I
nd
o
ne
s
ia
n J
o
urna
l o
f
E
lect
rica
l En
g
ineering
a
nd
Co
m
p
u
t
er
Science
Vo
l.
12
,
No
.
2
,
N
o
v
e
m
b
er
201
8
,
p
p
.
7
8
3
~
7
9
3
I
SS
N:
2
5
0
2
-
4
7
5
2
,
DOI
:
1
0
.
1
1
5
9
1
/i
j
ee
cs.v
1
2
.i
2
.
p
p
783
-
7
9
3
783
J
o
ur
na
l ho
m
ep
a
g
e
:
h
ttp
:
//ia
e
s
co
r
e.
co
m/jo
u
r
n
a
ls
/in
d
ex
.
p
h
p
/
ijeec
s
O
bje
ct
R
ecog
niti
o
n
I
nspirin
g
HVS
M
o
ha
m
m
a
de
s
m
a
eil
A
kb
a
rp
o
ur,
Na
s
s
er
M
ehrsh
a
d,
Sey
y
ed
-
M
o
ha
m
m
a
d Ra
za
v
i
El
e
c
tri
c
a
l
En
g
in
e
e
rin
g
,
Un
iv
e
rsit
y
o
f
Birj
a
n
d
,
Birj
a
n
d
,
Ira
n
Art
icle
I
nfo
AB
ST
RAC
T
A
r
ticle
his
to
r
y:
R
ec
eiv
ed
A
p
r
9
,
2
0
1
8
R
ev
i
s
ed
Ma
y
2
0
,
2
0
1
8
A
cc
ep
ted
Ju
l
11
,
2
0
1
8
Hu
m
a
n
re
c
o
g
n
ize
o
b
jec
ts
in
c
o
m
p
lex
n
a
tu
ra
l
i
m
a
g
e
s
v
e
r
y
f
a
st
w
it
h
in
a
f
ra
c
ti
o
n
o
f
a
se
c
o
n
d
.
M
a
n
y
c
o
m
p
u
tatio
n
a
l
o
b
jec
t
re
c
o
g
n
it
io
n
m
o
d
e
ls
in
sp
ired
f
ro
m
th
is
p
o
w
e
r
f
u
l
a
b
il
it
y
o
f
h
u
m
a
n
.
T
h
e
Hu
m
a
n
V
isu
a
l
S
y
ste
m
(HV
S
)
re
c
o
g
n
ize
s
o
b
jec
t
in
se
v
e
ra
l
p
r
o
c
e
ss
in
g
la
y
e
rs
w
h
ich
w
e
k
n
o
w
th
e
m
a
s
h
iera
rc
h
ica
ll
y
m
o
d
e
l.
Du
e
to
a
m
a
z
in
g
c
o
m
p
lex
it
y
o
f
HV
S
a
n
d
th
e
c
o
n
n
e
c
ti
o
n
s
i
n
v
isu
a
l
p
a
th
w
a
y
,
c
o
m
p
u
tatio
n
a
l
m
o
d
e
li
n
g
o
f
HV
S
d
irec
tl
y
f
ro
m
it
s
p
h
y
sio
lo
g
y
is
n
o
t
p
o
ss
i
b
le.
S
o
it
c
o
n
sid
e
re
d
a
s
a
so
m
e
b
lo
c
k
s
a
n
d
e
a
c
h
b
l
o
c
k
m
o
d
e
led
se
p
a
ra
tely
.
On
e
m
o
d
e
ls
in
sp
i
rin
g
o
f
H
V
S
is
HMA
X
w
h
ich
it
s
m
a
in
p
ro
b
lem
is
se
lec
ti
n
g
p
a
tch
e
s
in
ra
n
d
o
m
w
a
y
.
A
s
H
M
A
X
is
a
h
iera
rc
h
ica
l
m
o
d
e
l,
HMA
X
c
a
n
e
n
h
a
n
c
e
d
w
it
h
e
n
h
a
n
c
in
g
e
a
c
h
la
y
e
r
se
p
a
ra
tel
y
.
In
th
is
p
a
p
e
r
in
ste
a
d
o
f
ra
n
d
o
m
p
a
tch
e
x
trac
ti
o
n
,
De
sira
b
le
P
a
tch
e
s
f
o
r
HMA
X
(DP
HMA
X
)
w
il
l
e
x
trac
ted
.
HV
S
f
o
r
e
x
tra
c
ti
n
g
p
a
tch
f
irst
se
l
e
c
ted
p
a
tch
e
s
w
it
h
m
o
re
in
f
o
r
m
a
ti
o
n
.
F
o
r
sim
u
latin
g
t
h
is
b
lo
c
k
p
a
tch
e
s
w
it
h
m
o
re
v
a
rian
c
e
w
il
l
b
e
se
lec
ted
.
T
h
e
n
HV
S
w
il
l
c
h
o
se
p
a
tch
e
s
w
it
h
m
o
re
si
m
il
a
rit
y
in
a
c
las
s
.
F
o
r
sim
u
latin
g
t
h
is
b
lo
c
k
o
n
e
a
l
g
o
rit
h
m
is
u
se
d
.
F
o
r
e
v
a
lu
a
ti
n
g
p
ro
p
o
se
d
m
e
th
o
d
,
Ca
lt
e
c
h
5
a
n
d
Ca
lt
e
c
h
1
0
1
d
a
tas
e
ts
a
r
e
u
se
d
.
Re
su
lt
s
sh
o
w
th
a
t
th
e
p
ro
p
o
se
d
m
e
th
o
d
(D
P
M
A
X
)
p
ro
v
id
e
s
a
sig
n
if
ica
n
t
p
e
r
f
o
rm
a
n
c
e
o
v
e
r
H
M
A
X
a
n
d
o
th
e
r
m
o
d
e
ls
w
it
h
th
e
sa
m
e
f
ra
m
e
w
o
rk
.
K
ey
w
o
r
d
s
:
Ob
j
ec
t r
ec
o
g
n
itio
n
HM
AX
HVS
H
ier
ar
ch
ical
m
o
d
el
Co
p
y
rig
h
t
©
2
0
1
8
In
stit
u
te o
f
A
d
v
a
n
c
e
d
E
n
g
i
n
e
e
rin
g
a
n
d
S
c
ien
c
e
.
Al
l
rig
h
ts
re
se
rv
e
d
.
C
o
r
r
e
s
p
o
nd
ing
A
uth
o
r
:
Mo
h
a
m
m
ad
es
m
ae
il
Ak
b
ar
p
o
u
r
E
lectr
ical
E
n
g
i
n
ee
r
i
n
g
,
U
n
i
v
er
s
it
y
o
f
B
ir
j
an
d
,
B
ir
j
an
d
,
I
r
an
.
E
m
ail:
es
m
ae
i
l.a
k
b
ar
p
o
u
r
@
b
i
r
j
an
d
.
ac
.
ir
1.
I
NT
RO
D
UCT
I
O
N
T
h
e
Hu
m
a
n
V
is
u
al
S
y
s
te
m
(
HVS)
is
ab
le
to
r
ec
o
g
n
ize
o
b
j
ec
ts
ea
s
il
y
i
n
a
c
lu
t
ter
ed
s
ce
n
e
i
n
les
s
t
h
a
n
a
s
ec
o
n
d
.
R
ec
en
t
w
o
r
k
s
in
s
p
i
r
in
g
t
h
e
H
u
m
an
Vi
s
u
a
l
S
y
s
te
m
(
HV
S)
in
i
m
a
g
e
p
r
o
ce
s
s
in
g
ar
e
s
u
ch
as
i
m
ag
e
en
h
a
n
ce
m
en
t
[
1
]
,
d
ata
h
id
in
g
[
2
,
3
]
,
d
ig
ital
i
m
a
g
e
f
u
s
io
n
[
4
]
,
r
o
b
u
s
t
o
b
j
ec
t
r
ec
o
g
n
itio
n
[
5
]
.
HVS
p
r
o
ce
s
s
es
i
m
a
g
es
ea
s
il
y
,
w
h
ile
t
h
e
m
o
s
t
p
o
w
er
f
u
l
co
m
p
u
ter
s
y
s
te
m
s
a
r
e
g
e
n
er
all
y
n
o
t
ca
p
ab
le
o
f
d
o
in
g
s
o
.
Du
e
to
t
h
e
tr
e
m
en
d
o
u
s
co
m
p
le
x
it
y
o
f
H
VS
an
d
a
m
az
i
n
g
co
n
n
ec
tio
n
s
in
v
is
u
al
p
at
h
w
a
y
,
co
m
p
u
ta
t
io
n
al
m
o
d
eli
n
g
o
f
HVS
f
o
r
i
m
a
g
e
p
r
o
ce
s
s
i
n
g
ap
p
licatio
n
s
d
ir
ec
tl
y
f
r
o
m
it
s
o
v
er
all
an
ato
m
y
a
n
d
p
h
y
s
io
lo
g
y
is
n
o
t
p
o
s
s
ib
le
[
6
]
.
On
e
o
f
w
a
y
to
o
v
er
co
m
e
t
h
e
li
m
ita
tio
n
is
t
h
e
i
n
p
u
t
-
o
u
tp
u
t
m
o
d
eli
n
g
o
f
th
e
v
is
u
al
s
y
s
te
m
(
i.e
.
th
e
s
alie
n
c
y
m
ap
)
[
7
,
8
]
.
An
o
th
er
w
a
y
is
m
o
d
elin
g
o
f
t
h
e
s
i
m
p
le
s
u
b
s
y
s
te
m
s
a
n
d
t
h
e
ir
s
y
s
te
m
atica
ll
y
co
m
b
i
n
atio
n
b
ased
o
n
th
e
HV
S
s
tr
u
ct
u
r
e
(
i.e
.
ed
g
e
a
n
d
li
n
e
d
etec
tio
n
,
co
n
to
u
r
e
x
tr
ac
tio
n
a
n
d
tex
t
u
r
e
d
ia
g
n
o
s
e)
[
9
,
10
]
.
I
t
s
ee
m
s
t
h
at
th
e
m
a
n
n
er
o
f
th
e
H
VS
in
th
e
o
b
j
ec
t
d
escr
ip
tio
n
s
tag
e
an
d
o
b
j
ec
t
r
ec
o
g
n
izin
g
(
m
atch
i
n
g
)
p
r
o
ce
s
s
i
s
o
p
tim
ized
.
I
n
th
e
f
ir
s
t
s
tep
o
f
m
o
d
elin
g
th
e
v
is
u
al
s
y
s
te
m
b
eh
av
io
r
in
o
b
j
ec
t
r
ec
o
g
n
itio
n
,
an
ap
p
r
o
p
r
iate
o
b
j
ec
t d
escr
ip
to
r
s
h
o
u
ld
b
e
p
r
esen
ted
.
T
h
is
d
escr
ip
to
r
m
u
s
t
b
e
in
d
ep
en
d
en
t to
s
ca
le
an
d
r
o
tatio
n
[
5
,
11
,
12
]
.
HVS
f
o
r
o
b
j
ec
t
d
escr
ip
tio
n
u
s
es
p
r
o
ce
s
s
es
s
u
c
h
as
s
a
lien
c
y
m
ap
,
ed
g
e
d
etec
tio
n
,
l
in
e
d
etec
t
io
n
,
co
n
to
u
r
ex
tr
ac
tio
n
an
d
te
x
t
u
r
e
d
iag
n
o
s
e.
