I
n
d
o
n
e
s
i
a
n
J
o
u
r
n
a
l
o
f
E
l
e
c
tr
i
c
a
l
E
n
g
i
n
e
e
r
i
n
g
a
n
d
C
o
m
p
u
te
r
S
c
i
e
n
c
e
V
ol
.
8,
N
o.
1,
O
c
t
ob
er
20
17
,
pp
.
5
9
~
6
8
D
O
I
:
10.
115
91/
i
j
eec
s
.
v
8
.i
1
.
pp
5
9
-
6
8
59
R
ec
ei
v
ed
Ma
y
9
,
2
01
7
;
R
e
v
i
s
ed
A
ugus
t
2
1
,
201
7
;
A
c
c
ept
e
d
S
ept
e
mb
er
1
3,
20
17
A
n
E
d
ge
E
x
p
os
ur
e
us
i
ng C
a
l
i
b
e
r
Fuz
z
y
C
-
me
a
ns
W
i
th
Can
n
y
A
l
g
o
r
it
h
m
G
o
w
r
i
Jeyar
a
m
an
*
1
,
Jan
a
ki
r
am
an
S
u
b
b
i
ah
2
1
B
har
at
h
i
ar
U
n
i
v
er
s
i
t
y
,
C
oi
m
b
at
or
e,
T
am
i
l
n
adu,
I
nd
i
a
2
D
epar
t
m
ent
of
B
a
nk
i
ng
T
ec
h
nol
ogy
S
c
hoo
l
of
M
anagem
e
nt
,
P
ondi
c
h
er
r
y
U
ni
v
er
s
i
t
y
,
P
ond
i
c
her
r
y
,
I
ndi
a
*C
or
r
es
po
ndi
ng a
ut
hor
,
e
-
m
a
i
l:
gow
r
i
v
gr
@
y
ahoo.
c
o
m
A
b
st
r
act
E
dge
ex
p
os
ur
e or
ed
ge d
et
e
c
t
i
o
n i
s
a
n i
m
por
t
ant
and
c
l
a
s
s
i
c
al
s
t
udy
of
t
he
m
edi
c
al
f
i
el
d
and
c
om
put
er
v
i
s
i
o
n.
C
al
i
ber
F
u
z
z
y
C
-
m
eans
(
C
F
C
M
)
c
l
us
t
er
i
n
g
A
l
gor
i
t
hm
f
or
edg
e det
ec
t
i
on
depen
ds
o
n t
he
s
el
e
c
t
i
on
of
i
ni
t
i
al
c
l
u
s
t
er
c
e
nt
er
v
a
l
ue.
T
h
i
s
ende
av
o
r
t
o
pu
t
i
n
or
der
a
c
ol
l
ec
t
i
on
of
pi
x
el
s
i
n
t
o a
c
l
us
t
er
,
s
uc
h
t
hat
a
pi
x
el
w
i
t
hi
n
t
he
c
l
u
s
t
er
m
us
t
be
m
or
e
c
om
par
ab
l
e
t
o
ev
er
y
ot
h
er
pi
x
el
.
U
s
i
n
g
C
F
C
M
t
ec
hni
q
ues
f
i
r
s
t
c
l
u
s
t
er
t
he
B
S
D
S
i
m
a
ge,
nex
t
t
he c
l
u
s
t
er
ed
i
m
ag
e
i
s
gi
v
en
a
s
an i
np
ut
t
o t
he ba
s
i
c
c
an
ny
edg
e
d
et
e
c
t
i
o
n
a
l
gor
i
t
hm
.
T
he
ap
pl
i
c
at
i
on
of
n
ew
p
ar
am
et
er
s
w
i
t
h
f
ew
er
oper
at
i
o
ns
f
or
C
F
C
M
i
s
f
r
ui
t
f
u
l
.
A
c
c
or
di
ng
t
o t
he c
al
c
ul
a
t
i
on
,
a r
es
u
l
t
a
c
qu
i
r
ed
by
us
i
ng C
F
C
M
c
l
us
t
e
r
i
ng
f
un
c
t
i
on d
i
v
i
des
t
he i
m
ag
e i
nt
o
f
our
c
l
u
s
t
er
s
i
n
c
om
m
on.
T
he
pr
opo
s
ed
m
et
hod
i
s
ev
i
de
nt
l
y
r
obus
t
i
nt
o t
h
e m
odi
f
i
c
a
t
i
o
n of
f
u
z
z
y
c
-
m
eans
and c
an
ny
al
gor
i
t
hm
.
T
he c
o
nv
er
g
enc
e of
t
hi
s
al
gor
i
t
hm
i
s
v
er
y
s
peed
y
c
om
par
e t
o t
he ent
i
r
e e
dg
e
det
e
c
t
i
o
n al
g
or
i
t
hm
s
.
T
he
c
on
s
equ
enc
es
of
t
hi
s
pr
op
os
e
d a
l
gor
i
t
hm
m
ak
e enh
anc
ed ed
g
e det
e
c
t
i
o
n and
bet
t
er
r
e
s
u
l
t
t
han any
ot
h
er
t
r
a
di
t
i
o
nal
i
m
age edge
det
ec
t
i
on t
ec
hn
i
qu
es
.
Ke
y
w
o
rd
s
:
F
u
z
zy
C
-
m
eans
c
l
us
t
er
i
ng,
i
m
age
s
egm
e
nt
at
i
on,
c
an
ny
e
dge det
e
c
t
i
on,
S
el
f
-
O
r
gani
z
ed
M
ap
C
o
p
y
r
i
g
h
t
©
2
01
7
I
n
s
t
i
t
u
t
e
o
f
A
d
v
a
n
c
e
d
E
n
g
i
n
e
e
r
i
n
g
a
n
d
S
c
i
e
n
c
e
.
A
l
l
r
i
g
h
t
s r
es
er
ved
.
1
.
I
n
tr
o
d
u
c
ti
o
n
T
he pur
pos
e
of
ed
ge
ex
p
os
ur
e i
s
f
or
O
bj
ec
t
D
et
ec
t
i
on
.
T
her
e
ar
e n
um
er
ous
ed
ge
det
ec
t
i
on
al
g
or
i
t
hm
s
av
ai
l
a
bl
e t
o c
om
put
e t
he s
egr
e
ga
t
i
on
of
bound
ar
i
es
i
n a
n i
m
age.
T
her
e ar
e
t
hr
ee f
undam
ent
al
t
y
p
es
of
di
s
c
ont
i
nu
i
t
i
es
i
n an
i
m
a
ge,
w
h
i
c
h ar
e p
o
i
nt
s
,
l
i
n
e
s
and edg
es
.
E
dg
es
por
t
r
a
y
obj
ec
t
b
oun
dar
i
es
and
t
he
i
r
us
ef
ul
f
eat
ur
es
f
or
s
egm
ent
at
i
on.
E
d
ge
d
et
ec
t
i
on
i
s
us
ed
t
o
ex
t
r
ac
t
s
a
l
i
ent
f
eat
ur
es
of
an
i
m
age.
T
her
e
ar
e
s
om
e
appl
i
c
at
i
o
ns
of
edg
e
det
ec
t
i
o
n
i
n
r
eal
l
i
f
e,
w
hi
c
h ar
e:
a)
F
ac
e det
ec
t
i
o
n
–
now
ad
a
y
s
t
hi
s
i
s
us
e
d i
n s
oc
i
a
l
m
edi
a
apps
f
or
ex
am
pl
e F
ac
ebo
o
k
.
b)
P
eo
pl
e C
o
unt
i
ng
-
i
t
i
s
br
oug
ht
i
nt
o
pl
a
y
i
n
ana
l
y
z
i
ng
s
t
o
r
e per
f
or
m
anc
e dur
i
ng c
ar
n
i
v
al
s
.
c)
V
eh
i
c
l
e det
ec
t
i
o
n
-
us
ed
t
o d
et
ec
t
a t
y
p
e of
s
hi
p en
t
er
i
n
g a p
or
t
or
t
r
ac
k
i
ng t
h
e s
pe
ed of
a c
ar
.
d)
S
ec
ur
i
t
y
-
us
e
d t
o r
ec
o
gn
i
z
e
anom
al
i
es
i
n a
bom
b ex
pl
o
s
i
v
e.
E
dg
e
d
et
ec
t
i
on
of
an
i
m
age
i
s
c
ar
r
i
e
d
o
ut
us
i
ng
gr
a
di
ent
a
nd
La
pl
ac
i
an
op
er
at
i
o
n.
T
he
boun
dar
y
i
s
r
e
pr
es
ent
ed
b
y
i
t
s
l
e
ngt
h
a
nd
r
e
gu
l
ar
i
t
y
.
T
he
boun
dar
y
des
c
r
i
p
t
or
i
s
c
l
as
s
i
f
i
ed
i
nt
o
F
our
i
er
d
es
c
r
i
pt
or
an
d po
l
y
nom
i
al
appr
ox
i
m
at
i
on m
et
hods
.
B
ot
h m
et
hods
ar
e us
e
d i
n c
l
us
t
er
i
n
g.
