or
m
a
nc
e
,
w
it
h
f
li
p
a
nd
D
C
G
A
N
s
how
in
g
th
e
m
os
t
c
on
s
is
te
nt
r
e
s
ul
ts
.
A
t
th
e
s
a
m
e
ti
m
e
,
c
ut
-
o
ut
m
a
y
in
tr
oduc
e
m
or
e
c
om
pl
e
xi
ty
due
to
im
a
ge
oc
c
lu
s
io
n.
T
he
th
r
e
e
c
onf
us
io
n
m
a
tr
ic
e
s
s
how
th
a
t
f
iv
e
ni
ti
k
ba
ti
k
m
ot
if
s
a
r
e
in
c
or
r
e
c
tl
y
c
la
s
s
if
ie
d:
s
e
ka
r
ke
b
e
n,
s
e
ka
r
da
ng
a
n,
ge
dha
nga
n,
s
e
ka
r
pa
la
,
a
nd
s
e
ka
r
ke
n
a
nga
.
F
ig
ur
e
7 s
how
s
t
he
f
iv
e
ni
ti
k ba
ti
k
m
ot
if
s
th
a
t
a
r
e
i
nc
or
r
e
c
tl
y c
la
s
s
if
ie
d.
S
e
ka
r
ke
be
n
S
e
ka
r
da
nga
n
G
e
dha
nga
n
S
e
ka
r
pa
l
a
S
e
ka
r
ke
na
nga
F
ig
ur
e
7
. T
he
f
iv
e
ni
ti
k ba
ti
k
m
ot
if
s
t
ha
t
w
e
r
e
i
nc
or
r
e
c
tl
y c
la
s
s
i
f
ie
d
T
he
b
a
ti
k
m
ot
if
s
di
s
pl
a
ye
d
(
s
e
k
a
r
ke
be
n,
s
e
k
a
r
da
nga
n,
ge
dha
nga
n,
s
e
ka
r
pa
la
,
a
nd
s
e
ka
r
ke
na
n
g
a
)
m
a
y
be
c
ha
ll
e
ngi
ng
to
pr
e
di
c
t
by
th
e
m
ode
l
due
to
s
e
ve
r
a
l
f
a
c
t
or
s
:
i)
t
he
f
iv
e
ni
ti
k
ba
ti
k
m
ot
if
s
ha
ve
s
im
il
a
r
ge
om
e
tr
ic
s
ha
pe
s
or
r
e
pe
a
ti
ng
pa
tt
e
r
n
s
(
f
or
e
xa
m
pl
e
,
s
e
ka
r
ke
be
n
a
nd
s
e
ka
r
pa
la
ha
ve
c
ir
c
ul
a
r
a
nd
s
ym
m
e
tr
ic
a
l
e
le
m
e
nt
s
)
,
s
o
th
e
y
c
a
n
c
onf
us
e
th
e
m
ode
l.
I
f
th
e
f
e
a
tu
r
e
s
e
xt
r
a
c
te
d
by
th
e
m
ode
l
f
oc
us
m
or
e
on
ge
om
e
tr
ic
s
tr
uc
tu
r
e
th
a
n
f
in
e
de
ta
il
s
,
th
is
c
a
n
le
a
d
to
m
i
s
c
la
s
s
if
ic
a
ti
on;
ii
)
s
ubt
le
te
xt
ur
e
di
f
f
e
r
e
nc
e
s
c
a
u
s
e
th
e
te
xt
ur
e
di
f
f
e
r
e
nc
e
s
to
be
s
ubt
le
a
nd
not
e
a
s
il
y
c
a
pt
ur
e
d
by
th
e
B
S
I
F
f
e
a
tu
r
e
e
xt
r
a
c
ti
on
m
e
th
od;
a
nd
ii
i)
a
ll
m
ot
if
s
ha
ve
a
s
im
il
a
r
c
ol
or
s
c
h
e
m
e
(
m
a
in
ly
da
r
k
ba
c
kgr
ounds
w
it
h
li
ght
e
r
pa
tt
e
r
ns
)
.