T
h
e
s
alie
n
c
y
m
ap
[
7
,
8
]
is
th
e
f
ir
s
t
to
p
o
g
r
ap
h
i
ca
ll
y
ar
r
an
g
ed
m
a
p
th
at
r
ep
r
esen
ts
v
is
u
al
s
alie
n
c
y
o
f
a
co
r
r
esp
o
n
d
in
g
v
is
u
al
s
ce
n
e.
Fo
r
ed
g
es
d
etec
tio
n
[
13
,
1
4
]
th
e
r
etin
a
an
d
L
GN
ce
ll
s
ar
e
in
s
p
ir
ed
.
T
h
e
y
d
o
n
’
t h
a
v
e
d
ir
ec
tio
n
al
s
elec
tio
n
b
ec
au
s
e
o
f
t
h
e
cir
c
u
lar
r
ec
ep
tiv
e
f
ield
.
Evaluation Warning : The document was created with Spire.PDF for Python.
I
SS
N
:
2
5
0
2
-
4752
I
n
d
o
n
esia
n
J
E
lec
E
n
g
&
C
o
m
p
Sci,
Vo
l.
12
,
No
.
2
,
No
v
e
m
b
er
2
0
1
8
:
7
8
3
–
7
9
3
784
Fo
r
lin
e
d
etec
tio
n
[
15
]
t
h
e
h
u
m
an
p
r
i
m
ar
y
v
is
u
al
co
r
tex
(
V
1
)
s
i
m
p
le
ce
lls
ar
e
i
n
s
p
ir
ed
b
ec
au
s
e
o
f
th
eir
d
ir
ec
tio
n
al
r
ec
ep
tiv
e
f
iel
d
.
A
s
et
o
f
ad
ap
ti
v
e
f
i
lter
s
is
d
er
iv
ed
b
y
lear
n
in
g
m
ec
h
an
i
s
m
w
h
ich
e
m
u
late
s
th
e
V1
s
i
m
p
le
ce
lls
.
T
h
ese
f
ilter
s
ar
e
ap
p
lied
at
ev
er
y
p
o
s
itio
n
o
f
t
h
e
i
n
p
u
t
i
m
a
g
e
to
g
et
a
li
n
e
f
ea
t
u
r
e
re
p
r
esen
tatio
n
.
No
n
-
clas
s
ical
r
ec
ep
tiv
e
f
ield
(
N
-
C
R
F)
i
n
h
ib
it
io
n
m
ec
h
an
is
m
i
s
a
n
ex
a
m
p
le
to
d
esi
g
n
p
h
y
s
io
lo
g
icall
y
p
la
u
s
ib
le
co
n
t
o
u
r
d
etec
tio
n
m
o
d
els
[
13
,
16
,
17
]
.
N
-
C
R
F
m
ec
h
a
n
i
s
m
s
u
p
p
r
ess
es
ed
g
es
w
h
ich
m
ak
e
p
ar
t o
f
t
h
e
tex
tu
r
e,
w
h
ile
it d
o
es n
o
t su
p
p
r
ess
ed
g
e
s
th
a
t b
elo
n
g
to
th
e
co
n
to
u
r
s
o
f
o
b
j
ec
ts
.
Sin
ce
HVS
at
f
ir
s
t
i
s
o
lates
co
n
to
u
r
o
f
o
b
j
ec
ts
f
r
o
m
s
ce
n
e
i
m
ag
e
s
i
n
it
s
ea
r
l
y
s
ta
g
es
o
f
v
i
s
u
al
co
r
te
x
,
it
is
ab
le
to
d
is
t
in
g
u
is
h
t
h
e
tex
t
u
r
e
ed
g
e
s
a
n
d
b
o
u
n
d
ar
y
o
f
o
b
j
ec
ts
in
s
ce
n
e
i
m
a
g
es,
k
n
o
w
n
as
th
e
co
n
tr
ast
[
18
,
19
]
.
Af
ter
o
b
j
ec
t
d
escr
ip
tio
n
,
it
s
r
e
co
g
n
itio
n
ca
n
b
e
m
o
d
eled
b
y
u
s
i
n
g
t
h
e
e
x
tr
ac
ted
f
ea
t
u
r
es
[
20
,
21
]
.
Fo
r
th
is
p
u
r
p
o
s
e,
i
n
ad
d
itio
n
to
s
i
m
p
le
ce
ll
s
,
co
m
p
le
x
ce
lls
o
f
t
h
e
co
r
tex
m
u
s
t
b
e
m
o
d
eled
.
T
h
u
s
,
t
h
e
m
o
d
els
ar
e
co
n
s
is
t
o
f
t
w
o
k
i
n
d
s
o
f
la
y
er
s
,
ea
ch
o
f
w
h
ic
h
e
m
u
lates
th
e
f
u
n
ctio
n
s
o
f
V1
s
i
m
p
le
a
n
d
c
o
m
p
le
x
ce
ll
s
.
T
h
u
s
,
h
ier
ar
ch
ical
m
o
d
els
ar
e
cr
ea
t
ed
.
On
e
ca
s
es
in
s
p
ir
ed
b
y
th
e
h
ier
ar
ch
ical
n
at
u
r
e
o
f
p
r
i
m
ate
v
i
s
u
al
co
r
tex
,
i
s
HM
AX
h
ier
ar
ch
ical
m
o
d
el
[
22
]
(
th
e
n
eu
r
al
n
et
w
o
r
k
m
o
d
el
f
o
r
i
m
a
g
e
clas
s
i
f
icatio
n
)
.
T
h
e
HM
AX
m
o
d
el
ca
n
b
e
d
escr
ib
ed
as
a
f
o
u
r
-
le
v
el
ar
ch
itect
u
r
e
w
it
h
a
f
ir
s
t
le
v
el
co
n
s
i
s
ti
n
g
o
f
m
u
l
ti
-
s
ca
le
a
n
d
m
u
lti
-
o
r
ien
tatio
n
lo
ca
l
f
ilter
s
(
i.e
.
Gau
s
s
ia
n
d
er
iv
ati
v
es
o
r
Gab
o
r
f
ilter
s
)
.
T
h
ese
n
etw
o
r
k
s
co
m
b
in
e
t
h
e
lo
w
le
v
el
r
ep
r
esen
tatio
n
s
i
n
to
o
b
j
ec
t le
v
el
r
ep
r
esen
tatio
n
s
s
u
itab
le
f
o
r
r
ec
o
g
n
itio
n
tas
k
s
[
20
]
.
Ser
r
e
et
al.
[
5
]
ex
ten
d
ed
th
e
o
r
ig
in
al
H
M
A
X
m
o
d
el
to
ad
d
m
u
l
ti
-
s
ca
le
r
ep
r
esen
ta
tio
n
s
as
w
ell
as
m
o
r
e
co
m
p
le
x
v
i
s
u
al
f
ea
t
u
r
es.
Hu
an
g
et
al.
[
3
8
]
also
im
p
r
o
v
ed
th
e
HM
A
X
m
o
d
el
w
it
h
co
n
s
tr
ain
t
s
,
a
d
if
f
er
en
t
p
o
o
lin
g
s
tr
ate
g
y
an
d
a
f
ee
d
b
ac
k
m
ec
h
a
n
i
s
m
to
i
m
p
r
o
v
e
f
ea
tu
r
e
lear
n
i
n
g
.
Da
v
id
et
al.
[
23
]
h
as
s
h
o
w
n
h
o
w
HM
AX
f
ilter
s
ca
n
o
u
tp
er
f
o
r
m
s
tate
-
of
-
t
h
e
-
ar
t
f
il
ter
s
s
u
c
h
as
SIFT
u
n
d
er
v
ar
io
u
s
co
n
tr
o
lle
d
in
v
ar
ia
n
ce
ta
s
k
s
o
n
s
y
n
t
h
etic
i
m
a
g
es.
T
h
er
iau
lt
et
al.
[
22
]
First,
HM
A
X
w
as
i
m
p
r
o
v
ed
b
y
i
n
te
g
r
atin
g
th
e
lo
ca
l
f
il
ter
s
at
t
h
e
f
ir
s
t
lev
el
i
n
to
m
o
r
e
co
m
p
le
x
f
ilter
s
at
t
h
e
la
s
t
le
v
el,
p
r
o
v
id
i
n
g
a
f
lex
ib
le
d
escr
ip
tio
n
o
f
o
b
j
ec
t
r
eg
io
n
s
a
n
d
co
m
b
i
n
i
n
g
lo
ca
l
in
f
o
r
m
atio
n
o
f
m
u
ltip
le
s
ca
le
s
an
d
o
r
ien
tatio
n
s
.
Seco
n
d
,
a
m
u
l
ti
-
r
e
s
o
lu
tio
n
s
p
atial
p
o
o
lin
g
w
a
s
in
tr
o
d
u
ce
d
.
T
h
is
p
o
o
lin
g
en
co
d
es b
o
th
lo
c
al
an
d
g
lo
b
al
s
p
atial
in
f
o
r
m
a
ti
o
n
to
p
r
o
d
u
ce
d
is
cr
im
i
n
ati
v
e
i
m
ag
e
s
i
g
n
at
u
r
es.
I
tti
e
t
al.
[
7
]
h
as
in
tr
o
d
u
ce
d
th
e
h
ier
ar
ch
ical
ap
p
r
o
ac
h
b
ased
o
n
th
r
ee
p
ar
am
eter
s
:
in
te
n
s
it
y
,
o
r
ien
tatio
n
a
n
d
co
lo
r
s
f
o
r
i
m
a
g
e
s
al
ien
c
y
m
ap
d
iag
n
o
s
is
.
T
h
e
f
in
al
m
o
d
el
o
b
tain
ed
b
y
co
m
b
in
i
n
g
t
h
e
o
u
tp
u
t
m
o
d
el
s
o
f
t
h
ese
f
ea
tu
r
e
s
.
I
n
[
3
4
,
3
5
]
s
alien
c
y
m
ap
ar
e
af
f
ec
ted
b
y
t
h
e
p
r
o
p
er
ties
o
f
th
e
o
b
j
ec
t.
P
o
u
r
asad
[
24
]
p
r
o
p
o
s
ed
a
Mo
d
if
ied
HM
AX
m
o
d
el
b
ased
o
n
co
m
b
i
n
e
d
w
it
h
th
e
v
i
s
u
al
f
ea
t
u
r
ed
m
o
d
el
f
o
r
HM
AX
(
h
en
ce
f
o
r
th
r
e
f
er
r
ed
to
as
MH
MA
X)
b
y
ca
lc
u
lat
in
g
o
p
ti
m
u
m
p
atc
h
es
b
a
s
ed
o
n
t
h
eir
in
f
o
r
m
atio
n
.
G
h
o
d
r
ati
et
al.
[
25
]
f
o
u
n
d
t
h
e
b
etter
p
atch
es f
o
r
HM
A
X
b
y
u
s
i
n
g
g
e
n
eti
c
alg
o
r
ith
m
(
h
e
n
ce
f
o
r
th
r
ef
er
r
ed
to
as GM
A
X)
.