C
l
us
t
er
i
ng
i
s
a
s
u
bdi
v
i
s
i
on
of
uns
uper
v
i
s
ed
l
e
ar
ni
ng
pr
oc
es
s
es
[
1]
.
I
t
i
s
a
m
et
hod
f
or
c
l
as
s
i
f
y
i
n
g
dat
a i
n an i
m
age and t
o f
i
nd c
l
us
t
er
s
w
i
t
h t
he m
os
t
l
i
k
enes
s
i
n t
he i
den
t
i
c
a
l
c
l
u
s
t
er
and
m
o
s
t
unl
i
k
enes
s
am
ong di
v
er
s
e
c
l
us
t
er
s
[
2]
.
S
egm
ent
at
i
on b
y
c
l
us
t
er
i
ng
i
s
done b
y
t
hr
e
e s
t
eps
w
hi
c
h ar
e:
-
t
he i
n
i
t
i
al
s
t
ep i
s
t
o
def
i
ne
t
he c
o
l
or
f
eat
ur
es
;
t
h
e s
ec
o
nd s
t
e
p i
s
t
o t
r
ans
f
or
m
t
he
pi
x
el
s
i
nt
o c
o
l
or
f
eat
ur
e s
p
ac
e an
d f
i
na
l
l
y
c
l
us
t
er
t
h
e p
i
x
el
s
i
n
c
ol
or
f
eat
ur
e s
p
ac
e.
F
u
z
z
y
l
og
i
c
pr
o
v
i
des
a
t
ec
h
ni
q
ue
t
o
m
a
k
e of
f
i
c
i
al
r
eas
oni
ng.
T
he
c
odi
ng
of
i
n
put
i
m
age
dat
a
i
s
c
al
l
ed
as
f
uz
z
i
f
i
c
at
i
on.
T
he
dec
o
di
ng
of
t
he
out
pu
t
i
m
age
i
s
k
now
n
as
def
u
z
z
i
f
i
c
at
i
on
i
n
f
uz
z
y
t
ec
hn
i
q
ues
.
T
he
m
odi
f
i
c
at
i
on
of
m
e
m
b
er
s
hi
p
v
al
ues
i
s
t
he i
m
por
t
ant
s
t
ep i
n f
u
z
z
y
c
l
us
t
er
i
n
g.
F
u
z
z
y
s
et
de
c
l
ar
es
gr
adat
i
o
n
of
al
l
t
o
ne
s
i
n bet
w
e
en
bl
ac
k
and
w
h
i
t
e.
F
u
z
z
y
s
e
t
s
dea
l
w
i
t
h
gr
a
dua
l
i
t
y
of
c
onc
ept
s
and
f
u
z
z
y
m
em
ber
s
hi
p
f
unc
t
i
ons
(
S
ee F
i
gur
e
1)
.
Evaluation Warning : The document was created with Spire.PDF for Python.
I
SSN
:
25
02
-
4
752
I
J
E
EC
S
V
o
l.
8
,
N
o.
1,
O
c
t
o
ber
20
17
:
5
9
–
6
8
60
F
i
gur
e
1.
F
u
z
z
y
c
l
us
t
er
i
ng
F
u
z
z
y
c
l
us
t
er
i
ng t
ec
hn
i
qu
es
hav
e a f
eat
ur
e of
t
h
e f
u
z
z
y
s
et
t
he
or
y
.
I
n i
m
age
s
egm
ent
at
i
on
,
F
u
z
z
y
C
-
M
eans
(
F
C
M)
t
ec
hn
i
q
ue
i
s
one
of
t
he
ge
ner
a
l
l
y
app
l
i
ed t
ec
hn
i
qu
es
.
F
C
M a
l
gor
i
t
hm
i
s
an
ex
t
e
ns
i
on
of
t
he
k
-
m
eans
c
l
us
t
er
i
ng
t
ec
hn
i
q
ue
w
i
t
h a
f
u
z
z
y
s
et
.
T
he
s
el
ec
t
i
on
of
a
par
t
i
c
u
l
ar
t
ec
hni
que
w
i
l
l
d
epe
nd
on
w
h
i
c
h t
y
pe
of
ou
t
put
i
s
des
i
r
ed
and
ho
w
m
uc
h
per
f
or
m
anc
e,
t
he pr
oc
es
s
r
equi
r
ed.
F
C
M has
been
us
ed i
n m
edi
c
i
ne i
m
agi
n
g and pat
t
er
n
r
ec
ogni
t
i
o
n.
T
hi
s
pap
er
i
s
pl
a
nne
d as
f
ol
l
o
w
s
:
-
I
n d
i
v
i
s
i
on 1.
1,
t
h
e F
C
M
i
s
an
al
y
z
e
d
a
nd t
he
par
am
et
er
s
el
ec
t
i
ons
ar
e
t
al
k
ed ab
out
.
I
n
di
v
i
s
i
on
2
,
a c
om
pr
ehens
i
v
e
deb
at
e
of
pr
opos
ed
al
g
or
i
t
hm
i
s
pr
es
e
nt
e
d.
T
hi
s
c
onf
er
s
t
he
s
t
r
o
ng
pr
op
er
t
i
es
of
c
l
us
t
er
a
l
g
or
i
t
hm
.
I
n
di
v
i
s
i
on
3,
t
h
e
f
i
nal
o
ut
c
om
e of
t
he pr
op
os
ed s
y
s
t
em
i
s
l
i
s
t
e
d o
ut
.
1.
1.
R
el
at
ed
W
o
r
ks
C
l
us
t
er
i
ng p
er
f
or
m
s
di
v
i
d
i
n
g pi
x
e
l
po
i
nt
s
i
n
t
o hom
oge
neous
c
l
us
t
er
s
,
i
n
w
h
i
c
h t
h
e pi
x
e
l
s
i
n
t
he
s
am
e
c
l
as
s
ar
e
r
el
at
ed
a
nd
t
he
p
i
x
el
s
i
n
di
f
f
er
ent
c
l
as
s
es
ar
e
u
nr
el
at
ed
.
I
n
v
ar
i
ous
s
t
eps
c
l
us
t
er
i
n
g i
s
per
f
or
m
ed:
-
1)
F
eat
ur
e s
e
l
ec
t
i
on or
ex
t
r
ac
t
i
on
i
s
done
us
i
ng t
he
i
np
ut
i
m
age and t
he
r
es
ul
t
i
s
a pat
t
er
n r
epr
es
e
nt
at
i
on.
2)
T
hi
s
pat
t
er
n i
s
gi
v
en as
i
nput
f
or
i
nt
er
-
pat
t
er
n s
i
m
i
l
ar
i
t
y
oper
at
i
o
n an
d t
h
e o
ut
pu
t
i
s
gi
v
en
t
o gr
o
up
i
ng
op
er
at
i
on
.
T
he f
i
nal
r
es
u
l
t
i
s
a c
l
us
t
er
ed i
m
age.
T
hi
s
s
ec
t
i
on di
s
c
us
s
es
t
he i
nf
l
uenc
e of
F
u
z
z
y
C
-
m
eans
on ed
ge
ex
pos
ur
e i
n i
m
age
pr
oc
es
s
i
ng.
Bo
o
n
-
S
eng C
h
e
w
et
al
.
[
3]
b
ui
l
t
a f
u
z
z
y
c
l
us
t
er
i
n
g a
l
gor
i
t
hm
w
h
i
c
h
us
es
t
he d
at
a
r
es
em
bl
anc
e w
i
t
hi
n t
he f
r
am
ew
or
k
s
t
r
uc
t
ur
e of
a v
i
r
t
ual
c
har
ac
t
er
(
V
C
)
m
odel
and i
s
t
oget
her
c
ons
i
der
e
d
w
i
t
h
t
he
t
em
por
al
c
o
her
enc
e
i
n
t
he
m
o
v
em
ent
dat
a.
J
af
er
z
a
deh
,
K
.
e
t
a
l
.
[
4
]
ex
pl
a
i
ne
d
t
h
e
dom
ai
n
an
d
r
ange
b
l
oc
k
s
c
at
egor
i
z
ed
b
y
a
f
u
z
z
y
c
-
m
ean
-
c
l
us
t
er
i
n
g
m
et
hod
and
c
o
m
par
ed
w
i
t
h t
he us
e
d
i
s
c
r
et
e c
os
i
n
e t
r
ans
f
or
m
c
oe
f
f
i
c
i
ent
.
E
v
ans
A
.
N
et
al
.
[
5]
pr
op
o
s
ed
a
ne
w
c
o
l
or
edg
e
det
e
c
t
or
bas
ed
on
v
ec
t
or
d
i
s
s
i
m
i
l
ar
i
t
y
and
i
t
s
per
f
or
m
anc
e
get
be
t
t
er
i
n
t
h
e
pr
es
enc
e
of
no
i
s
e.
S
.
K
r
i
n
i
d
i
s
et
al
.