B
e
c
a
u
s
e
c
ol
or
in
f
or
m
a
ti
on i
s
l
e
s
s
r
e
le
va
nt
or
di
s
c
a
r
de
d e
nt
ir
e
ly
i
n gr
a
ys
c
a
le
f
e
a
tu
r
e
e
xt
r
a
c
ti
on me
th
ods
, t
he
m
ode
l
m
a
y ne
e
d
he
lp
di
f
f
e
r
e
nt
ia
ti
ng e
f
f
e
c
ti
ve
ly
ba
s
e
d on te
xt
ur
e
a
lo
ne
.
Evaluation Warning : The document was created with Spire.PDF for Python.
I
nt
J
A
r
ti
f
I
nt
e
ll
I
S
S
N
:
2252
-
8938
O
pt
imi
z
in
g ni
ti
k
bat
ik
c
la
s
s
if
ic
at
io
n t
hr
ough c
om
pa
r
at
iv
e
analy
s
is
of
i
m
age
augme
nt
at
io
n
(
Supr
apt
o
)
3979
5.
C
O
N
C
L
U
S
I
O
N
T
hi
s
r
e
s
e
a
r
c
h
ha
s
c
om
pr
e
he
n
s
iv
e
ly
e
va
lu
a
te
d
va
r
io
us
im
a
ge
da
ta
a
ugm
e
nt
a
ti
on
te
c
hni
que
s
in
th
e
c
ont
e
xt
of
ni
ti
k
ba
ti
k
m
ot
if
c
la
s
s
if
ic
a
ti
on
.
B
y
c
om
p
a
r
in
g
ge
om
e
tr
ic
tr
a
ns
f
or
m
a
ti
ons
,
in
te
ns
it
y
tr
a
ns
f
or
m
a
ti
ons
,
non
-
in
s
ta
nc
e
le
ve
l
a
ugm
e
nt
a
ti
on,
a
nd
unc
on
di
ti
ona
l
im
a
ge
ge
ne
r
a
ti
on
,
w
e
id
e
nt
if
ie
d
th
e
ir
im
pa
c
t
on
th
e
pe
r
f
or
m
a
nc
e
of
r
a
ndom
f
or
e
s
t
-
ba
s
e
d
c
la
s
s
if
ic
a
ti
on
m
ode
ls
.
T
he
r
e
s
ul
ts
d
e
m
ons
tr
a
te
th
a
t
s
pe
c
if
ic
a
ugm
e
nt
a
ti
on
te
c
hni
que
s
,
na
m
e
ly
f
li
ppi
ng,
c
ut
-
out
,
a
n
d
D
C
G
A
N
,
s
ig
ni
f
ic
a
nt
ly
e
nha
nc
e
th
e
a
c
c
ur
a
c
y
of
th
e
ni
ti
k
ba
ti
k
c
la
s
s
if
ic
a
ti
on
m
ode
l.
F
li
ppi
ng
a
c
hi
e
ve
d
th
e
hi
ghe
s
t
a
c
c
ur
a
c
y
im
pr
ove
m
e
nt
of
20.20%
c
om
pa
r
e
d
to
th
e
ba
s
e
li
ne
m
ode
l
w
it
hout
a
ugm
e
nt
a
ti
on,
f
ol
l
ow
e
d
by
c
ut
-
out
a
t
19.27%
a
nd
D
C
G
A
N
a
t
16.25%
.
N
ot
a
bl
y,
th
e
D
C
G
A
N
a
ugm
e
nt
a
ti
on
te
c
hni
que
e
xh
ib
it
e
d
th
e
hi
ghe
s
t
s
ta
bi
li
ty
,
w
it
h
a
s
ta
nda
r
d
de
vi
a
ti
on
of
0.78%
,
in
di
c
a
ti
ng
c
ons
is
te
nt
p
e
r
f
or
m
a
nc
e
a
c
r
os
s
va
li
da
ti
on
f
ol
ds
.
T
he
f
in
di
ngs
h
a
ve
im
por
ta
nt
im
pl
ic
a
ti
ons
f
or
bot
h
pr
a
c
ti
c
a
l
a
nd
th
e
or
e
ti
c
a
l
dom
a
in
s
.
P
r
a
c
ti
c
a
ll
y,
th
e
im
pr
ove
d
c
la
s
s
if
ic
a
ti
on
a
c
c
ur
a
c
y
c
a
n
be
us
e
d
f
or
th
e
de
ve
lo
pm
e
nt
of
a
ut
om
a
te
d
b
a
ti
k
m
ot
if
r
e
c
ogni
ti
on
s
ys
te
m
s
,
s
uppor
ti
ng
c
ul
tu
r
a
l
pr
e
s
e
r
va
ti
on
e
f
f
or
ts
,
m
us
e
um
di
gi
ti
z
a
ti
on
,
a
nd
a
ppl
ic
a
ti
ons
in
th
e
c
r
e
a
ti
ve
in
dus
tr
y.