2.
RE
VI
E
W
O
F
B
ACK
G
RO
U
ND
WO
RK
S
I
n
th
i
s
s
ec
tio
n
w
e
i
n
tr
o
d
u
ce
a
b
r
ief
r
ev
ie
w
o
f
HM
A
X
m
o
d
el
s
.
T
h
e
o
r
ig
in
al
HM
AX
m
o
d
el
is
in
s
p
ir
i
n
g
t
h
e
h
ier
ar
ch
ical
th
eo
r
y
o
f
v
is
u
al
p
r
o
ce
s
s
i
n
g
.
I
ts
ar
ch
itect
u
r
e
is
d
er
iv
ed
f
r
o
m
t
h
e
w
ell
-
k
n
o
w
n
m
o
d
el
in
tr
o
d
u
ce
d
b
y
H
u
b
el
&
W
i
esel
[
3
,
1
8
s
ab
o
u
r
i4
]
.
HM
A
X
m
o
d
els
th
e
v
e
n
tr
al
v
is
u
al
p
ath
w
a
y
f
r
o
m
f
ir
s
t
p
r
o
ce
s
s
in
g
p
ar
t
i
n
t
h
e
v
i
s
u
al
c
o
r
tex
(
V1
)
to
h
ig
h
er
lev
el
s
o
f
v
is
u
al
co
r
tex
(
e.
g
.
I
T
an
d
P
F
C
A
)
.
Sch
e
m
atic
o
f
th
e
HM
A
X
m
o
d
el
is
s
h
o
w
n
i
n
Fig
u
r
e
1
.
A
s
s
h
o
w
n
:
T
h
e
b
asic
ar
ch
itect
u
r
e
o
f
th
e
HM
AX
m
o
d
el
h
as
f
o
u
r
m
o
d
u
les
ca
l
led
an
d
.
T
h
e
s
elec
tiv
it
y
a
n
d
in
v
ar
ian
ce
i
n
cr
e
ase
as
th
e
la
y
er
s
p
r
o
g
r
es
s
alo
n
g
t
h
e
h
ier
ar
ch
ical
s
tr
u
ct
u
r
e
o
f
m
o
d
el.
T
h
ese
la
y
er
s
i
m
itate
th
e
b
eh
a
v
io
r
o
f
ce
l
ls
f
r
o
m
V1
to
I
T
co
r
tex
.
As
t
h
e
m
ai
n
p
r
o
b
le
m
o
f
HM
AX
is
r
an
d
o
m
p
atch
ex
tr
ac
tio
n
an
d
in
th
i
s
p
ap
er
w
e
s
o
lv
e
it,
it
is
co
n
s
id
er
ed
as
a
s
ep
ar
ate
m
o
d
u
le.
P
atch
es a
r
e
ex
tr
ac
ted
in
tr
ain
i
n
g
s
tep
a
n
d
u
s
ed
i
n
tes
tin
g
s
te
p
.
I
n
th
e
f
o
llo
w
i
n
g
,
w
e
e
x
p
lai
n
t
h
ese
m
o
d
u
le
s
.
L
et
th
e
n
u
m
b
er
o
f
tr
ai
n
i
n
g
i
m
a
g
e
b
e
d
en
o
ted
b
y
Ntr
,
f
o
r
ea
ch
m
o
d
u
le
w
e
w
ill
d
is
c
u
s
s
ab
o
u
t
in
p
u
t
an
d
o
u
tp
u
t
o
f
m
o
d
u
le.
Evaluation Warning : The document was created with Spire.PDF for Python.
I
n
d
o
n
esia
n
J
E
lec
E
n
g
&
C
o
m
p
Sci
I
SS
N:
2502
-
4752
O
b
ject
R
ec
o
g
n
itio
n
I
n
s
p
ir
in
g
HV
S
(
Mo
h
a
mma
d
esma
eil
A
kb
a
r
p
o
u
r
)
785
Fig
u
r
e
1
.
Sch
e
m
atic
o
f
t
h
e
H
MA
X
m
o
d
el
2
.
1
.
S1
M
o
du
le
T
h
e
f
ir
s
t
la
y
er
o
f
th
e
HM
A
X
m
o
d
el
ca
lled
1
S
,
r
ec
eiv
es
i
m
ag
e
s
as
its
in
p
u
t.
Af
ter
w
ar
d
,
th
ese
i
m
a
g
es
ar
e
u
s
ed
to
a
s
et
o
f
ed
g
e
d
etec
to
r
f
ilter
s
to
d
etec
t
t
h
eir
ed
g
es.
T
h
ese
f
i
lter
s
ar
e
b
u
ilt
b
ased
o
n
t
h
e
Gab
o
r
f
u
n
ctio
n
.
T
h
ese
Gab
o
r
f
ilter
f
o
r
m
u
la
i
s
m
e
n
tio
n
ed
as E
q
u
atio
n
1
an
d
E
q
u
atio
n
2:
(
)
(
)
(
)
(
1
)
(
2
)
W
h
er
e
th
e
p
ar
a
m
eter
s
γ
,
σ
,
θ,
an
d
λ
ar
e
asp
ec
t
r
atio
,
ef
f
ec
t
iv
e
w
id
t
h
,
o
r
ien
tatio
n
a
n
d
t
h
e
w
a
v
ele
n
g
t
h
o
f
th
e
Gab
o
r
f
ilter
,
r
esp
ec
tiv
el
y
.
T
h
e
o
u
tp
u
t o
f
t
h
is
m
o
d
u
le
o
b
tain
e
d
b
y
co
n
v
o
lv
i
n
g
th
e
i
n
p
u
t i
m
a
g
eb
y
Gab
o
r
f
ilter
s
.
T
h
is
la
y
er
p
ar
a
m
eter
s
(
Gab
o
r
f
ilter
p
ar
a
m
eter
s
)
ar
e
s
h
o
w
n
i
n
T
ab
le
1
.
A
s
s
h
o
w
n
i
n
T
ab
le
1
,
Gab
o
r
f
ilter
p
ar
a
m
eter
s
ar
e
i
n
1
6
r
o
w
s
.
I
m
p
o
r
tan
t
n
o
te
i
s
t
h
ese
p
ar
a
m
eter
ar
e
f
o
r
ea
ch
o
r
ie
n
tatio
n
(
θ=
{0
˚
,
4
5
˚
,
9
0
˚
,
1
3
5
˚
}
)
.
Fo
r
a
n
in
p
u
t
i
m
a
g
e,
6
4
im
a
g
es
a
r
e
p
r
o
d
u
ce
d
s
u
ch
th
at
ed
g
es
ar
e
ex
tr
ac
ted
w
it
h
d
if
f
er
e
n
t size
s
an
d
o
r
ien
tat
io
n
s
.
Gab
o
r
f
ilter
s
m
e
n
tio
n
ed
in
T
ab
le
1
(
f
o
r
1
6
s
izes a
n
d
4
o
r
i
en
tatio
n
s
)
ar
e
s
h
o
w
n
in
Fi
g
u
r
e
2
.
T
ab
le
1
.
S1
an
d
C
1
P
ar
am
eter
s
f
o
r
ea
ch
o
r
ien
tatio
n
(
θ)
mo
d
u
l
e
mo
d
u
l
e
G
a
b
o
r
F
i
l
t
e
r
p
a
r
a
me
t
e
r
M
A
X
p
a
r
a
me
t
e
r
O
F
i
l
t
e
r
si
z
e
9
2
.
8
3
.
6
3
.
5
4
.
6
4
B
a
n
d
-
1
13
4
.
5
5
.
4
5
.
6
6
.
8
5
B
a
n
d
-
2
17
6
.
3
7
.
3
7
.
9
9
.
1
6
B
a
n
d
-
3
21
8
.
2
9
.
2
1
0
.
3
1
1
.
5
7
B
a
n
d
-
4
25
1
0
.
2
1
1
.
3
1
2
.
7
1
4
.
1
8
B
a
n
d
-
5
29
1
2
.
3
1
3
.
4
1
5
.
4
1
6
.
8
9
B
a
n
d
-
6
33
1
4
.
6
1
5
.
8
1
8
.
2
1
9
.
7
10
B
a
n
d
-
7
37
1
7
.
0
1
8
.
2
2
1
.
2
2
2
.
8
11
B
a
n
d
-
8
Evaluation Warning : The document was created with Spire.PDF for Python.
I
SS
N
:
2
5
0
2
-
4752
I
n
d
o
n
esia
n
J
E
lec
E
n
g
&
C
o
m
p
Sci,
Vo
l.
12
,
No
.
2
,
No
v
e
m
b
er
2
0
1
8
:
7
8
3
–
7
9
3
786
No
te
th
at
th
e
s
ize
o
f
o
u
tp
u
t
a
n
d
in
p
u
t
ar
e
eq
u
al
in
th
is
la
y
er
.
A
s
S1
m
o
d
u
le
p
r
o
d
u
ce
s
6
4
o
u
tp
u
t
s
,
th
e
in
p
u
t
s
o
f
t
h
is
m
o
d
u
le
ar
e
Ntr
t
r
ain
in
g
i
m
a
g
es a
n
d
t
h
e
o
u
tp
u
ts
ar
e
6
4
Ntr
im
ag
e
s
(
S1
l,θ
)
.
Fig
u
r
e
2
.
6
4
Gab
o
r
f
ilter
s
(
1
6
s
ca
les [
7
×7
to
3
7
×
3
7
]
b
y
4
o
r
ien
tatio
n
s
[
0
˚
,
4
5
˚
,
9
0
˚
,
1
3
5
˚
]
)
[
1
8
]
2
.
2
.
C1
M
o
du
le:
T
h
is
m
o
d
u
le
i
s
t
h
e
s
ec
o
n
d
la
y
er
in
HM
A
X
m
o
d
el
w
h
ic
h
e
m
u
lates
th
e
co
m
p
le
x
ce
l
ls
ac
t
iv
it
y
i
n
th
e
co
r
tex
.
T
h
is
la
y
er
p
ar
a
m
e
ter
is
s
h
o
w
n
i
n
T
ab
le
1
.
A
s
s
h
o
w
n
B
an
d
-
7
co
n
tai
n
s
s
ca
les
1
3
,
1
4
w
h
ic
h
ar
e
p
r
o
d
u
ce
d
b
y
Gab
o
r
f
ilter
s
w
it
h
s
ize
s
3
1
×3
1
an
d
3
3
×3
3
.
L
et
t
w
o
co
n
s
ec
u
ti
v
e
s
ca
les
b
e
d
en
o
ted
b
y
S1
l,θ
an
d
S1
l+1
,
θ.
T
h
ese
i
m
ag
e
s
ar
e
s
eg
m
e
n
ted
i
n
to
b
lo
ck
s
w
it
h
Ns×
Ns s
ize.
O
u
tp
u
t o
f
th
i
s
la
y
er
i
s
ca
lcu
la
te
d
as a
s
E
q
u
atio
n
3:
(
)
ax
{
(
)
(
)
}
(
3
)
w
h
er
e
b
is
b
an
d
,
θ
is
o
r
ien
tatio
n
,
an
d
l
ar
e
th
e
n
u
m
b
er
o
f
s
ca
les.