[
6]
pr
op
os
ed
al
gor
i
t
hm
t
hat
i
nc
or
p
or
at
es
t
he
l
oc
al
s
pat
i
al
de
t
ai
l
s
a
nd
gr
a
y
l
e
v
el
det
ai
l
s
i
n a
n
e
w
f
u
z
z
y
m
et
hod
an
d
w
hi
c
h
us
es
a f
u
z
z
y
l
oc
a
l
s
i
m
i
l
ar
i
t
y
m
eas
ur
e a
nd p
l
a
nn
ed t
o
guar
ant
e
e n
oi
s
e
i
ns
ens
i
t
i
v
enes
s
,
i
nf
or
m
at
i
on pr
es
er
v
a
t
i
o
n.
T
o c
al
c
ul
at
e t
he c
ol
or
di
f
f
er
enc
e pr
oper
l
y
,
t
he d
i
g
i
t
a
l
p
i
c
t
ur
e c
ol
or
s
ar
e
s
y
m
bol
i
z
e
d
i
n
a
m
odi
f
i
ed
L*
u
*
v
c
ol
or
s
pac
e
[
7]
,
t
he
c
ol
or
r
e
duc
t
i
on
i
s
ex
p
ec
t
e
d i
nt
o
a
s
et
of
m
odel
s
us
i
ng
s
el
f
-
or
gan
i
z
i
ng m
ap (
S
O
M)
l
e
ar
ni
n
g.
T
he
w
ei
g
ht
e
d f
u
z
z
y
f
ac
t
or
u
s
es
t
he s
p
ac
e
di
s
t
anc
e
of
al
l
a
dj
ac
ent
pi
x
el
s
and
t
h
ei
r
gr
a
y
-
l
e
v
el
d
i
s
s
i
m
i
l
ar
i
t
y
s
i
m
ul
t
ane
ous
l
y
.
B
y
us
i
ng t
hi
s
i
s
s
ue i
n [
8]
,
t
h
e ne
w
m
et
hod c
an ac
c
ur
at
el
y
c
a
l
c
ul
a
t
e t
he d
am
pi
ng am
ount
of
adj
a
c
ent
p
i
x
el
s
.
F
u
z
z
y
C
ond
i
t
i
on
al
C
l
us
t
er
i
ng bas
e
d Mo
de
l
i
ng m
et
hod [
9]
,
w
hi
c
h pr
oduc
es
f
u
z
z
y
r
ul
es
r
epeat
ed
l
y
us
i
ng t
h
e c
ond
i
t
i
on
al
F
u
z
z
y
C
-
M
eans
a
l
go
r
i
t
hm
,
and i
t
pr
opos
es
t
h
e
us
e of
a new
appr
o
ac
h
f
or
at
t
r
i
but
es
gr
o
upi
ng
on
t
he
c
ont
ex
t
def
i
ni
t
i
on
s
t
e
p,
us
i
n
g
he
ur
i
s
t
i
c
s
e
ar
c
h
bas
ed
on
t
he b
es
t
r
ec
i
t
a
l
.
D
z
i
uk
,
M.
A
.
et
al
.
[
10
]
i
m
pl
em
ent
ed t
h
e
f
uz
z
y
l
o
gi
c
c
ont
r
o
l
f
or
aut
o
pi
l
ot
s
w
h
i
c
h i
s
at
t
a
i
ne
d b
y
t
w
o s
et
s
of
f
uz
z
y
r
u
l
es
,
one f
or
c
ont
r
ol
l
i
ng t
h
e t
r
a
ns
f
or
m
at
i
on i
n h
ead
i
ng
an
d t
h
e
ot
her
f
or
s
c
hem
i
ng
t
he
r
ev
ol
ut
i
o
ni
z
e
i
n
el
ev
a
t
i
o
n
i
n
t
he
ai
r
c
r
a
f
t
.
T
he
bas
i
c
t
hou
g
ht
s
of
f
u
z
z
y
s
e
t
and t
h
e t
he
or
i
es
of
f
uz
z
y
C
-
m
ean c
l
us
t
er
i
ng t
ec
h
ni
que
ar
e di
s
c
us
s
ed i
n [
11]
.
T
he F
u
z
z
y
C
-
M
ea
n
al
g
or
i
t
hm
w
as
us
ed
t
o
c
l
u
s
t
er
dat
a f
i
r
s
t
,
t
h
en
t
he
F
u
z
z
y
I
nf
er
enc
e
S
y
s
t
em
had
bee
n c
r
eat
ed
bas
ed on t
h
es
e c
l
us
t
er
s
by
v
ar
i
ous
r
ul
es
,
m
ax
i
m
u
m
num
ber
s
o
f
i
nput
s
and f
l
o
oded h
y
dr
a
ul
i
c
c
onduc
t
i
v
i
t
y
as
a
r
es
u
l
t
i
n [
12]
.
K
i
an
i
,
S
.
et
al
.
[
1
3]
pr
o
pos
ed
a m
et
hod
w
h
i
c
h
us
e
s
a s
pec
i
a
l
t
y
p
e
of
f
r
ac
t
al
c
odi
ng.
I
t
s
c
ons
t
r
ai
nt
s
ar
e
c
ont
r
as
t
e
d
s
c
al
i
n
g
and
t
he
av
er
ag
e
r
ang
e
o
f
bl
oc
k
.
A
l
s
o,
i
t
em
pl
o
y
s
t
he
f
u
z
z
y
C
-
m
ean
c
l
us
t
er
i
n
g
t
o
t
al
k
t
o
t
he
w
at
er
m
ar
k
bi
t
s
.
Medi
c
a
l
i
m
agi
n
g
t
ec
hn
i
q
ue
i
nc
l
u
ded f
u
z
z
y
C
-
m
eans
(
F
C
M)
t
i
s
s
ue c
l
as
s
i
f
i
c
at
i
on
an
d i
m
age ac
qui
r
em
ent
has
b
een b
egu
n as
Evaluation Warning : The document was created with Spire.PDF for Python.
I
J
E
EC
S
IS
S
N
:
2
502
-
4
752
A
n
E
dg
e E
x
p
os
ur
e us
i
ng
C
al
i
ber
F
u
z
z
y
C
-
mea
ns
W
i
t
h C
anny
A
l
gor
i
t
hm
(
G
ow
r
i
J
e
y
ar
aman
)
61
a f
eas
i
bl
e ap
pr
oac
h f
or
ed
ge de
t
ec
t
i
on of
t
he
bo
ne f
r
om
s
o
f
t
t
i
s
s
ue [
14
-
15]
.
T
he G
ener
a
l
i
z
e
d
F
u
z
z
y
C
-
Me
ans
d
eal
s
w
i
t
h
t
he
de
v
i
at
i
on
of
F
C
M
al
g
or
i
t
hm
s
and
c
an
ef
f
or
t
l
es
s
l
y
l
ea
d
t
o
i
nn
ov
at
i
v
e
and s
t
i
m
ul
at
i
n
g
c
l
us
t
er
i
n
g a
l
gor
i
t
hm
s
[
16]
.
A
l
t
er
nat
i
v
e F
u
z
z
y
C
-
Mea
ns
ar
e us
ed
i
n
c
l
us
t
er
an
al
y
s
i
s
w
h
i
c
h r
e
pl
a
c
es
t
he E
uc
l
i
de
an m
odel
[
1
7]
.
1
.1
.1
. F
u
z
z
y
C
-
M
ean
s
F
u
z
z
y C
-
m
eans
i
s
t
he ex
t
e
ns
i
on of
K
-
m
eans
.
F
u
z
z
y
C
-
m
eans
al
l
o
w
s
p
i
x
el
s
p
oi
n
t
s
t
o be
as
s
i
gne
d
t
o
m
ul
t
i
pl
e
c
l
us
t
e
r
s
and
e
ac
h
p
i
x
el
po
i
nt
ha
s
a
de
gr
ee
of
m
e
m
ber
s
hi
p
i
n
a
c
l
us
t
er
t
o
w
hi
c
h
i
t
be
l
o
ngs
.
T
hi
s
a
l
g
or
i
t
hm
us
es
m
e
m
ber
s
hi
p
f
unc
t
i
on
an
d
c
l
us
t
er
c
ent
er
v
a
l
ues
w
h
i
c
h
ar
e
upda
t
ed
i
t
er
at
i
v
e
l
y
.
T
he F
C
M i
n
v
o
l
v
es
f
ol
l
o
w
i
ng s
t
e
ps
:
1.
C
ons
i
d
er
M X
N
di
m
ens
i
on
al
p
i
x
el
s
r
e
pr
es
ent
ed
b
y
x
i
.
2.
S
up
pos
e t
he
nu
m
ber
of
c
l
us
t
er
s
C
,
w
her
e
2≤
C
≤
N
.
3.
S
el
ec
t
t
h
e l
ev
el
of
c
l
us
t
er
f
uz
z
i
n
es
s
f
>
1
4.
S
et
t
he I
ni
t
i
a
l
v
al
ue f
or
m
em
ber
s
hi
p
m
at
r
i
x
U
.
5.
C
om
put
e t
he f
u
z
z
y
c
en
t
r
oi
d
f
or
j
=
1,
.
.
,
C
.
w
her
e m
i
s
t
he f
u
z
z
y
par
am
et
er
an
d n
i
s
t
h
e n
um
ber
of
pi
x
el
po
i
nt
s
.
6.