T
he
or
e
ti
c
a
ll
y,
th
i
s
r
e
s
e
a
r
c
h
de
m
ons
tr
a
te
s
how
di
f
f
e
r
e
nt
a
ugm
e
nt
a
ti
on
te
c
hni
que
s
c
a
n
ove
r
c
om
e
th
e
c
h
a
ll
e
nge
s
a
s
s
oc
ia
te
d
w
it
h
li
m
it
e
d
a
nd
c
om
pl
e
x
da
ta
s
e
ts
,
pr
ovi
di
ng
in
s
ig
ht
in
to
th
e
ir
e
f
f
e
c
ti
ve
ne
s
s
a
nd
s
ta
bi
li
ty
in
c
om
put
e
r
vi
s
io
n
ta
s
k
s
.
T
hi
s
r
e
s
e
a
r
c
h ha
s
s
om
e
l
im
it
a
ti
ons
. T
he
da
ta
s
e
t
us
e
d i
s
r
e
la
ti
ve
ly
s
m
a
ll
, c
ons
is
ti
ng of
onl
y 240 im
a
ge
s
, w
hi
c
h doe
s
not
f
ul
ly
r
e
p
r
e
s
e
nt
th
e
di
ve
r
s
it
y
of
ni
ti
k
ba
ti
k
m
ot
if
s
in
r
e
a
l
-
w
or
ld
s
c
e
na
r
io
s
.
M
or
e
ove
r
,
th
e
us
e
of
gr
a
ys
c
a
le
pr
e
pr
oc
e
s
s
in
g
m
a
y
ha
ve
ove
r
lo
oke
d
pot
e
nt
ia
l
c
ol
or
-
ba
s
e
d
f
e
a
tu
r
e
s
th
a
t
c
oul
d
im
pr
ove
th
e
c
la
s
s
if
ic
a
ti
on
pe
r
f
or
m
a
nc
e
f
or
s
pe
c
if
ic
m
ot
if
s
.
F
in
a
ll
y,
a
lt
hough
th
e
r
a
ndom
f
or
e
s
t
c
la
s
s
if
ie
r
pr
ove
d
to
be
e
f
f
e
c
ti
ve
,
it
m
a
y
not
f
ul
ly
e
xpl
oi
t
th
e
c
a
p
a
bi
li
ti
e
s
of
m
or
e
a
dv
a
nc
e
d
de
e
p
le
a
r
ni
ng
m
ode
ls
,
w
hi
c
h
c
oul
d
of
f
e
r
f
ur
th
e
r
im
pr
ove
m
e
nt
s
.
F
or
f
ut
u
r
e
r
e
s
e
a
r
c
h,
w
e
s
ugge
s
t
e
xpl
or
in
g
th
e
in
c
or
por
a
ti
on
of
m
ul
ti
pl
e
a
ugm
e
nt
a
ti
on
te
c
hni
que
s
to
e
nha
nc
e
th
e
m
od
e
l'
s
r
obus
tn
e
s
s
f
ur
th
e
r
.
I
n
a
d
di
ti
on,
e
xpe
r
im
e
nt
in
g
w
it
h
la
r
ge
r
a
nd
m
or
e
di
ve
r
s
e
da
ta
s
e
ts
,
in
c
lu
di
ng
im
a
ge
s
w
it
h
va
r
yi
ng
il
lu
m
in
a
ti
on
or
oc
c
lu
s
io
n,
m
a
y
e
nha
n
c
e
th
e
m
ode
l'
s
ge
ne
r
a
li
z
a
bi
li
ty
.
L
a
s
tl
y,
e
va
lu
a
ti
ng
th
e
pe
r
f
or
m
a
nc
e
of
de
e
p
l
e
a
r
ni
ng
a
r
c
hi
te
c
tu
r
e
a
lo
ng
w
it
h
a
ugm
e
nt
a
ti
on
te
c
hni
que
s
m
a
y pr
ovi
de
f
ur
th
e
r
i
ns
ig
ht
s
t
o
opt
im
iz
e
th
e
c
la
s
s
if
ic
a
ti
on of
ni
ti
k ba
ti
k
.