T
h
is
p
r
o
ce
s
s
is
f
o
r
all
o
r
ien
tatio
n
s
a
n
d
all
b
an
d
s
in
d
ep
en
d
en
tl
y
,
s
o
3
2
m
atr
ices
w
ith
K×
L
s
ize,
w
ill p
r
o
d
u
ce
f
o
r
an
y
i
n
p
u
t i
m
a
g
e
(
o
f
s
ize
K×
L
)
.
2
.
3
.
RP
E
M
o
du
le:
T
h
is
m
o
d
u
le
is
j
u
s
t
f
o
r
tr
ain
i
n
g
s
tep
,
th
i
s
m
o
d
u
le
e
x
tr
ac
ts
p
atch
es
f
r
o
m
ea
ch
o
f
th
e
3
2
o
u
tp
u
t
i
m
a
g
e
s
(
C
1
b
,
θ)
p
r
o
d
u
ce
d
b
y
C
1
m
o
d
u
le.
Fo
r
C
1
b
,
θ
f
r
o
m
i
-
t
h
tr
ai
n
in
g
i
m
a
g
e
(
i=1
,
…,
Ntr
)
,
p
atch
es
w
i
th
f
o
u
r
s
ize
s
(
4
h
×4
h
;
h
=1
,
2
,
3
,
4
)
a
r
e
ex
tr
ac
ted
.
Fo
r
ea
ch
s
ize,
m
p
atch
e
s
ar
e
ec
tr
ac
ed
.
Hen
ce
f
o
r
C
1
b
,
θ
f
r
o
m
i
-
th
tr
ai
n
in
g
i
m
a
g
e,
4
m
p
atc
h
es
w
i
ll b
e
ex
r
ac
ted
.
L
et
t
h
is
e
x
tr
ac
ted
p
atch
es b
e
d
en
o
ted
b
y
P
i,b
,
θ,
h
,
n
,
wh
er
e
i is
th
e
n
u
m
b
er
o
f
tr
ain
i
n
g
i
m
a
g
e,
b
is
t
h
e
n
u
m
b
er
o
f
th
e
b
a
n
d
,
θ
is
o
r
ien
tat
io
n
,
h
is
th
e
p
atc
h
s
ize
d
e
f
in
er
an
d
n
is
th
e
n
u
m
b
er
o
f
p
atch
.
T
h
ese
p
atch
es
w
ill
b
e
s
et
as
p
atch
s
.
T
h
e
in
p
u
ts
o
f
th
i
s
m
o
d
u
le
ar
e
C
1
b
,
θ
f
r
o
m
C
1
w
h
ic
h
f
o
r
ea
ch
b
an
d
an
d
o
r
ien
tatio
n
t
h
e
n
u
m
b
er
o
f
th
e
m
ar
e
Ntr
,
h
e
n
ce
4
m
Ntr
p
atch
s
f
o
r
ea
ch
b
a
n
d
an
d
o
r
ien
tatio
n
w
ill
b
e
p
r
o
d
u
ce
d
in
to
tal.
2
.
4
.
S2
M
o
du
le:
T
h
e
in
p
u
t
s
o
f
t
h
i
s
m
o
d
u
le
ar
e
tak
en
f
r
o
m
C
1
a
n
d
R
P
E
m
o
d
u
les.
T
h
is
m
o
d
u
le
ca
lc
u
late
s
t
h
e
te
m
p
late
m
atc
h
in
g
b
et
w
ee
n
C
1
b
,
θ
f
r
o
m
i
-
t
h
(
i=1
,
…,
Ntr
)
tr
ain
in
g
i
m
a
g
e
an
d
al
l
p
atch
s
i
n
b
an
d
b
an
d
o
r
ien
tatio
n
θ.
As
m
en
tio
n
ed
in
R
P
E
,
th
e
n
u
m
b
e
r
o
f
p
atch
s
in
b
an
d
b
an
d
o
r
ien
tatio
n
θ
ar
e
4
m
Ntr
.
Hen
ce
4
m
Ntr
m
atr
ice
s
w
ill
b
e
p
r
o
d
u
ce
d
f
o
r
ea
ch
b
an
d
an
d
o
r
ien
tatio
n
an
d
3
2
×4
m
Ntr
=
1
2
8
m
Ntr
m
a
tr
ices
f
o
r
i
-
th
tr
ai
n
in
g
i
m
a
g
e.
Fo
r
all
tr
ain
i
n
g
i
m
a
g
e,
S2
o
u
tp
u
t
i
s
a
ce
ll
w
i
th
Ntr
×1
2
8
m
Ntr
m
atr
ic
es.
I
n
th
i
s
m
o
d
u
le,
ea
ch
p
atch
P
i,b
,
θ,
h
,
n
is
s
lid
ed
acr
o
s
s
an
i
n
ter
m
ed
iate
o
u
tp
u
t
m
atr
i
x
C
1
b
,
θ,
an
d
th
e
te
m
p
lat
e
m
atc
h
i
n
g
i
s
ca
lc
u
lated
i
n
a
Gau
s
s
ia
n
-
l
ik
e
w
a
y
o
n
th
e
E
u
clid
ea
n
d
is
tan
ce
b
et
w
ee
n
t
h
e
lo
ca
l
C
1
b
,
θ
b
lo
ck
an
d
P
i,b
,
θ,
h
,
n
.
A
s
s
u
m
e
t
h
at
t
h
e
s
ize
o
f
t
h
e
p
atch
P
i,b
,
θ,
h
,
n
is
W
×W
.
L
et
th
e
W
×W
b
lo
ck
f
r
o
m
et
h
C
1
b
,
θ,
s
tar
tin
g
at
co
o
r
d
in
ate
(
p
,
q
)
b
e
d
en
o
ted
b
y
X.
T
h
e
o
u
tp
u
t S2
i,b
,
θ,
h
,
n
i
s
th
e
n
ca
lc
u
lated
as E
q
u
atio
n
4:
(
)
(
‖
‖
)
(
4
)
W
h
er
e
i
is
th
e
n
u
m
b
er
o
f
tr
ain
i
n
g
i
m
a
g
e
(
i=1
,
…,
Ntr
)
th
e
p
ar
am
eter
(
>0
)
d
ef
in
es
t
h
e
s
h
ar
p
n
es
s
o
f
th
e
ex
p
o
n
en
t
ial
f
u
n
ctio
n
a
n
d
‖
‖
is
a
n
E
u
clid
ia
n
n
o
r
m
.
2
.
5
.
C2
M
o
du
le:
T
h
e
in
p
u
t
s
o
f
t
h
is
m
o
d
u
le
ar
e
S2
o
u
tp
u
t
m
atr
ices
(
S2
i,b
,
θ,
h
,
n
)
.
I
n
th
is
m
o
d
u
le,
to
f
i
n
d
b
est
m
atc
h
i
n
g
,
th
e
m
a
x
i
m
u
m
o
f
t
h
ese
m
atr
ice
s
ar
e
co
m
p
u
ted
as E
q
u
atio
n
5.
Evaluation Warning : The document was created with Spire.PDF for Python.
I
n
d
o
n
esia
n
J
E
lec
E
n
g
&
C
o
m
p
Sci
I
SS
N:
2502
-
4752
O
b
ject
R
ec
o
g
n
itio
n
I
n
s
p
ir
in
g
HV
S
(
Mo
h
a
mma
d
esma
eil
A
kb
a
r
p
o
u
r
)
787
{
(
)
}
(
5
)
HM
AX
in
te
s
ti
n
g
s
tep
:
T
h
e
HM
A
X
test
i
n
g
m
o
d
el
i
s
v
er
y
s
i
m
ilar
to
t
h
e
tr
ain
i
n
g
m
o
d
el
ex
ce
p
t
th
at,
T
h
e
test
in
g
m
o
d
el
d
o
e
s
n
o
t
h
a
v
e
R
an
d
o
m
P
atch
E
x
tr
a
ctio
n
(
R
P
E
)
m
o
d
u
le.
As
te
s
ti
n
g
m
o
d
el
d
o
es
n
o
t
h
av
e
R
P
E
m
o
d
u
le,
S2
m
o
d
u
le
in
te
s
ti
n
g
m
o
d
el
u
s
es
th
e
s
to
r
ed
p
atch
ex
tr
ac
ted
i
n
tr
ai
n
i
n
g
m
o
d
el.
I
n
S2
m
o
d
u
le,
t
h
e
te
m
p
late
m
atc
h
in
g
i
s
ca
lcu
lated
i
n
a
Gau
s
s
ia
n
-
li
k
e
w
a
y
o
n
t
h
e
E
u
clid
ea
n
d
is
ta
n
ce
b
et
w
ee
n
t
h
e
C
1
b
,
θ
f
r
o
m
test
i
n
g
i
m
ag
e
s
a
n
d
s
to
r
ed
p
atch
es
in
tr
ain
i
n
g
m
o
d
el
(
P
i,
b
,
θ,
h
,
n
)
.
Hen
ce
in
th
e
s
a
m
e
w
a
y
,
S2
h
a
s
1
2
8
m
Ntr
m
atr
ices
f
o
r
ea
ch
test
i
n
g
i
m
a
g
e.
L
et
t
h
e
to
tal
n
u
m
b
er
o
f
test
in
g
i
m
a
g
e
s
b
e
d
en
o
ted
b
y
Nte
s
t.
T
h
e
o
u
tp
u
t
o
f
S2
m
o
d
u
le
w
ill
in
cl
u
d
e
is
a
ce
l
l
w
i
th
Nte
s
t×
1
2
8
m
Ntr
m
atr
ices.
C
2
ca
lc
u
lates
t
h
e
m
a
x
o
f
S2
o
u
tp
u
t
m
atr
ices.
Hen
ce
t
h
e
test
i
n
g
f
ea
t
u
r
es
w
il
l
b
e
a
m
at
r
ix
w
it
h
Ntes
t×1
2
8
m
Ntr
ele
m
en
ts
.
T
o
tal
task
s
o
f
HM
A
X
m
o
d
el
i
s
s
h
o
w
n
i
n
Fig
u
r
e
3.
Si
m
p
le
l
in
ea
r
cla
s
s
i
f
ier
:
Fo
r
i
m
a
g
e
clas
s
i
f
icatio
n
ap
p
licati
o
n
,
th
e
o
u
tp
u
t
i
m
a
g
e
f
ea
t
u
r
es
m
a
y
b
e
p
ass
ed
th
r
o
u
g
h
a
cla
s
s
i
f
ier
(
e.
g
.
SVM)
to
class
i
f
y
an
i
m
a
g
e.
Fig
u
r
e
3
.
T
o
tal
task
o
f
HM
A
X
m
o
d
el
[
5
]
3.
P
RO
P
O
SE
D
T
E
CH
NI
Q
U
E
T
h
e
b
asic
ar
ch
itectu
r
e
o
f
th
e
p
r
o
p
o
s
ed
m
o
d
el
h
as
s
ix
m
o
d
u
les
in
tr
ai
n
in
g
s
tep
an
d
f
o
u
r
m
o
d
u
les
i
n
test
i
n
g
s
tep
(
test
i
n
g
m
o
d
el
i
s
lik
e
HM
AX
tes
tin
g
m
o
d
el)
.