D
et
er
m
i
ne t
he
E
uc
l
i
dea
n d
i
s
t
anc
e be
t
w
ee
n p
i
x
el
an
d c
l
us
t
er
c
ent
r
oi
d
.
7.
Mod
i
f
y
t
he
f
u
z
z
y
m
e
m
ber
s
hi
p m
at
r
i
x
U
.
R
epe
at
t
h
e s
t
eps
5 t
o
7 u
nt
i
l
t
h
e c
ut
of
f
m
e
m
ber
s
hi
p i
s
o
bt
ai
ned
.
1.
1.
2
.
P
ar
am
et
er
s o
f
F
C
M
T
he
F
u
z
z
y
C
-
m
eans
l
ook
s
t
o
b
e
v
er
y
s
i
m
pl
e
a
nd
eas
y
t
o
u
nder
s
t
a
nd
but
i
t
c
ons
i
s
t
s
of
v
ar
i
ous
p
ar
am
et
er
s
or
f
ac
t
or
s
,
w
hi
c
h
af
f
ec
t
t
he ef
f
i
c
i
enc
y
of
t
h
i
s
al
gor
i
t
hm
.
T
he f
ac
t
or
s
,
w
hi
c
h
c
aus
e s
er
i
o
us
dam
age t
o t
hi
s
a
l
gor
i
t
hm
,
ar
e f
ol
l
o
w
ed:
a.
T
he
as
s
u
m
pt
i
on
of
s
t
ar
t
i
n
g
c
l
us
t
er
c
e
nt
r
oi
d:
t
he
ef
f
i
c
i
enc
y
of
F
C
M
f
ul
l
an
d
f
ul
l
y
d
epen
ds
on
t
hi
s
f
ac
t
or
,
t
h
e c
ent
r
o
i
d
v
al
ue s
el
ec
t
i
on m
us
t
be n
ear
e
r
t
o t
h
e e
ndi
ng c
e
nt
r
oi
d
v
a
l
ue.
I
f
i
t
i
s
a
good
c
ent
er
v
a
l
u
e,
t
h
en
i
t
i
s
c
onv
er
ged s
pee
di
l
y
and
pe
r
f
or
m
anc
e
t
i
m
e w
i
l
l
be
v
er
y
l
es
s
.
b.
T
he num
ber
of
c
l
us
t
er
s
C
:
t
he
nex
t
i
m
por
t
ant
par
am
et
er
i
s
c
l
us
t
er
n
um
ber
,
w
h
i
c
h
dec
i
d
es
t
he
k
e
y
s
t
e
ps
i
n
f
u
z
z
y
c
-
m
eans
a
l
g
or
i
t
hm
.
U
s
ual
l
y
,
t
h
e
num
ber
r
ang
es
f
r
o
m
2
t
o
am
ount
of
pi
x
el
s
i
n
an
i
m
age.
T
he s
el
ec
t
i
on
of
c
l
us
t
er
num
ber
pr
oduc
es
a
d
i
f
f
er
ent
r
es
ul
t
f
or
a d
i
f
f
er
ent
num
ber
of
c
l
us
t
er
s
.
F
u
z
z
y
par
am
et
er
m
:
t
hi
s
f
u
z
z
y
p
ar
am
et
er
m
pr
es
ent
s
i
n f
u
z
z
y
m
em
ber
s
hi
p m
at
r
i
x
U
.
B
y
def
aul
t
,
m
v
al
ue m
us
t
be gr
eat
er
t
han
on
e an
d a
l
s
o i
t
i
s
a r
eal
num
ber
.
2.
C
al
i
b
er
F
u
z
z
y
C
-
M
ean
s (
C
F
C
M
)
2.
1.
F
u
z
z
y
C
-
m
e
a
n
s
c
l
u
s
te
r
i
n
g
w
i
th
M
o
d
i
fi
c
a
ti
o
n
T
hi
s
paper
pr
oj
ec
t
ed a s
i
m
pl
e but
a v
er
y
w
el
l
-
or
ga
ni
z
ed F
u
z
z
y
C
-
m
eans
and
c
ann
y
edge
det
ec
t
i
o
n al
gor
i
t
hm
w
hi
c
h c
on
v
e
y
t
h
e v
i
e
w
s
of
di
g
i
t
a
l
i
m
age pr
oc
es
s
i
ng
.
C
om
par
e t
o al
l
f
uz
z
y
c
l
us
t
er
i
n
g m
et
hods
,
t
he f
u
z
z
y
c
-
m
eans
(
F
C
M)
al
g
or
i
t
hm
i
s
t
he
m
os
t
f
a
m
ous
t
ec
hni
que
,
s
i
nc
e
i
t
has
t
he
be
nef
i
t
of
r
obus
t
n
es
s
f
or
v
ague
nes
s
and
m
ai
nt
ai
ns
a
gr
eat
d
eal
of
i
nf
or
m
at
i
on
t
han e
v
er
y
h
ar
d c
l
us
t
er
i
ng t
ec
hn
i
qu
e.
I
t
i
s
ex
t
ens
i
v
el
y
us
ed an
d c
onc
er
ned i
n i
m
ag
e
s
egm
ent
at
i
on
a
nd
i
m
age
c
l
us
t
er
i
ng
.
T
he
C
al
i
b
er
f
uz
z
y
c
-
m
eans
c
ons
i
s
t
of
v
ar
i
ous
m
odi
f
i
c
at
i
on
s
i
n t
r
a
di
t
i
o
nal
f
u
z
z
y
C
-
m
ean
s
w
h
i
c
h ar
e as
f
ol
l
o
w
s
:
a)
T
he num
ber
of
c
l
us
t
er
s
el
e
c
t
i
on
i
s
do
ne
us
i
ng
S
el
f
-
O
r
gan
i
z
ed
Map (
S
O
M)
i
n n
eu
r
al
n
et
w
or
k
c
onc
ept
.
b)
T
he s
t
ar
t
i
ng
c
l
us
t
er
c
ent
r
o
i
d v
al
ue
i
s
f
i
x
ed
us
i
ng
t
he
pr
oc
edur
e:
s
el
ec
t
c
en
t
r
oi
d
f
r
om
n
-
p
ix
e
ls
i
n
s
uc
h
a
w
a
y
t
h
at
t
he
C
or
r
el
at
i
on
d
i
s
t
anc
e
of
t
hat
pi
x
el
i
s
hi
g
h
f
r
o
m
ot
her
pi
x
el
s
E
quat
i
o
n
(1
).
c)
F
u
z
z
y
par
am
et
er
m
s
el
ec
t
i
on
i
s
do
ne
us
i
ng
t
he r
a
nd
om
s
el
ec
t
i
o
n
m
et
hod
w
h
i
c
h
t
ak
es
t
he
v
a
l
ue
i
n
t
h
e
r
ange
1.
5
t
o
2
.
5.
I
f
t
he
m
ax
i
m
u
m
v
al
ue
o
f
t
he
f
uz
z
i
f
i
er
m
i
s
out
s
i
de
t
he
up
per
boun
dar
y
v
al
ue,
t
hen
t
he
u
n
w
ant
ed
noi
s
e
i
nf
or
m
at
i
on
i
s
i
nc
l
u
ded
i
n
t
he r
es
ul
t
ant
i
m
age.
Evaluation Warning : The document was created with Spire.PDF for Python.
I
SSN
:
25
02
-
4
752
I
J
E
EC
S
V
o
l.
8
,
N
o.
1,
O
c
t
o
ber
20
17
:
5
9
–
6
8
62
T
he
S
O
M
us
es
t
he
t
r
ai
n
i
ng
w
h
i
c
h
ex
pl
oi
t
s
c
om
pet
i
t
i
v
e
l
ear
n
i
ng
i
s
a
f
eed
f
or
w
ar
d
net
w
or
k
.
T
hi
s
S
O
M
c
al
c
ul
a
t
es
t
he
E
uc
l
i
d
ean
d
i
s
t
anc
e
al
l
w
ei
g
ht
v
ec
t
or
s
.
N
eur
ons
w
e
i
gh
t
v
ec
t
or
i
s
s
i
m
i
l
ar
t
o t
h
e i
nput
v
al
u
e
w
hi
c
h
i
s
i
d
ent
i
f
i
ed
as
t
he
b
es
t
m
at
c
hi
ng un
i
t
.
T
hi
s
un
i
t
i
s
us
ed t
o f
i
nd
out
t
he c
l
us
t
er
c
oun
t
of
t
hi
s
pr
opos
e
d a
l
gor
i
t
hm
.
W
(
s
+1
)
=W
(s
)+
Θ
(u
,
v
,
s
)α
(s
)(
D
(
t)
-
W
(s
)
)
I
n
t
h
e
ab
ov
e
f
or
m
ul
a,
W
i
s
w
ei
ght
v
ec
t
or
,
D
(
t
)
i
s
t
h
e
or
i
gi
na
l
i
npu
t
v
ec
t
or
,
Θ
(
u,
v
,
s
)
i
s
t
he ne
i
g
hbor
h
ood f
unc
t
i
on,
‘
s
’
i
s
s
t
ep i
ndex
,
‘
u’
i
s
i
ndex
of
bes
t
m
at
c
hi
ng uni
t
f
or
D
(
t
)
,
‘
t
’
i
s
i
n
dex
i
nt
o
t
he
t
r
ai
ni
ng
s
am
pl
e
a
n
d
α
(
s
)
i
s
a
l
ear
n
i
ng
c
oef
f
i
c
i
ent
.