F
U
N
D
I
N
G
I
N
F
O
R
M
A
T
I
O
N
A
ut
hor
s
s
ta
te
no f
undi
ng i
nvol
ve
d.
A
U
T
H
O
R
C
O
N
T
R
I
B
U
T
I
O
N
S
S
T
A
T
E
M
E
N
T
T
hi
s
jo
ur
na
l
us
e
s
th
e
C
ont
r
ib
ut
or
R
ol
e
s
T
a
xonomy
(
C
R
e
di
T
)
to
r
e
c
ogni
z
e
in
di
vi
dua
l
a
ut
hor
c
ont
r
ib
ut
io
ns
, r
e
duc
e
a
ut
hor
s
hi
p di
s
put
e
s
,
a
nd f
a
c
il
it
a
te
c
ol
la
bo
r
a
ti
on.
N
am
e
o
f
A
u
t
h
or
C
M
So
Va
Fo
I
R
D
O
E
Vi
Su
P
Fu
S
upr
a
pt
o
✓
✓
✓
✓
✓
✓
✓
✓
✓
✓
M
e
il
a
ny N
ons
i
T
e
nt
ua
✓
✓
✓
✓
✓
✓
A
hm
a
d R
iz
ki
M
a
ul
a
n
a
✓
✓
✓
✓
✓
✓
C
:
C
onc
e
pt
ua
l
i
z
a
t
i
on
M
:
M
e
t
hodol
ogy
So
:
So
f
t
w
a
r
e
Va
:
Va
l
i
da
t
i
on
Fo
:
Fo
r
m
a
l
a
na
l
ys
i
s
I
:
I
nve
s
t
i
ga
t
i
on
R
:
R
e
s
our
c
e
s
D
:
D
a
t
a
C
ur
a
t
i
on
O
:
W
r
i
t
i
ng
-
O
r
i
gi
na
l
D
r
a
f
t
E
:
W
r
i
t
i
ng
-
R
e
vi
e
w
&
E
di
t
i
ng
Vi
:
Vi
s
ua
l
i
z
a
t
i
on
Su
:
Su
pe
r
vi
s
i
on
P
:
P
r
oj
e
c
t
a
dm
i
ni
s
t
r
a
t
i
on
Fu
:
Fu
ndi
ng a
c
qui
s
i
t
i
on
C
O
N
F
L
I
C
T
O
F
I
N
T
E
R
E
S
T
S
T
A
T
E
M
E
N
T
A
ut
hor
s
s
ta
te
no c
onf
li
c
t
of
i
nt
e
r
e
s
t.
D
A
T
A
A
V
A
I
L
A
B
I
L
I
T
Y
T
he
da
ta
th
a
t
s
uppor
t
th
e
f
in
di
ngs
o
f
th
is
s
tu
dy
a
r
e
ope
nl
y
a
va
il
a
bl
e
in
M
e
nde
le
y
D
a
ta
a
t
ht
tp
:/
/d
oi
.or
g/
10.17632/s
gh484jxz
y.3
.
R
E
F
E
R
E
N
C
E
S
[
1]
A
.
D
.
W
i
ba
w
a
,
E
.
A
.
W
i
c
a
ks
o
no,
S
.
D
.
S
ur
ya
n
i
,
a
nd
R
.
R
u
m
a
di
,
“
J
a
va
ne
s
e
ba
t
i
k
i
m
a
g
e
c
l
a
s
s
i
f
i
c
a
t
i
on
us
i
n
g
s
e
l
f
-
o
r
ga
ni
z
i
n
g
m
a
p,”
i
n
202
3
I
nt
e
r
na
t
i
on
al
C
on
f
e
r
e
n
c
e
o
n
C
om
p
ut
e
r
Sc
i
e
nc
e
,
I
n
f
or
m
at
i
o
n
T
e
c
h
no
l
o
gy
and
E
ng
i
ne
e
r
i
n
g
(
I
C
C
oS
I
T
E
)
,
2
02
3,
pp.
47
2
–
477
,
doi
:
10.
11
09
/
I
C
C
oS
I
T
E
5
764
1.
202
3.
101
27
783
.
Evaluation Warning : The document was created with Spire.PDF for Python.