T
h
e
ar
ch
itectu
r
e
i
s
b
ased
o
n
HM
AX
m
o
d
el.
W
e
in
tr
o
d
u
ce
t
w
o
m
ai
n
co
n
tr
ib
u
ti
o
n
s
to
HM
A
X
m
o
d
el
t
h
at
i
m
p
r
o
v
e
th
e
r
ec
o
g
n
itio
n
r
ate
o
f
H
MA
X
m
o
d
el.
First
co
n
tr
ib
u
tio
n
is
eli
m
i
n
ati
n
g
o
f
b
ac
k
g
r
o
u
n
d
to
a
v
o
id
p
atch
e
x
tr
ac
tio
n
f
r
o
m
b
ac
k
g
r
o
u
n
d
th
at
th
er
e
i
s
n
o
ap
p
r
o
p
r
iate
in
f
o
r
m
a
tio
n
,
s
ec
o
n
d
co
n
tr
ib
u
t
io
n
i
s
u
s
i
n
g
o
p
ti
m
u
m
p
atc
h
f
o
r
HM
A
X
m
o
d
el
in
s
tead
o
f
r
an
d
o
m
p
atch
,
th
ir
d
co
n
tr
ib
u
tio
n
is
u
s
in
g
1
2
o
r
ien
tatio
n
s
i
n
s
tead
o
f
4
o
r
ien
tatio
n
s
(
th
e
t
h
ir
d
co
n
tr
ib
u
tio
n
h
a
v
e
d
o
n
e
b
ef
o
r
e)
.
T
h
e
p
r
o
p
o
s
ed
tr
ain
in
g
s
tep
is
d
i
f
f
er
en
t
a
n
d
h
as
s
ix
m
o
d
u
les
:
E
B
,
S1
,
C
1
,
OP
E
,
S
2
,
C
2
.
E
B
is
a
n
e
w
m
o
d
u
le,
OP
E
is
in
s
tead
o
f
R
P
E
,
S1
an
d
C
1
a
r
e
b
ased
o
n
H
MA
X
m
o
d
el
b
u
t
th
e
y
ar
e
s
li
g
h
tl
y
d
i
f
f
er
en
t,
f
i
n
all
y
Evaluation Warning : The document was created with Spire.PDF for Python.
I
SS
N
:
2
5
0
2
-
4752
I
n
d
o
n
esia
n
J
E
lec
E
n
g
&
C
o
m
p
Sci,
Vo
l.
12
,
No
.
2
,
No
v
e
m
b
er
2
0
1
8
:
7
8
3
–
7
9
3
788
o
th
er
m
o
d
u
le
s
(
S2
,
C
2
)
ar
e
a
s
th
e
s
a
m
e
as
th
e
H
M
A
X
m
o
d
u
le
s
.
I
n
th
e
f
o
llo
w
i
n
g
w
e
ex
p
lain
o
u
r
m
o
d
u
les
s
ep
ar
atel
y
:
3
.
1
.
E
B
:
I
n
HM
A
X
w
e
e
x
tr
ac
ted
R
an
d
o
m
p
atc
h
es
f
r
o
m
i
m
a
g
e,
f
ir
s
t
s
tep
f
o
r
e
x
tr
ac
ti
n
g
o
p
ti
m
u
m
p
atc
h
i
s
eli
m
i
n
ati
n
g
b
ac
k
g
r
o
u
n
d
.
W
h
en
o
u
r
s
y
s
te
m
u
s
e
p
atch
e
s
f
r
o
m
b
ac
k
g
r
o
u
n
d
th
er
e
w
i
ll
b
e
t
w
o
i
m
p
o
r
tan
t
p
r
o
b
lem
s
:
f
ir
s
t
w
e
w
ill
f
o
r
ce
s
y
s
te
m
to
d
o
ad
d
itio
n
al
co
m
p
u
t
atio
n
al
w
o
r
k
s
ec
o
n
d
w
e
w
i
ll
d
ir
ec
t
o
u
r
s
y
s
te
m
to
th
e
w
r
o
n
g
r
es
u
lt,
b
ec
au
s
e
o
f
o
u
r
w
r
o
n
g
i
n
f
o
r
m
atio
n
.
A
n
e
x
a
m
p
le
o
f
eli
m
i
n
ati
n
g
b
ac
k
g
r
o
u
n
d
o
n
air
p
lan
e
f
r
o
m
C
AL
T
E
C
H1
0
1
d
ataset
is
s
h
o
w
n
i
n
Fi
g
u
r
e
4
.
I
n
t
h
is
m
o
d
u
le
w
e
f
ir
s
t,
el
i
m
in
ate
s
o
m
e
r
o
w
s
o
f
p
ict
u
r
es
f
r
o
m
u
p
an
d
d
o
w
n
o
f
it,
b
ec
au
s
e
al
m
o
s
t
t
h
er
e
is
n
o
i
m
ag
e
in
f
o
r
m
a
ti
o
n
in
th
e
s
e
p
lace
s
b
u
t
t
h
er
e
is
s
o
m
e
i
n
f
o
r
m
atio
n
lik
e
ex
p
lai
n
ab
o
u
t
i
m
a
g
es
t
h
at
th
e
y
ar
e
n
o
t
u
s
e
f
u
l.
T
h
en
w
e
u
s
e
eli
m
in
a
te
b
ac
k
g
r
o
u
n
d
alg
o
r
ith
m
[
2
6
]
.
A
f
ter
u
s
i
n
g
eli
m
i
n
ate
alg
o
r
it
h
m
w
e
w
il
l
e
x
tr
ac
t
f
o
u
r
p
o
in
ts
:
m
i
n
i
m
u
m
r
o
w
,
m
ax
i
m
u
m
r
o
w
,
m
i
n
i
m
u
m
co
lu
m
n
a
n
d
m
ax
i
m
u
m
co
lu
m
n
.
W
e
ex
tr
a
cted
w
i
n
d
o
w
i
n
g
i
m
ag
e
w
i
th
r
o
w
s
f
r
o
m
m
i
n
i
m
u
m
r
o
w
to
m
ax
i
m
u
m
r
o
w
a
n
d
co
lu
m
n
f
r
o
m
m
in
i
m
u
m
co
lu
m
n
to
m
a
x
i
m
u
m
co
lu
m
n
.
Fig
u
r
e
4
.
C
A
L
T
E
C
1
0
1
air
p
lan
e
i
m
ag
e
a
f
ter
eli
m
i
n
ati
n
g
b
ac
h
g
r
o
u
n
d
3
.
2
.
S1
:
I
n
th
i
s
m
o
d
u
le
w
e
u
s
e
Gab
o
r
f
ilter
w
i
th
T
ab
le
1
.
P
a
r
am
eter
s
,
lik
e
o
r
ig
i
n
al
HM
AX
m
o
d
el.
B
u
t
w
e
u
s
e
Gab
o
r
f
ilter
s
in
1
2
o
r
ien
tatio
n
as E
q
u
atio
n
6.
*
+
(
6
)
3
.
3
.
C1
:
I
n
o
r
ig
in
al
HM
AX
m
o
d
el
th
e
p
atch
es
ar
e
ex
tr
ac
ted
f
r
o
m
C
1
m
o
d
u
le
ac
r
o
s
s
all
f
o
u
r
o
r
ien
tatio
n
s
,
p
atch
es
s
ize
s
ar
e
n
×n
(
n
=4
,
8
,
1
2
,
1
6
)
.
I
n
o
u
r
m
o
d
el
w
e
e
x
tr
a
cted
p
atch
es
w
ith
o
n
e
s
ize
f
o
r
ev
er
y
b
an
d
.
As
w
e
ex
p
lain
b
e
f
o
r
e,
w
e
g
et
m
a
x
w
i
th
w
i
n
d
o
w
s
f
r
o
m
T
ab
le
1
.
So
t
h
e
o
u
tp
u
t
o
f
C
1
m
o
d
u
le
is
d
if
f
er
en
t,
f
o
r
ex
a
m
p
le
th
e
s
ize
o
f
o
u
tp
u
t
o
f
C
1
m
o
d
u
le
f
o
r
i
m
ag
e
w
it
h
s
ize
o
f
1
4
0
×1
4
0
is
f
r
o
m
3
0
×3
0
to
1
0
×1
0
.
T
h
is
s
h
o
u
ld
b
e
m
en
tio
n
ed
t
h
at
i
f
a
f
ter
eli
m
i
n
atin
g
b
ac
k
g
r
o
u
n
d
,
i
f
t
h
e
s
ize
o
f
o
u
tp
u
t
i
s
s
m
all
w
e
s
h
o
u
ld
r
esize
i
t
to
p
r
ep
ar
e
s
ize
(
i.e
,
1
4
0
×
1
4
0
)
.
3
.
4
.
O
P
E
:
I
n
HM
A
X
m
o
d
el
p
atch
e
s
ar
e
ex
tr
ac
ted
r
an
d
o
m
l
y
.
I
n
o
u
r
m
o
d
el,
f
ir
s
t
b
ac
k
g
r
o
u
n
d
s
ar
e
eli
m
i
n
ated
,
th
en
a
m
o
n
g
t
h
e
i
m
ag
e
t
h
e
o
p
tim
u
m
p
atc
h
es a
r
e
ex
tr
ac
ted
.
W
e
ex
tr
ac
ted
o
p
tim
u
m
e
in
t
w
o
s
tep
s
:
Step
1
:
As
HV
S
is
s
e
n
s
it
iv
e
t
o
ed
g
es
an
d
li
n
es,
s
o
it
s
ee
m
s
t
h
at
o
p
ti
m
u
m
p
atc
h
es
s
h
o
u
ld
b
e
i
n
v
o
lv
e
m
o
r
e
ed
g
es
an
d
li
n
es.
He
n
ce
f
o
r
o
u
r
m
o
d
el,
p
atch
es
w
i
th
m
o
r
e
ed
g
es
an
d
li
n
es
ar
e
ex
tr
ac
ted
i
n
s
tead
o
f
t
h
e
d
en
s
e
in
p
u
t
s
a
n
d
b
li
n
d
p
atch
s
elec
ti
o
n
i
n
HM
AX
m
o
d
el.
A
s
s
ee
n
i
n
t
h
e
Alg
o
r
it
h
m
1
,
ch
o
o
s
i
n
g
o
p
ti
m
u
m
p
atc
h
i
s
b
ased
o
n
ch
o
o
s
in
g
m
m
r
an
d
o
m
p
atch
a
n
d
ar
r
an
g
e
t
h
e
m
b
a
s
ed
o
n
v
ar
ian
ce
.
Su
p
p
o
s
e
t
h
e
n
u
m
b
er
o
f
o
p
ti
m
u
m
p
atch
is
m
a
n
d
in
th
i
s
alg
o
r
it
h
m
w
e
f
i
n
d
less
o
p
ti
m
u
m
p
at
ch
th
a
n
m
.
So
it
is
t
h
e
p
r
o
b
le
m
o
f
th
is
al
g
o
r
ith
m
.
Fo
r
s
o
lv
i
n
g
t
h
i
s
p
r
o
b
lem
,
f
ir
s
t
w
e
s
elec
t
m
r
an
d
o
m
p
atc
h
es
an
d
s
av
e
th
eir
v
al
u
e
as
a
n
ar
r
ay
.