T
her
e ar
e
v
ar
i
ous
d
i
s
t
an
c
e
c
al
c
ul
at
i
on
f
or
m
ul
as
a
r
e a
v
ai
l
a
bl
e
s
uc
h
as
E
u
c
l
i
de
an
di
s
t
anc
e
,
Ma
nhat
t
an d
i
s
t
a
nc
e,
Mi
nk
ow
s
k
i
di
s
t
anc
e,
and C
or
r
e
l
at
i
o
n di
s
t
a
nc
e.
A
m
ong al
l
,
t
he
c
or
r
el
at
i
on
di
s
t
a
nc
e i
s
s
e
l
e
c
t
ed f
or
i
ni
t
i
a
l
c
e
nt
r
oi
d v
al
u
e s
et
t
i
ng
pr
oc
es
s
bec
aus
e
of
i
t
s
ef
f
i
c
i
enc
y
.
T
he c
or
r
el
at
i
on d
i
s
t
anc
e
r
xy
is
c
al
c
ul
at
e
d us
i
ng t
he f
or
m
ul
a
E
q
uat
i
on
(
1)
(
1)
w
her
e X
an
d
Y
ar
e
t
h
e
pi
x
el
v
a
l
ue,
C
o
v
(
X
,
Y
)
m
ean
s
c
ov
ar
i
anc
e
of
X
an
d
Y
,
V
a
r(X
)
r
epr
es
ent
s
t
he
v
ar
i
anc
e of
X
and V
ar
(
Y
)
r
epr
es
ent
s
t
he v
ar
i
anc
e of
Y
.
T
he c
ov
ar
i
anc
e a
nd
v
ar
i
anc
e f
or
m
ul
as
ar
e l
i
s
t
ed
bel
o
w
E
qu
at
i
on
(2
) t
o
(4
):
(
2)
(
3)
(
4)
(
a)
(
b)
F
i
gur
e
2.
(
a)
O
r
i
g
i
n
al
i
m
age (
b)
c
l
us
t
er
ed
i
m
age
T
he abo
v
e
-
c
l
us
t
er
ed
i
m
age i
s
d
er
i
v
ed
af
t
er
a
pp
l
y
i
ng
C
al
i
ber
F
u
z
z
y
C
-
m
eans
(
C
F
C
M)
t
o
t
he or
i
gi
na
l
i
m
age (
S
ee F
i
g
ur
es
.
2)
.
T
he pr
ogr
es
s
of
nex
t
s
t
ep
i
s
s
ho
w
n
i
n
t
he
ne
x
t
di
v
i
s
i
on.
2.
2.
C
an
n
y
E
d
g
e D
et
ect
i
o
n
T
he C
ann
y
edg
e
det
ec
t
or
i
s
a g
ood
de
t
ec
t
i
on
and
g
oo
d l
oc
al
i
z
at
i
on
op
er
at
or
.
T
he
hi
gh
-
qua
l
i
t
y
d
et
ec
t
i
on
m
eans
opt
i
m
al
det
ec
t
or
d
i
m
i
ni
s
h t
h
e pr
ob
ab
i
l
i
t
y
of
f
al
s
e pos
i
t
i
v
es
a
nd f
al
s
e
nega
t
i
v
es
.
T
he
hi
g
h
-
qu
al
i
t
y
l
oc
al
i
z
at
i
on
m
eans
t
he
ed
ges
det
ec
t
ed
ar
e
c
l
os
e
t
o
t
he
t
r
ue
e
dges
.
C
ann
y
edg
e
det
ec
t
or
s
at
i
s
f
i
es
S
i
ng
l
e
R
es
po
ns
e
C
ons
t
r
ai
nt
a
nd
r
e
t
ur
ns
o
ne
po
i
nt
onl
y
f
or
e
v
er
y
edge
po
i
n
t
.
A
l
s
o,
t
h
i
s
h
as
t
he f
eat
ur
es
s
uc
h
as
C
o
nv
ol
ut
i
o
n (
S
m
oot
hi
ng)
w
i
t
h a
der
i
v
a
t
i
v
e
of
G
aus
s
i
an,
N
on
-
m
ax
i
m
u
m
S
uppr
es
s
i
o
n,
a
nd H
y
s
t
er
es
i
s
T
hr
es
hol
di
ng.
1
)
(
)
(
1
2
−
−
=
∑
=
n
X
x
X
V
ar
n
i
i
Evaluation Warning : The document was created with Spire.PDF for Python.
I
J
E
EC
S
IS
S
N
:
2
502
-
4
752
A
n
E
dg
e E
x
p
os
ur
e us
i
ng
C
al
i
ber
F
u
z
z
y
C
-
mea
ns
W
i
t
h C
anny
A
l
gor
i
t
hm
(
G
ow
r
i
J
e
y
ar
aman
)
63
F
i
gur
e
3.
C
ann
y
E
dg
e d
et
e
c
t
i
on s
t
e
ps
I
n
C
an
n
y
edg
e
det
ec
t
i
o
n,
t
he
f
i
r
s
t
s
t
ep
i
s
t
o
d
es
t
r
o
y
n
oi
s
e
b
y
ed
ge
s
m
oot
hi
n
g
us
i
ng
t
he
G
aus
s
i
an.
I
f
t
he m
as
k
i
s
ov
er
s
i
z
e
d,
t
h
en s
om
e edg
es
i
nf
or
m
at
i
on
w
i
l
l
b
e l
os
t
.
T
he
s
ec
ond s
t
ep
i
s
a
di
r
ec
t
i
on
al
t
r
a
ns
f
or
m
at
i
on
i
n
t
he
i
nt
ens
i
t
y
of
a
d
i
gi
t
al
i
m
age.
T
he
t
hi
r
d
s
t
ep
i
s
us
ed
t
o
m
a
k
e
a
t
hi
n
ed
ge.
A
l
s
o
,
no
n
-
m
ax
i
m
u
m
s
uppr
es
s
i
on
h
el
ped
t
o s
uppr
es
s
e
nt
i
r
e
gr
ad
i
e
nt
v
a
l
ues
t
o
z
er
o
ex
c
ept
s
har
pes
t
i
nt
ens
i
t
y
v
al
u
e.
T
he
l
as
t
s
t
e
p
i
s
us
ed
t
o
c
r
eat
e
c
o
nnec
t
e
d
-
c
om
p
onen
t
.
T
hat
i
s
h
y
s
t
er
es
i
s
t
hr
es
h
ol
d
i
n
g
c
om
pl
et
e t
h
e
edg
es
(
F
i
g
ur
e
3)
.
T
he bas
i
c
c
ann
y
e
dge
det
ec
t
i
o
n
oper
at
i
o
n i
s
per
f
or
m
ed b
y
us
i
ng t
he ab
ov
e
-
c
l
us
t
er
e
d i
m
age (
F
i
gur
e
2)
w
hi
c
h i
s
d
er
i
v
ed b
y
us
i
ng
C
al
i
ber
f
u
z
z
y
c
-
m
eans
c
l
us
t
er
i
ng
al
gor
i
t
hm
.
A
f
t
er
appl
y
i
n
g c
an
n
y
edg
e
de
t
ec
t
i
on,
t
he r
es
u
l
t
an
t
i
m
age
i
s
an
ed
ge
det
ec
t
ed
i
m
age.
I
n
F
i
g
ur
e
4,
w
i
t
ho
u
t
c
l
us
t
er
i
ng
i
m
a
ge
i
s
d
er
i
v
e
d
b
y
a
ppl
y
i
ng
c
ann
y
edg
e
det
ec
t
or
t
o
t
he
or
i
gi
nal
i
m
age
and
w
i
t
h
c
l
us
t
er
i
ng
i
m
age
der
i
v
ed
b
y
app
l
y
i
ng
c
a
nn
y
edge
det
ec
t
or
t
o
a c
a
l
i
b
er
f
u
z
z
y
c
-
m
eans
c
l
us
t
er
ed
i
m
age.
(
a)
(
b)
F
i
gur
es
4
.
(
a)
W
i
t
hout
C
l
us
t
er
i
ng (
b)
w
i
t
h C
l
us
t
er
i
ng
3.
E
xp
er
i
m
en
t
an
d
R
esu
l
t
s
T
hi
s
di
v
i
s
i
on
i
l
l
us
t
r
at
es
t
he
det
a
i
l
s
a
bout
t
he
ex
per
i
m
ent
s
c
ond
uc
t
ed
us
i
ng
pr
o
pos
ed
al
g
or
i
t
hm
.
T
he pr
opos
ed
al
g
or
i
t
hm
i
s
i
m
pl
em
ent
ed on t
he Ma
t
l
a
bR
2
012
b w
i
t
h t
he B
er
k
el
e
y
S
egm
ent
at
i
on
D
at
as
et
(
B
S
D
S
)
i
m
age.