W
e
ar
r
an
g
e
t
h
e
m
ac
co
r
d
in
g
to
th
eir
v
ar
ia
n
ce
f
r
o
m
w
o
r
s
t
to
b
est.
W
e
r
ep
ea
t
p
atch
s
elec
tio
n
i
n
r
an
d
o
m
m
o
d
e
f
o
r
m
m
ti
m
es.
I
f
v
ar
ian
ce
o
f
n
e
w
p
atc
h
i
s
m
o
r
e
th
a
n
t
h
e
v
ar
ian
ce
o
f
th
e
b
est
r
a
n
d
o
m
it
w
ill
b
e
o
n
e
s
tep
f
o
r
th
i
s
al
g
o
r
ith
m
.
W
e
w
il
l
s
h
i
f
t
all
p
atch
e
s
an
d
r
ep
lace
th
e
b
est
p
atch
w
i
th
n
e
w
p
at
ch
.
I
n
t
h
is
s
tep
w
e
w
ill
e
x
tr
ac
t
p
atch
es
w
it
h
m
o
r
e
in
f
o
r
m
atio
n
(
w
i
th
m
o
r
e
lin
e
s
an
d
ed
g
es).
Evaluation Warning : The document was created with Spire.PDF for Python.
I
n
d
o
n
esia
n
J
E
lec
E
n
g
&
C
o
m
p
Sci
I
SS
N:
2502
-
4752
O
b
ject
R
ec
o
g
n
itio
n
I
n
s
p
ir
in
g
HV
S
(
Mo
h
a
mma
d
esma
eil
A
kb
a
r
p
o
u
r
)
789
Alg
o
rit
h
m
1
ch
o
o
s
i
n
g
o
p
ti
m
u
m
p
atch
1)
E
xtra
ct
m
r
a
n
d
o
m
p
a
tch
es
(
like
HMAX
)
,
a
r
r
a
n
g
e
th
em
fr
o
m
w
o
r
s
t
to
b
est
a
n
d
s
a
ve
th
e
m
a
s
m
b
est p
a
tch
es.
2)
R
ep
ea
t ra
n
d
o
m
p
a
tch
ex
tr
a
ctio
n
fo
r
mm
times,
I
f v
a
r
(
n
ew p
a
tch
)
>va
r
(
o
ld
p
a
tch
)
th
en
:
(
I
)
s
h
ift a
ll sa
ve
d
p
a
tch
es fr
o
m
b
est to
w
o
r
s
t.
(
I
I
)
r
ep
la
ce
th
e
b
est p
a
tch
w
ith
n
ew p
a
tch
.
W
e
s
u
r
e
th
a
t th
ere
w
ill b
e
m
b
est p
a
tch
es in
(
m+mm
)
iter
a
ti
o
n
s
.
3)
Th
ese
m
p
a
tch
es
a
r
e
in
p
u
t
o
f
f(
)
a
cc
o
r
d
in
g
to
E
q
u
a
tio
n
8
.
a
n
d
w
e
w
ill
fin
d
th
e
min
imu
m
o
f
th
is
fu
n
ctio
n
.
Th
en
w
ill b
e
th
e
b
est p
a
tch
es.
Step
2
:
I
n
th
i
s
s
tep
w
e
w
ill e
x
tr
ac
t p
at
ch
es
w
i
th
th
e
b
es
t p
er
f
o
r
m
a
n
c
e,
a
m
o
n
g
t
h
e
p
atch
e
s
t
h
at
w
er
e
ex
tr
ac
ted
in
s
tep
1
.
HVS
i
n
clas
s
i
f
icatio
n
f
in
d
s
a
p
atch
e
s
t
h
at
ar
e
t
h
e
s
a
m
e
in
o
n
e
ca
te
g
o
r
y
a
n
d
h
a
s
a
m
o
s
t
d
if
f
er
en
ce
w
it
h
o
th
er
ca
te
g
o
r
y
.
So
w
e
w
il
l
ex
tr
ac
t
p
atch
e
s
th
at
h
av
e
m
o
s
t
s
i
m
ila
r
it
y
w
it
h
o
th
er
p
atch
es
i
n
th
e
s
a
m
e
ca
teg
o
r
y
a
n
d
in
o
th
er
h
a
n
d
h
av
e
t
h
e
m
o
s
t
d
i
f
f
er
en
ce
w
i
th
p
atch
es
i
n
o
t
h
er
ca
teg
o
r
ies.
F
o
r
d
o
in
g
t
h
is
w
it
h
d
ef
in
e
a
f
o
r
m
u
la
li
k
e
E
q
u
atio
n
8
.
W
ith
th
e
b
es
t
p
atch
e
s
t
h
is
f
o
r
m
u
la
w
i
ll
b
e
m
i
n
i
m
u
m
.
So
in
s
i
m
u
la
tio
n
i
f
w
e
f
i
n
d
m
i
n
i
m
u
m
o
f
t
h
is
f
o
r
m
u
l
a,
th
e
p
atch
e
s
w
ill
b
e
t
h
e
b
es
t
p
atch
es
t
h
at
h
av
e
th
e
m
a
x
i
m
u
m
s
i
m
ilar
it
y
w
it
h
o
th
er
p
atch
es
i
n
t
h
e
s
a
m
e
ca
t
eg
o
r
y
a
n
d
h
a
v
e
m
o
s
t
d
if
f
er
e
n
ce
w
ith
o
th
er
p
atc
h
es
in
o
th
er
ca
teg
o
r
ies.
W
e
u
s
e
eq
u
atio
n
f
o
r
ea
c
h
b
a
n
d
an
d
o
r
ien
tatio
n
s
ep
ar
atel
y
.
W
e
s
h
o
w
th
e
clas
s
n
u
m
b
er
o
f
p
atc
h
w
i
t
h
i
an
d
th
e
j
is
th
e
n
u
m
b
er
o
f
tr
ain
in
g
i
m
ag
e
t
h
at
p
atch
is
ex
tr
ac
ted
(
P
i,
j
)
.
T
h
en
th
e
f
u
n
ctio
n
d
e
f
in
itio
n
i
s
as E
q
u
atio
n
7
:
(
)
∑
∑
∑
‖
‖
∑
∑
∑
‖
‖
(
7
)
(
)
(
)
W
h
er
e
C
N
in
t
h
e
n
u
m
b
er
o
f
c
lass
an
d
T
N
is
th
e
n
u
m
b
er
o
f
T
r
ain
in
g
i
m
ag
e
i
n
ea
ch
cla
s
s
.
T
h
e
to
tal
n
u
m
b
er
o
f
p
air
p
atc
h
es
in
ea
c
h
cla
s
s
ar
e
T
N(
T
N
-
1
)
,
s
o
f
o
r
C
N
cla
s
s
w
e
w
ill
h
a
v
e
T
N(
T
N
-
1
)
×CN p
air
p
atc
h
es
in
s
a
m
e
ca
te
g
o
r
y
.
I
n
t
h
e
s
a
m
e
w
a
y
w
e
w
ill
h
av
e
C
N(
C
N
-
1
)
×T
N
p
air
p
atch
es
in
d
i
f
f
er
en
t
ca
teg
o
r
y
.
I
t
is
i
m
p
o
r
tan
t
th
a
t
v
al
u
e
o
f
p
air
s
b
e
eq
u
al
s
o
w
e
m
u
ltip
le
t
h
e
r
esu
lt
w
it
h
s
ec
o
n
d
ter
m
.
T
h
is
m
o
d
u
le
ac
tiv
it
y
i
s
ex
p
lain
ed
i
n
Alg
o
r
it
h
m
1
.
A
s
ex
p
lain
ed
b
ef
o
r
e,
in
HM
A
X
m
o
d
el
p
atch
es
ar
e
ex
tr
ac
ted
r
an
d
o
m
l
y
i
n
R
P
E
m
o
d
u
le,
w
h
ic
h
m
a
y
n
o
t
b
e
o
p
ti
m
al.
T
h
e
ex
t
r
ac
ted
p
atch
es
b
y
t
h
e
HM
A
X
m
o
d
el
ar
e
f
r
o
m
r
a
n
d
o
m
p
o
s
itio
n
.
So
,
th
ese
p
atch
e
s
m
a
y
co
m
e
f
r
o
m
b
ac
k
g
r
o
u
n
d
o
r
o
t
h
er
ir
r
el
ev
an
t
o
b
j
ec
ts
r
ath
er
t
h
an
t
h
e
t
ar
g
et
o
b
j
ec
t.
I
n
th
e
o
th
er
h
an
d
th
e
y
m
a
y
co
m
e
f
r
o
m
tar
g
et
o
b
j
ec
t
b
u
t
all
th
e
p
ix
els
ar
e
th
e
s
a
m
e,
s
o
th
er
e
is
n
o
in
f
o
r
m
atio
n
in
it.
T
h
e
C
2
f
ea
t
u
r
es
th
at
ar
e
o
b
tain
ed
f
r
o
m
t
h
ese
p
atch
es
ar
e
n
o
t
v
er
y
u
s
e
f
u
l
f
o
r
r
ec
o
g
n
izi
n
g
a
tar
g
et
o
b
j
ec
t.
T
h
e
y
m
a
y
also
m
a
k
e
t
h
e
f
ea
t
u
r
e
s
p
ac
e
m
o
r
e
co
m
p
le
x
f
o
r
clas
s
i
f
i
ca
tio
n
[
2
2
]
.
T
h
e
p
r
o
p
o
s
ed
m
o
d
u
le
(
OP
E
)
s
elec
t
s
p
atch
es
w
it
h
m
o
r
e
i
n
f
o
r
m
ati
o
n
b
y
e
m
p
lo
y
in
g
o
p
ti
m
u
m
p
atch
s
elec
tio
n
m
et
h
o
d
w
h
ic
h
is
e
x
p
lain
ed
i
n
alg
o
r
ith
m
1
.
As
a
r
es
u
lt,
le
s
s
p
atch
es
i
n
tr
ai
n
i
n
g
s
tep
i
n
o
u
r
m
o
d
el
h
a
v
e
t
h
e
s
a
m
e
r
e
s
u
l
t
in
tr
ai
n
i
n
g
s
tep
i
n
HM
AX
m
o
d
el
th
a
t is
g
o
o
d
f
o
r
s
y
s
te
m
s
p
ee
d
.
3
.
5
.
S2
a
nd
C2
:
T
w
o
m
o
d
u
les S2
an
d
C
2
ar
e
a
s
th
e
s
a
m
e
as o
r
ig
in
al
H
M
A
X
m
o
d
el
a
n
d
w
e
h
a
v
e
ex
p
lai
n
ed
it b
ef
o
r
e.
4.
E
XP
E
R
I
M
E
NT
A
L
RE
SUL
T
S
W
e
h
av
e
test
ed
o
u
r
m
o
d
el
o
n
th
e
C
alT
ec
h
5
an
d
C
alT
ec
h
1
0
1
d
ataset
o
f
i
m
a
g
es.
W
e
h
a
v
e
co
m
p
ar
ed
th
e
p
r
o
p
o
s
ed
m
o
d
el
(
DP
HM
A
X)
w
it
h
SIFT
[
?]
,
HM
A
X
[
?
]
,
GM
A
X
[
]
an
d
MH
M
A
X
m
o
d
el
s
.
Ou
r
r
esu
lt
s
s
h
o
w
th
at
t
h
er
e
ar
e
s
ig
n
if
ica
n
t
i
m
p
r
o
v
e
m
e
n
ts
to
class
i
f
ic
atio
n
in
o
u
r
m
o
d
el.
T
h
e
p
r
o
p
o
s
ed
alg
o
r
ith
m
i
s
i
m
p
le
m
en
ted
w
it
h
M
A
T
L
A
B
.
4
.
1
.
I
m
a
g
e
Da
t
a
s
et
s
T
o
ev
alu
ate
o
u
r
m
o
d
el
o
n
cla
s
s
if
ica
tio
n
ta
s
k
s
,
w
e
u
s
e
C
alT
ec
h
5
,
C
altec
h
1
0
1
an
d
GR
A
Z
-
0
1
d
atasets
.
C
alT
ec
h
5
:
T
h
is
d
ataset
[
26
]
c
o
n
tain
s
f
i
v
e
c
las
s
es
o
f
o
b
j
ec
ts
:
th
e
f
r
o
n
ta
l
-
f
ac
e,
m
o
to
r
c
y
c
le,
r
ea
r
-
ca
r
,
air
p
lan
e
an
d
leaf
.
T
h
e
s
a
m
p
le
i
m
a
g
es
o
f
ea
ch
ca
te
g
o
r
y
is
s
h
o
w
n
i
n
Fig
u
r
e
5
.
C
alT
ec
h
1
0
1
:
T
h
is
d
ataset
[
27
]
co
n
tain
s
1
0
1
class
es
o
f
o
b
j
ec
ts
s
u
c
h
as
:
b
o
at,
ca
r
-
s
id
e,
b
ik
e,
air
p
lan
e
,
etc.
T
h
e
s
a
m
p
le
i
m
a
g
e
s
ar
e
s
h
o
w
n
in
Fi
g
u
r
e
6
.
GR
AZ
-
0
1
:
T
h
i
s
d
ataset
[
28
]
co
n
tain
s
m
a
n
y
d
i
f
f
er
en
t
o
b
j
ec
t
lik
e:
b
ik
e
s
,
p
eo
p
le,
p
lan
t,
b
u
ild
in
g
,
s
h
o
es,
etc.
Evaluation Warning : The document was created with Spire.PDF for Python.
I
SS
N
:
2
5
0
2
-
4752
I
n
d
o
n
esia
n
J
E
lec
E
n
g
&
C
o
m
p
Sci,
Vo
l.
12
,
No
.
2
,
No
v
e
m
b
er
2
0
1
8
:
7
8
3
–
7
9
3
790
p
eo
p
le
an
d
b
ik
e
co
n
s
id
er
ed
as
a
p
o
s
itiv
e
i
m
a
g
es
a
n
d
o
th
er
i
m
ag
e
s
ar
e
as
b
ac
k
g
r
o
u
n
d
.
T
h
e
s
a
m
p
le
i
m
a
g
es
ar
e
s
h
o
w
n
in
F
ig
u
r
e
7
.
Fig
u
r
e
5
.
Sa
m
p
le
i
m
ag
e
s
f
r
o
m
C
altec
h
5
d
atab
ase.
Fro
m
le
f
t t
o
r
ig
h
t:
m
o
to
r
b
ik
e,
le
af
a
n
d
b
ac
k
g
r
o
u
n
d
Fig
u
r
e
6
.
Sa
m
p
le
i
m
ag
e
s
f
r
o
m
C
altec
h
1
0
1
.
Fro
m
lef
t to
r
ig
h
t
: a
i
r
p
lan
e,
ca
r
-
s
id
e
an
d
b
ac
k
g
r
o
u
n
d
Fig
u
r
e
7
.
Sa
m
p
le
i
m
ag
e
s
f
r
o
m
GR
AZ
-
0
1
.
Fro
m
le
f
t to
r
ig
h
t,
th
e
clas
s
es a
r
e
b
ik
es,
p
eo
p
le
an
d
b
ac
k
g
r
o
u
n
d
s
4
.
2
.
P
er
f
o
r
m
a
nce
M
ea
s
ures
Af
ter
clas
s
i
f
icatio
n
,
th
e
i
m
a
g
e
s
ca
teg
o
r
ize
i
n
4
g
r
o
u
p
:
T
r
u
e
P
o
s
itiv
e
(
T
P)
is
a
co
r
r
ec
t
class
if
ica
tio
n
o
f
a
p
o
s
itiv
e
(
o
b
j
ec
t)
,
a
T
r
u
e
Neg
ativ
e
(
T
N)
is
a
co
r
r
ec
t
class
i
f
icatio
n
o
f
a
n
e
g
ati
v
e
(
b
ac
k
g
r
o
u
n
d
)
,
Fals
e
P
o
s
itiv
e
(
FP
)
is
an
i
n
co
r
r
ec
t
o
b
j
ec
t
class
if
icatio
n
a
n
d
Fa
ls
e
Neg
a
tiv
e
(
FN)
is
a
n
i
n
c
o
r
r
ec
t
b
ac
k
g
r
o
u
n
d
class
i
f
icatio
n
.
W
e
ch
o
s
e
th
e
e
v
alu
a
tio
n
m
etr
ics o
f
class
if
ica
t
io
n
r
ate
d
ef
in
ed
as E
q
u
at
io
n
8
:
Evaluation Warning : The document was created with Spire.PDF for Python.
I
n
d
o
n
esia
n
J
E
lec
E
n
g
&
C
o
m
p
Sci
I
SS
N:
2502
-
4752
O
b
ject
R
ec
o
g
n
itio
n
I
n
s
p
ir
in
g
HV
S
(
Mo
h
a
mma
d
esma
eil
A
kb
a
r
p
o
u
r
)
791
(
)
(
8
)
4
.
3
.
Cla
s
s
if
ica
t
io
n r
esu
lt
s
W
e
r
an
d
o
m
l
y
c
h
o
s
e
1
0
,
2
0
,
3
0
,
4
0
an
d
5
0
im
a
g
es
f
r
o
m
e
ac
h
ca
teg
o
r
y
o
f
th
e
d
ataset
s
as
tr
ain
i
n
g
i
m
a
g
es
,
o
th
er
i
m
a
g
es
ar
e
co
n
s
id
er
ed
as
test
in
g
i
m
a
g
es.
f
o
r
C
altec
h
5
an
d
C
altec
h
1
0
1
d
a
tasets
,
o
b
j
ec
ts
ar
e
p
o
s
itiv
e
i
m
ag
e
s
a
n
d
b
ac
k
g
r
o
u
n
d
is
n
eg
ati
v
e
i
m
a
g
es.
f
o
r
GR
A
-
0
1
d
ataset,
p
eo
p
le
an
d
b
ik
e
ar
e
p
o
s
itiv
e
i
m
ag
e
s
an
d
o
th
er
i
m
a
g
es
ar
e
n
eg
at
iv
e
i
m
a
g
es
(
w
e
m
a
k
e
a
n
e
g
ati
v
e
i
m
a
g
es
f
r
o
m
co
m
b
in
i
n
g
o
t
h
er
ca
te
g
o
r
ies).
A
ll
r
esu
lt
s
r
ep
o
r
ted
w
er
e
g
e
n
er
ate
d
w
it
h
1
0
r
an
d
o
m
s
p
lit
s
.
T
h
e
r
esu
lt
s
ar
e
s
h
o
w
n
i
n
Fi
g
u
r
e
8
.
A
s
s
h
o
w
n
,
ex
ce
p
t
in
ca
r
-
s
id
e
f
r
o
m
C
al
tech
1
0
1
,
th
e
p
r
o
p
o
s
ed
m
et
h
o
d
p
r
o
v
id
es
a
s
ig
n
i
f
ica
n
t
p
er
f
o
r
m
an
ce
i
m
p
r
o
v
e
m
en
t
o
v
er
o
th
er
m
et
h
o
d
s
.
Fig
u
r
e
8
(
a)
.
C
lass
i
f
icatio
n
ac
cu
r
ac
y
f
o
r
ea
ch
class
:
C
altec
h
5
d
ataset
class
i
f
icat
io
n
ac
cu
r
ac
y
(
m
o
to
r
an
d
f
ac
e)
Fig
u
r
e
8
(
b
)
.
C
lass
if
icatio
n
ac
c
u
r
ac
y
f
o
r
ea
ch
class
:
C
altec
h
1
0
1
d
ataset
class
if
ica
tio
n
ac
c
u
r
ac
y
(
ca
r
-
s
id
e
an
d
air
p
lan
e)
60
65
70
75
80
85
90
95
100
10
20
30
40
50
C
l
a
ssi
f
i
c
a
t
i
o
n
a
c
c
u
r
a
c
y
N
u
mb
e
r
o
f
t
r
a
i
n
Mo
to
r
SI
FT
H
M
AX
M
H
M
A
X
G
M
AX
D
PHM
AX
60
65
70
75
80
85
90
95
100
10
20
30
40
50
C
l
a
ssi
f
i
c
a
t
i
o
n
a
c
c
u
r
a
c
y
N
u
mb
e
r
o
f
t
r
a
i
n
Face
s
SI
FT
H
M
AX
M
H
M
AX
G
M
AX
D
PHM
A
X
60
65
70
75
80
85
90
95
100
10
20
30
40
50
C
l
a
ssi
f
i
c
a
t
i
o
n
a
c
c
u
r
a
c
y
N
u
mb
e
r
o
f
t
r
a
i
n
C
ar
_
s
id
e
SI
FT
H
M
AX
M
H
M
AX
G
M
AX
D
PHM
A
X
60
65
70
75
80
85
90
95
100
10
20
30
40
50
C
l
a
ssi
f
i
c
a
t
i
o
n
a
c
c
u
r
a
c
y
N
u
mb
e
r
o
f
t
r
a
i
n
A
ir
p
lan
e
SI
FT
H
M
AX
M
H
M
AX
G
M
AX
D
PHM
A
X
Evaluation Warning : The document was created with Spire.PDF for Python.
I
SS
N
:
2
5
0
2
-
4752
I
n
d
o
n
esia
n
J
E
lec
E
n
g
&
C
o
m
p
Sci,
Vo
l.
12
,
No
.
2
,
No
v
e
m
b
er
2
0
1
8
:
7
8
3
–
7
9
3
7
92
F
ig
u
r
e
8
(
c)
.
C
lass
i
f
icatio
n
ac
cu
r
ac
y
f
o
r
ea
ch
class
:
GR
AZ
-
0
1
d
ataset
class
i
f
icatio
n
ac
c
u
r
a
c
y
(
p
eo
p
le
an
d
b
ik
e)
5
.
CO
NCLUS
I
O
N
I
n
t
h
is
p
ap
er
a
n
e
w
f
r
a
m
e
w
o
r
k
f
o
r
r
o
b
u
s
t
o
b
j
ec
t
r
ec
o
g
n
itio
n
b
ased
o
n
p
r
ev
io
u
s
h
ier
ar
c
h
ica
l
m
o
d
el
is
p
r
o
p
o
s
ed
.