B
S
D
S
c
ons
i
s
t
s
of
100 t
es
t
i
m
ages
a
nd 2
00 t
r
a
i
n
i
n
g
i
m
ages
w
h
i
c
h ar
e
us
ed f
or
t
hi
s
pr
o
pos
ed
w
or
k
.
F
or
al
l
ex
per
i
m
ent
s
,
t
he m
ax
i
m
um
a
m
ount
of
i
t
er
at
i
ons
i
s
s
et
t
o 5
0,
t
h
e t
er
m
i
nat
i
on c
on
di
t
i
o
n i
s
0.
00
01
(
F
i
g
ur
e 5)
.
T
he ex
per
i
m
ent
r
es
ul
t
s
ar
e
v
i
s
u
al
i
z
ed o
ne.
F
i
gur
e
5.
F
u
nc
t
i
o
na
l
d
i
agr
a
m
s
f
or
a pr
opos
e
d m
et
hod
Evaluation Warning : The document was created with Spire.PDF for Python.
I
SSN
:
25
02
-
4
752
I
J
E
EC
S
V
o
l.
8
,
N
o.
1,
O
c
t
o
ber
20
17
:
5
9
–
6
8
64
(
i)
(
ii)
(
iii
)
(
i
v)
F
i
gur
e 6.
-
P
l
a
ne (
3
096)
(
i)
(
ii)
(
iii)
(
i
v)
F
i
gur
e 7.
-
E
ag
l
e (
4
204
9)
(
i)
(
ii)
(
iii)
(
i
v)
F
i
gur
e 8.
-
V
es
s
el
(
22
709
2)
(
i)
(
ii)
(
iii)
(
i
v)
F
i
gur
e 9.
-
H
um
an (
189
080)
(
i)
(
ii)
(
iii)
(
i
v)
F
i
gur
e 10.
N
at
ur
e (
176
03
5)
Evaluation Warning : The document was created with Spire.PDF for Python.
I
J
E
EC
S
IS
S
N
:
2
502
-
4
752
A
n
E
dg
e E
x
p
os
ur
e us
i
ng
C
al
i
ber
F
u
z
z
y
C
-
mea
ns
W
i
t
h C
anny
A
l
gor
i
t
hm
(
G
ow
r
i
J
e
y
ar
aman
)
65
E
dg
e de
t
ec
t
i
on:
(
i
)
I
np
ut
i
m
age,
(
i
i
)
C
al
i
ber
F
u
z
z
y
C
-
m
eans
C
l
us
t
er
e
d i
m
age,
(
i
i
i
)
C
ann
y
i
m
age,
(
i
v
)
B
ot
h C
a
l
i
ber
F
u
z
z
y
C
-
m
eans
C
l
us
t
er
i
n
g an
d C
an
n
y
S
om
e
s
a
m
pl
e
ou
t
put
s
gen
er
at
ed
b
y
t
h
i
s
t
ec
hn
i
q
ue
ar
e
s
ho
w
n
ab
ov
e
(
F
i
g
ur
e
6
-
1
0
)
.
B
y
s
eei
n
g
t
h
at
,
an
y
o
ne
c
a
n u
nder
s
t
an
d
a
n
ov
er
ed
ge
d
et
ec
t
ed
i
s
av
oi
ded
i
n
pr
o
p
os
ed
a
l
g
or
i
t
hm
.
C
om
par
e t
o c
ann
y
a
l
g
or
i
t
h
m
,
t
he pr
opos
ed
al
gor
i
t
hm
det
ec
t
s
f
e
w
er
ed
ges
.
T
he per
f
or
m
anc
e
ev
al
uat
i
o
n i
s
done us
i
n
g P
S
N
R
an
d MS
E
.
T
he peak
s
i
gnal
t
o noi
s
e
r
at
i
o (
P
S
N
R
)
i
s
t
he pr
opor
t
i
on bet
w
ee
n t
he m
ax
i
m
u
m
pos
s
i
bl
e v
al
u
e of
a pi
x
el
a
n
d t
he i
nf
l
uenc
e
of
al
t
er
i
n
g
no
i
s
e
t
hat
c
h
ang
es
t
h
e
v
a
l
u
e
of
i
t
s
r
epr
es
e
nt
at
i
on.
I
t
i
s
us
ual
l
y
m
eas
u
r
ed
i
n
t
er
m
s
o
f
t
he l
o
gar
i
t
hm
i
c
dec
i
bel
s
c
al
e.
T
he
m
ean s
quar
ed
er
r
or
(
MS
E
)
i
s
us
ed t
o
ev
a
l
uat
e t
he
per
f
or
m
anc
e of
a pr
edi
c
t
or
.
E
v
en t
h
oug
h ex
per
i
m
ent
r
es
ul
t
i
s
t
h
e v
i
s
ua
l
on
e,
her
e
MS
E
an
d P
S
N
R
v
al
ues
f
or
t
he
pr
opos
e
d al
gor
i
t
hm
ar
e l
i
s
t
ed i
n T
ab
l
e 1.
P
S
N
R
a
nd M
S
E
v
a
l
u
e of
pr
opos
e
d m
et
hod ar
e
s
i
gni
f
i
c
an
t
l
y
b
et
t
er
t
ha
n
t
r
a
di
t
i
on
al
c
a
nn
y
a
nd
Log
a
l
g
or
i
t
hm
s
.
A
nd
al
s
o
t
he
c
om
par
i
s
on
r
es
ul
t
s
f
or
t
r
adi
t
i
ona
l
c
an
n
y
an
d l
o
g al
gor
i
t
hm
s
ar
e l
i
s
t
ed
be
l
o
w
(
T
abl
e 2
and F
i
g
ur
e 1
1)
.
T
abl
e 1.
P
r
op
os
ed
A
l
gor
i
t
h
m
R
es
ul
t
C
om
par
i
s
on
T
abl
e 2.
C
om
par
i
s
on
w
i
t
h
C
ann
y
a
nd
Log
R
es
ul
t
I
m
age
PSN
R
MS
E
T
im
e
P
l
ane
66.
305
0.
02
24.
554
E
agl
e
62.
9578
0.
03
13.
915
V
es
s
el
61.
521
0.
05
234.
24
H
um
an
63.
0686
0.
03
15.
334
N
at
ur
e
62.
6492
0.
04
11.
731
E
l
ephant
59.
7611
0.
07
17.
269
E
dge D
et
e
c
t
i
on
c
o
m
par
i
s
on
w
i
t
h o
t
her
T
ec
hni
que
s
I
m
age
C
anny
Log
PSN
R
MS
E
PSN
R
MS
E
P
l
ane
62.
364
0.
04
65.
7959
0.
02
E
agl
e
61.
5277
0.
05
62.
661
0.
04
V
es
s
el
58.
2106
0.
10
59.
4014
0.
08
H
um
an
59.
423
0.
07
61.
795
0.
04
N
at
ur
e
57.
6902
0.
11
60.
3688
0.
06
E
l
ephant
57.
385
0.
12
59.
489
0.
07
F
i
gur
e 11.
P
S
N
R
v
a
l
ue
T
he
c
l
us
t
er
c
ent
r
o
i
d
v
a
l
u
e
es
t
i
m
at
i
on
i
s
bas
e
d
o
n
c
or
r
el
at
i
on
di
s
t
anc
e
c
a
l
c
ul
at
i
o
n,
an
d
t
hat
v
a
l
ue
c
on
v
er
ges
t
o t
h
e
f
i
x
ed po
i
nt
unt
i
l
t
he
er
r
or
v
al
u
e i
s
l
es
s
t
ha
n 0.
000
1.
T
he b
el
o
w
t
a
bl
e
deno
t
es
t
h
e
c
on
v
er
g
enc
e
of
c
l
us
t
er
c
ent
r
o
i
d
v
a
l
u
e
s
dur
i
n
g
i
t
er
at
i
on
an
d
f
i
n
al
l
y
i
t
r
e
ac
hed
s
at
u
r
at
i
on po
i
nt
.
T
he c
l
us
t
e
r
c
ent
r
oi
d v
al
u
e c
onv
er
ge
n
c
e i
s
l
i
s
t
ed out
i
n T
abl
e 3 a
nd t
he n
um
ber
of
c
l
us
t
er
s
c
hanges
t
hat
af
f
ec
t
ed t
he
P
S
N
R
v
al
ue
i
s
s
ho
w
n
i
n F
i
gur
e
12.
T
abl
e 3.
P
r
op
os
ed
A
l
gor
i
t
h
m
C
l
us
t
er
C
ent
r
oi
d R
es
u
l
t
C
on
v
er
ge
nc
e
I
m
age
C
l
us
t
er
c
ent
r
i
od
v
a
l
ue
v
ar
i
at
i
on i
n
ev
er
y
i
t
er
at
i
on
P
l
ane
103.
871
91.
6634
81.
8104
72.
2520
43.
5809
E
agl
e
66.
2412
62.
1441
61.
7362
61.
6949
61.
6917
V
es
s
el
93.
0232
78.
0792
70.
1599
68.
5935
68.