A
n
o
v
el
De
s
ir
ab
le
P
atch
ex
tr
ac
tio
n
m
e
th
o
d
f
o
r
HM
A
X
h
ier
ar
ch
ical
m
o
d
el
(
OP
HM
A
X)
is
p
r
o
p
o
s
ed
.
Fo
r
th
is
p
u
r
p
o
s
e,
f
ir
s
t
w
e
eli
m
in
a
te
th
e
b
ac
k
g
r
o
u
n
d
,
th
en
w
e
ex
tr
ac
ted
th
e
b
est p
atch
es t
h
at
t
h
e
y
ar
e
th
e
s
a
m
e
in
t
h
e
s
a
m
e
ca
te
g
o
r
y
an
d
t
h
e
y
ar
e
m
o
r
e
d
i
f
f
er
e
n
t i
n
th
e
d
i
f
f
er
en
t
ca
teg
o
r
y
.
T
h
is
i
m
p
r
o
v
e
m
e
n
t
is
o
n
l
y
in
tr
ai
n
i
n
g
s
tag
e
s
o
it
d
o
n
o
t
n
ee
d
m
o
r
e
ti
m
e
i
n
r
ec
o
g
n
itio
n
s
tag
e.
Op
ti
m
u
m
p
atc
h
es
ar
e
th
e
b
est
p
atc
h
es
to
r
ep
r
esen
t
d
is
cr
i
m
i
n
ati
v
e
an
d
i
n
v
ar
ia
n
t
f
ea
tu
r
es.
T
h
e
y
s
o
l
v
e
th
e
m
ai
n
li
m
itat
io
n
in
t
h
e
H
A
X
“
d
en
s
e
in
p
u
ts
an
d
b
lin
d
f
ea
t
u
r
e
s
elec
tio
n
’
’
.
E
x
p
er
i
m
e
n
ts
b
ased
o
n
th
r
ee
d
if
f
er
en
t
k
in
d
s
o
f
d
atasets
h
a
v
e
d
em
o
n
s
tr
ated
th
a
t
th
e
DP
HM
A
X
p
er
f
o
r
m
s
m
u
c
h
b
etter
th
a
n
t
h
e
o
th
er
m
et
h
o
d
s
in
v
ar
io
u
s
v
is
u
al
r
ec
o
g
n
itio
n
t
ask
s
.
RE
F
E
R
E
NC
E
[1
]
H.
Yin
,
G
.
Ly
u
,
X
.
L
u
o
,
a
n
d
C.
L
i,
"
A
s
e
g
m
e
n
tatio
n
-
b
a
se
d
a
d
a
p
ti
v
e
i
m
a
g
e
e
n
h
a
n
c
e
m
e
n
t
m
e
th
o
d
i
n
sp
ired
b
y
th
e
se
lf
-
a
d
ju
st
f
e
a
tu
re
s
o
f
HV
S
,
"
In
ter
n
a
ti
o
n
a
l
J
o
u
rn
a
l
o
f
M
a
c
h
in
e
L
e
a
rn
in
g
a
n
d
Cy
b
e
rn
e
ti
c
s,
v
o
l.
8
,
p
p
.
1
8
9
5
-
1
9
0
5
,
2
0
1
7
.
[2
]
R.
Ku
m
a
r,
S
.
Ch
a
n
d
,
a
n
d
S
.
S
in
g
h
,
"
A
n
I
m
p
ro
v
e
d
Histo
g
ra
m
-
S
h
if
ti
n
g
-
I
m
it
a
ted
re
v
e
rsib
le
d
a
ta
h
i
d
in
g
b
a
se
d
o
n
HV
S
c
h
a
ra
c
teristics
,
"
M
u
lt
ime
d
ia
T
o
o
ls
a
n
d
Ap
p
li
c
a
t
io
n
s,
p
p
.
1
-
1
3
,
2
0
1
7
.
[3
]
S
.
S
h
a
f
e
e
a
n
d
B.
Ra
jae
i,
"
A
se
c
u
re
ste
g
a
n
o
g
ra
p
h
y
a
lg
o
rit
h
m
u
sin
g
c
o
m
p
re
ss
iv
e
se
n
sin
g
b
a
s
e
d
o
n
HV
S
f
e
a
tu
re
,
"
in
Eme
rg
in
g
S
e
c
u
rity T
e
c
h
n
o
l
o
g
ies
(
ES
T
),
2
0
1
7
S
e
v
e
n
t
h
I
n
ter
n
a
ti
o
n
a
l
Co
n
fer
e
n
c
e
o
n
,
2
0
1
7
,
p
p
.
7
4
-
7
8
.
[4
]
V
.
Ra
d
h
ik
a
,
K.
V
e
e
ra
sw
a
m
y
,
a
n
d
S
.
S
.
K
u
m
a
r,
"
Di
g
it
a
l
Im
a
g
e
F
u
sio
n
Us
in
g
HV
S
i
n
Blo
c
k
Ba
se
d
T
ra
n
sf
o
r
m
s,"
J
o
u
rn
a
l
o
f
S
i
g
n
a
l
Pro
c
e
ss
in
g
S
y
ste
ms
,
p
p
.
1
-
1
1
,
2
0
1
7
.
[5
]
T
.
S
e
rre
,
L
.
W
o
lf
,
S
.
Bil
e
sc
h
i,
M
.
Ries
e
n
h
u
b
e
r,
a
n
d
T
.
P
o
g
g
io
,
"
Ro
b
u
st
o
b
jec
t
re
c
o
g
n
it
i
o
n
w
it
h
c
o
rtex
-
li
k
e
m
e
c
h
a
n
is
m
s,"
IEE
E
tra
n
sa
c
ti
o
n
s
o
n
p
a
t
ter
n
a
n
a
lys
is
a
n
d
ma
c
h
in
e
i
n
telli
g
e
n
c
e
,
v
o
l.
2
9
,
2
0
0
7
.
[6
]
M
.
M
is
h
k
in
a
n
d
L
.
G
.
Un
g
e
rle
id
e
r,
"
Co
n
tr
ib
u
ti
o
n
o
f
striate
in
p
u
ts
t
o
t
h
e
v
isu
o
sp
a
ti
a
l
f
u
n
c
ti
o
n
s
o
f
p
a
rieto
-
p
re
o
c
c
ip
it
a
l
c
o
r
tex
in
m
o
n
k
e
y
s,"
Beh
a
v
io
u
ra
l
b
ra
i
n
re
se
a
rc
h
,
v
o
l
.
6
,
p
p
.
5
7
-
7
7
,
1
9
8
2
.
[7
]
L
.
Itt
i
a
n
d
C.
Ko
c
h
,
"
A
sa
li
e
n
c
y
-
b
a
se
d
se
a
rc
h
m
e
c
h
a
n
is
m
f
o
r
o
v
e
rt
a
n
d
c
o
v
e
rt
sh
if
ts
o
f
v
isu
a
l
a
tt
e
n
ti
o
n
,
"
Vi
sio
n
re
se
a
rc
h
,
v
o
l.
4
0
,
p
p
.
1
4
8
9
-
1
5
0
6
,
2
0
0
0
.
[8
]
W
.
Zh
a
n
g
,
Q.
J.
W
u
,
G
.
Wan
g
,
a
n
d
H.
Yi
n
,
"
A
n
a
d
a
p
ti
v
e
c
o
m
p
u
tatio
n
a
l
m
o
d
e
l
f
o
r
sa
li
e
n
t
o
b
jec
t
d
e
t
e
c
ti
o
n
,
"
IEE
E
T
ra
n
sa
c
ti
o
n
s
o
n
M
u
lt
ime
d
i
a
,
v
o
l.
1
2
,
p
p
.
3
0
0
-
3
1
6
,
2
0
1
0
.
[9
]
D.
W
ies
e
l
a
n
d
T
.
Hu
b
e
l,
"
Re
c
e
p
ti
v
e
f
ield
s
o
f
sin
g
le
n
e
u
ro
n
e
s
in
th
e
c
a
t’s
striate
c
o
rtex
,
"
J
.
Ph
y
sio
l,
v
o
l.
1
4
8
,
p
.
2
,
1
9
5
9
.
[1
0
]
B.
A
.
Olsh
a
u
se
n
,
"
E
m
e
rg
e
n
c
e
o
f
si
m
p
le
-
c
e
ll
re
c
e
p
ti
v
e
f
ield
p
ro
p
e
rti
e
s
b
y
lea
rn
in
g
a
sp
a
rse
c
o
d
e
f
o
r
n
a
tu
ra
l
im
a
g
e
s,
"
Na
tu
re
,
v
o
l.
3
8
1
,
p
p
.
6
0
7
-
6
0
9
,
1
9
9
6
.
[1
1
]
F
.
J.
H
u
a
n
g
,
Y.
-
L
.
B
o
u
re
a
u
,
a
n
d
Y
.
L
e
Cu
n
,
"
Un
su
p
e
rv
ise
d
l
e
a
rn
in
g
o
f
in
v
a
rian
t
f
e
a
tu
re
h
ie
ra
rc
h
ies
w
it
h
a
p
p
li
c
a
ti
o
n
s
to
o
b
jec
t
re
c
o
g
n
it
io
n
,
"
in
Co
mp
u
ter
Vi
sio
n
a
n
d
Pa
tt
e
rn
Rec
o
g
n
it
i
o
n
,
2
0
0
7
.
CV
PR
'0
7
.
IEE
E
Co
n
fer
e
n
c
e
o
n
,
2
0
0
7
,
p
p
.
1
-
8.
[1
2
]
M
.
Ries
e
n
h
u
b
e
r
a
n
d
T
.
P
o
g
g
io
,
"
Hie
ra
rc
h
ica
l
m
o
d
e
ls
o
f
o
b
jec
t
re
c
o
g
n
it
io
n
i
n
c
o
rtex
,
"
Na
tu
re
n
e
u
ro
sc
ien
c
e
,
v
o
l
.
2
,
p
p
.
1
0
1
9
-
1
0
2
5
,
1
9
9
9
.
[1
3
]
J.
C.
Be
z
d
e
k
,
R.
Ch
a
n
d
ra
se
k
h
a
r,
a
n
d
Y.
A
tt
ik
o
u
z
e
l,
"
A
g
e
o
m
e
tri
c
a
p
p
ro
a
c
h
t
o
e
d
g
e
d
e
tec
ti
o
n
,
"
IEE
E
T
ra
n
s
a
c
ti
o
n
s
o
n
F
u
zz
y
S
y
ste
ms
,
v
o
l.
6
,
p
p
.
5
2
-
7
5
,
1
9
9
8
.
60
65
70
75
80
85
90
10
20
30
40
50
Cl
assific
at
i
o
n
ac
cu
r
ac
y
N
u
mb
e
r
o
f
t
r
a
i
n
p
eo
p
le
SI
FT
H
M
AX
M
H
M
AX
G
M
AX
D
PHM
A
X
60
65
70
75
80
85
90
95
100
10
20
30
40
50
C
l
a
ssi
f
i
c
a
t
i
o
n
a
c
c
u
r
a
c
y
N
u
mb
e
r
o
f
t
r
a
i
n
B
ik
e
SI
FT
H
M
AX
M
H
M
AX
G
M
AX
D
PHM
A
X
Evaluation Warning : The document was created with Spire.PDF for Python.