5896
H
um
an
69.
4017
74.
3634
75.
0307
75.
1312
75.
1345
N
at
ur
e
67.
3549
70.
2766
70.
3861
70.
4128
70.
4194
E
l
ephant
73.
6764
69.
9866
68.
2896
67.
9683
67.
9403
Evaluation Warning : The document was created with Spire.PDF for Python.
I
SSN
:
25
02
-
4
752
I
J
E
EC
S
V
o
l.
8
,
N
o.
1,
O
c
t
o
ber
20
17
:
5
9
–
6
8
66
F
i
gur
e
12.
N
um
ber
of
C
l
us
t
er
v
er
s
us
P
S
N
R
v
al
u
e
I
n
edg
e
ex
pos
ur
e
,
pr
ec
i
s
i
o
n
i
s
us
ed
t
o
r
et
r
i
e
v
e
t
h
e
p
i
x
el
s
t
hat
ar
e
r
e
l
at
ed
E
qu
at
i
on
(
5)
.
T
he R
ec
al
l
i
s
t
he
r
el
ev
ant
pi
x
e
l
s
t
h
at
ar
e
r
et
r
i
e
v
ed
E
qu
at
i
on (
6)
.
A
c
al
c
u
l
at
i
o
n t
h
at
m
er
ges
pr
ec
i
s
i
o
n an
d r
ec
a
l
l
ar
e k
now
n as
a
n F
-
m
eas
ur
e E
qua
t
i
on (
7)
.
P
r
e
c
is
io
n
=
+
(
5)
R
ec
al
l
=
+
(
6)
F
m
eas
ur
e
=
2
∗
∗
+
(
7)
P
r
ec
i
s
i
o
n,
r
ec
al
l
an
d F
-
m
eas
ur
e ar
e us
ed t
o m
eas
ur
e t
he per
f
or
m
anc
e o
f
under
s
t
and
i
ng
i
n a
n i
m
age b
as
ed
on c
o
l
or
v
a
l
ue
.
F
or
a
s
am
pl
e of
t
hr
ee i
m
ages
,
p
er
f
or
m
anc
es
ar
e l
i
s
t
ed b
el
o
w
i
n T
abl
e 4.
P
r
ec
i
s
i
on
v
er
s
u
s
r
ec
al
l
gr
ap
hi
c
a
l
r
epr
es
en
t
at
i
on
f
or
v
ar
i
o
us
al
g
or
i
t
hm
s
i
s
s
how
n
i
n
F
i
gur
e 13.
T
abl
e 4.
P
er
f
or
m
anc
e (
P
r
ec
i
s
i
on,
R
ec
al
l
,
and F
-
m
eas
ur
e)
f
or
P
r
opos
e
d
an
d S
ob
el
A
lg
o
r
it
h
m
ID
M
et
hod
P
r
ec
i
s
i
on
R
ec
al
l
F
-
m
eas
ur
e
E
agl
e
P
r
opos
ed
0.
2545
0.
40246
0.
47
S
obel
0.
2497
0.
29625
0.
46
H
um
an
P
r
opos
ed
0.
1134
0.
17285
0.
28
S
obel
0.
3038
0.
29487
0.
50
N
at
ur
e
P
r
opos
ed
0.
1252
0.
17761
0.
23
S
obel
0.
4743
0.
48740
0.
20
F
i
gur
e 13.
P
r
ec
i
s
i
on
v
er
s
us
r
ec
al
l
Evaluation Warning : The document was created with Spire.PDF for Python.
I
J
E
EC
S
IS
S
N
:
2
502
-
4
752
A
n
E
dg
e E
x
p
os
ur
e us
i
ng
C
al
i
ber
F
u
z
z
y
C
-
mea
ns
W
i
t
h C
anny
A
l
gor
i
t
hm
(
G
ow
r
i
J
e
y
ar
aman
)
67
F
or
t
hi
s
ex
per
i
m
ent
t
ot
al
l
y
100
B
S
D
S
i
m
ages
w
er
e
s
el
ec
t
ed
w
h
i
c
h
w
er
e
bas
e
d
o
n
t
hei
r
s
egm
ent
s
and c
ol
or
c
ount
s
.
I
n e
ac
h a
nd e
v
er
y
c
at
e
go
r
y
,
t
w
o
i
m
ages
w
er
e t
ak
en f
or
ex
per
i
m
ent
pur
pos
e.
T
hos
e r
es
u
l
t
s
ar
e
s
ho
w
n
i
n T
abl
e 5.
T
abl
e
5.
P
er
f
or
m
anc
e on 1
00 B
S
D
S
T
es
t
s
I
m
ages
f
or
P
r
opos
e
d M
et
h
od
M
et
hod
P
r
ec
i
s
i
on
R
ec
al
l
F
-
m
eas
ur
e
P
r
opos
ed
0.
1734 ±
0
.
002
0.
3761±
0.
0015
0.
2418±
0.
00155
S
obel
0.
1736
0.
3576
0.
2337
4
.
C
o
n
c
l
u
s
i
o
n
T
he pr
opos
ed u
ni
que
edg
e
ex
pos
ur
e ap
pr
oac
h
i
n
B
S
D
S
i
m
ages
us
i
n
g C
a
l
i
b
er
F
u
z
z
y
c
-
m
eans
w
i
t
h c
an
n
y
al
g
or
i
t
hm
c
an
m
a
k
e up t
he c
o
m
ponent
of
s
of
t
c
o
m
put
i
ng
w
h
i
c
h ar
e
appr
o
pr
i
at
e
f
or
gr
i
ppi
n
g
t
he
i
s
s
ues
r
el
at
ed
t
o
t
he
u
nder
s
t
anda
bi
l
i
t
y
of
i
nf
or
m
at
i
on.
I
t
c
an
pr
o
v
id
e
appr
ox
i
m
at
e s
ol
u
t
i
o
ns
f
as
t
er
t
han a
n
y
ot
her
m
et
hods
.
B
er
k
el
e
y
dat
a s
e
t
s
w
er
e us
ed t
o c
om
par
e
t
he
p
er
f
or
m
anc
e
of
t
r
adi
t
i
o
nal
and
t
he
pr
op
os
ed
al
gor
i
t
hm
s
.
E
x
per
i
m
ent
al
r
es
u
l
t
s
s
ho
w
t
h
at
t
he
c
o
m
put
at
i
on
c
os
t
i
s
r
educ
e
d
an
d
got
bet
t
er
p
er
f
or
m
anc
e
b
y
di
s
c
o
v
er
i
n
g
a
s
u
per
i
or
s
et
of
ear
l
y
c
l
us
t
er
c
ent
er
s
.
I
n t
h
i
s
pa
p
er
,
C
al
i
b
er
f
u
z
z
y
c
-
m
eans
al
g
or
i
t
hm
s
w
er
e m
eas
ur
ed
up
w
i
t
h
i
n
t
he
c
ont
ex
t
of
c
ol
or
qu
ant
i
z
at
i
o
n.
I
n
f
ut
ur
e,
t
he
n
oi
s
e
w
i
l
l
be
ad
ded
i
n
i
m
ages
bound
ar
y
an
d
w
i
l
l
be
f
ur
t
her
ana
l
y
z
e
d.
R
ef
er
en
ces
[1
]
Li
Z
hen
gz
hou,
Li
u M
ei
,
W
a
ng
H
ui
gai
,
Y
ang Y
ang,
C
hen J
i
n,
J
i
n G
ang.
G
r
ay
-
s
c
al
e E
d
ge D
e
t
ec
t
i
on and
I
m
ag
e S
eg
m
ent
at
i
on A
l
gor
i
t
h
m
B
as
ed o
n M
ean S
hi
f
t
.
T
EL
K
O
M
N
I
KA
.
2013;
11(
3)
:
1
414
-
1
421.
[2
]
X
i
ao
F
eng
,
G
uo
Li
,
G
uo
L
i
na.
A
N
ew
S
ub
-
p
i
x
el
E
dg
e
D
et
e
c
t
i
on
M
et
hod
of
C
ol
or
I
m
age
s
.
T
E
LK
O
M
N
I
K
A
I
ndone
s
i
a
n J
ou
r
nal
o
f
E
l
e
c
t
r
i
c
al
E
ngi
neer
i
ng
.
2014;
12(
5)
:
36
09
-
36
15.
[3
]
B
oon
-
S
en
g C
h
ew
,
C
hau L
-
P
,
Ki
m
-
H
ui
Y
ap.
A
F
uz
z
y
C
l
us
t
er
i
ng
A
l
g
or
i
t
h
m
f
or
V
i
r
t
ua
l
C
har
a
c
t
er
A
ni
m
a
t
i
on
R
e
pr
es
ent
a
t
i
o
n
.
I
EEE
T
ra
n
s
a
c
t
i
ons
o
n
M
ul
t
i
m
ed
i
a
.
20
11;
13(
1)
:
4
0
-
4
9
,
D
O
I:
10.
11
09/
T
M
M
.
2010.
20
8251
2.
[4
]
J
af
er
z
ad
eh
K
,
K
i
ani
K
,
M
oz
af
f
ar
i
S
.
A
c
c
el
er
a
t
i
o
n
of
F
r
a
c
t
al
I
m
ag
e
C
om
pr
es
s
i
on
U
s
i
ng
F
uz
z
y
C
l
us
t
er
i
ng
and
D
i
sc
r
e
t
e
-
c
os
i
ne
-
tr
a
n
s
fo
r
m
-
ba
s
ed
Me
t
r
i
c
.
I
E
T
I
m
age P
r
o
c
e
ssi
n
g
. 2
0
1
2;
6(
7)
:
1
024
-
1030,
D
O
I
:
10.
1
049/
i
et
-
i
pr
.
20
11.
01
81.
[5
]
E
v
ans
A
N
,
Li
u
X
U.
A
M
or
phol
ogi
c
al
gr
adi
ent
A
ppr
oac
h
to
C
o
l
or
E
dge
D
et
ec
t
i
on
.
I
EEE
T
ra
n
s
a
c
t
i
o
n
s
o
n
I
m
ag
e P
r
oc
es
s
i
n
g.
2
006;
15(
6
)
:
1454
-
1463
.
[6
]
Kri
n
i
d
i
s
S,
C
h
a
t
z
i
s
V.
A
R
o
bu
s
t
F
uz
z
y
Loc
al
I
nf
or
m
at
i
on
C
-
m
eans
C
l
u
s
t
er
i
ng A
l
gor
i
t
hm
.
I
EEE T
ra
n
s
.
I
m
ag
e P
r
oc
es
s
i
n
g,
M
ay
2010;
19(
5)
:
1
328
-
1337
.
[7
]
D
ong G
,
X
i
e
M
.
C
ol
or
C
l
us
t
er
i
ng
an
d
Lear
ni
ng
fo
r
I
m
a
ge S
egm
e
nt
at
i
on.
B
as
e
d on N
eur
al
N
et
w
or
k
s
.
I
E
E
E
T
r
an
s
ac
t
i
on
s
o
n N
eur
a
l
N
et
w
or
k
s
.
20
05;
1
6(
40)
:
92
5
–
936.
[8
]
M
aoguo G
ong,
Y
an
Li
an
g,
J
i
a
o
S
hi
,
W
e
n
pi
n
g
M
a,
J
i
ngj
i
ng
M
a.
F
u
z
zy
C
-
M
eans
C
l
us
t
e
r
i
ng
w
i
t
h
L
oc
al
I
nf
or
m
at
i
on and K
er
nel
M
et
r
i
c
f
or
I
m
age s
e
gm
ent
at
i
on
.
I
E
E
E
T
r
ans
ac
t
i
o
ns
on I
m
ag
e
P
r
oc
es
s
i
n
g,
2013;
22(
2)
:
573
-
584,
D
O
I
:
10.
1109/
T
I
P
.
20
12.
2
219
547.
[9
]
N
oguei
ra
T
M
,
de A
r
r
uda C
a
m
ar
g
o
H
.
F
uz
z
y
R
ul
es
G
e
ne
r
at
i
on
us
i
ng C
ondi
t
i
o
nal
C
l
us
t
er
i
ng.
Lat
i
n
A
m
er
i
c
a
T
r
ans
a
c
t
i
o
ns
,
I
EE
E
(R
e
v
i
s
t
a
I
EEE
A
m
er
i
c
a
Lat
i
na)
,
2
009;
7(
4)
:
463
-
4
7
1
,
D
O
I:
10.
11
09/
T
LA
.
200
9.
5
3490
46.
[1
0
]
D
zi
u
k
M
A
,
Ja
msh
i
d
i
M
.
F
u
z
zy
Logi
c
C
ont
r
o
l
l
ed
U
AV
A
ut
opi
l
ot
U
s
i
ng
C
-
M
ean
C
l
us
t
er
i
ng
. 6
th
I
nt
er
na
t
i
o
nal
C
o
nf
er
e
nc
e on S
y
s
t
em
o
f
S
y
s
t
e
m
s
E
ngi
n
eer
i
ng
(
S
oS
E
)
,
20
11;
3
05
–
310.
[1
1
]
C
hen
Y
u
-
Me
i
.
R
es
ear
c
h
on
t
h
e
I
nc
o
m
e
of
R
ur
al
R
e
s
i
d
ent
s
of
S
i
c
hu
an
P
r
ov
i
n
c
e
B
as
e
d
o
n
F
uz
z
y
C
-
m
ean C
l
us
t
er
i
n
g.
F
S
K
D
'
09.
S
i
x
t
h I
nt
er
nat
i
ona
l
C
onf
er
en
c
e on
F
uz
z
y
S
y
s
t
em
s
an
d K
now
l
edge
D
i
s
c
ov
er
y
,
2
009;
3:
15
1
-
1
55,
D
O
I
:
10.
11
09/
F
S
K
D
.
2
009.
235
[1
2
]
T
ar
aghi
B
,
J
al
a
l
i
V
R.
F
u
z
zy
c
-
m
ean (
F
C
M
)
C
l
us
t
er
i
ng
and G
enet
i
c
A
l
gor
i
t
hm
C
apa
bi
l
i
t
y
in
P
r
e
d
ic
t
i
ng
S
at
ur
at
e
d H
y
dr
aul
i
c
C
ondu
c
t
i
v
i
t
y
.
I
nt
er
nat
i
onal
C
ongr
es
s
on T
ec
hn
ol
o
gy
,
C
om
m
un
i
c
at
i
on
and
K
now
l
edge (
I
C
T
C
K
)
,
201
4;
1
-
6,
D
O
I
:
10.
1109
/
I
C
T
C
K
.
2014.
70335
21
[1
3
]
K
i
ani
S
,
M
oghad
dam
M
E.
F
r
ac
t
al
B
as
e
d
D
i
gi
t
al
I
m
a
ge
W
a
t
e
r
m
ar
k
i
ng
U
s
i
n
g
F
uz
z
y
C
-
M
ean
C
l
us
t
er
i
ng
.
I
nt
er
na
t
i
o
nal
C
o
nf
er
e
nc
e
on
I
nf
or
m
at
i
o
n
M
anagem
e
nt
an
d
E
ngi
ne
er
i
n
g,
I
C
I
M
E
'
09,
2009
;
638
-
6
42,
D
O
I
:
10.
11
09/
I
C
I
M
E
.
2009.
7
2
[1
4
]
M
oham
m
ad
H
ad
i
A
ar
a
bi
,
A
nahi
t
a
F
at
hi
K
az
er
o
oni
,
P
ar
i
s
a
K
ha
t
er
i
,
M
oham
m
ad
R
ez
a
A
y
and
H
am
i
dr
ez
a S
al
i
gh
eh R
ad.
A
R
obus
t
MR
B
as
ed
A
t
t
enuat
i
on M
ap
G
ener
at
i
o
n
i
n s
hor
t
-
T
E
M
R
i
m
age
s
of
t
he
H
e
ad
E
m
pl
oy
i
ng
H
y
br
i
d
S
pat
i
al
F
uz
z
y
C
-
M
eans
C
l
u
s
t
er
i
ng
and
In
t
e
n
s
i
ty
in
H
om
og
ene
i
t
y
C
or
r
ec
t
i
on
.
E
J
N
M
M
I
P
hy
s
i
c
s
,
2014;
D
O
I
:
10.
1
186/
219
7
-
736
4
-
1
-
S1
-
A
4
9.
Evaluation Warning : The document was created with Spire.PDF for Python.
I
SSN
:
25
02
-
4
752
I
J
E
EC
S
V
o
l.
8
,
N
o.
1,
O
c
t
o
ber
20
17
:
5
9
–
6
8
68
[1
5
]
Q
uan
W
e
n,
M
E
m
r
e C
el
eb
i
.
H
ar
d
V
e
r
su
s
F
u
zzy C
-
m
eans
c
l
us
t
er
i
ng
f
o
r
C
ol
o
r
Q
u
a
nt
i
z
a
t
ion
.
EU
R
ASI
P
J
our
n
al
o
n A
d
v
an
c
es
i
n S
i
gn
al
P
r
oc
e
s
s
i
ng
,
201
1:
1
-
12.
D
O
I:
10.
11
86/
1
687
-
6
180
-
201
1
-
118
[1
6
]
Y
u
J
,
Y
ang
M
S
.
O
p
ti
m
a
l
i
ty
Te
s
t
f
o
r
G
ener
al
i
z
ed
F
C
M
and i
t
s
A
ppl
i
c
at
i
o
n
to
P
ar
am
et
er
S
el
e
c
t
i
o
n
.
I
EEE
T
r
ans
ac
t
i
on
s
F
uz
z
y
S
y
s
t
em
s
.
2005;
13:
1
64
–
176.
[1
7
]
W
u K
L,
Ya
n
g M
S
.
A
l
t
er
nat
i
v
e c
-
m
e
ans
c
l
u
s
t
er
i
ng
a
l
gor
i
t
h
m
s
.
P
at
t
er
n R
e
c
ogn
i
t
i
on
,
2
002
;
35
:
22
67
–
2278.
Evaluation Warning : The document was created with Spire.PDF for Python.