TELKOM
NIKA
, Vol. 13, No. 4, Dece
mb
er 201
5, pp. 1337
~1
342
ISSN: 1693-6
930,
accredited
A
by DIKTI, De
cree No: 58/DIK
T
I/Kep/2013
DOI
:
10.12928/TELKOMNIKA.v13i4.1178
1337
Re
cei
v
ed Au
gust 27, 20
15
; Revi
sed
No
vem
ber 1
3
, 2015; Accepte
d
No
vem
ber
27, 2015
A Novel Image Segmentation Algorithm Based on
Graph Cut Optimization Problem
Zhang Gu
an
g-hua
*, Xiong Zhong-y
a
n
g
, Li Kuan, Xing Chang
-
y
u
an, Xia Shu-
y
i
n
Cho
ngq
in
g Uni
v
ersit
y
, Ch
on
g
q
in
g Cit
y
,
4
0
0
0
30, Chi
n
a
*Corres
p
o
ndi
n
g
author, e-ma
i
l
: guan
gh
ua.zh
ang
@outl
ook.c
om
A
b
st
r
a
ct
Imag
e s
e
g
m
en
tation, a
fun
d
a
m
e
n
tal
task
in
co
mp
uter v
i
si
on, h
a
s
be
en
w
i
dely
use
d
i
n
rece
nt
years in
many
fields. De
al
in
g w
i
th the gra
ph cu
t o
p
ti
mi
zation
prob
le
m
obtai
ns the i
m
age s
e
g
m
e
n
tat
i
o
n
results. In this study, a novel al
gorith
m
w
i
th
w
e
ighted g
r
aphs w
a
s co
nstructed to solve the i
m
ag
e
seg
m
e
n
tatio
n
prob
le
m
thr
o
u
gh mini
mi
z
a
tio
n
of an
en
erg
y
functio
n
. A
b
i
nary v
e
ctor
of the s
e
g
m
e
n
ta
tio
n
lab
e
l w
a
s defi
n
ed to d
e
scrib
e
both the for
e
g
r
oun
d an
d
the
backgr
oun
d of
an i
m
a
ge. T
o
de
mo
nstrate th
e
effectiveness
o
f
our prop
osed
meth
od, four
vario
u
s ty
pes
of images w
e
r
e
used to co
n
s
truct a series
of
exper
iments. E
x
peri
m
e
n
tal r
e
sults ind
i
cate t
hat co
mp
ared
w
i
th other methods, the
prop
osed
alg
o
rith
m can
effectively pr
o
m
ote th
e qu
al
ity of ima
ge
seg
m
e
n
ta
tio
n
und
er three
p
e
rformanc
e ev
alu
a
tion
metri
cs,
na
me
ly, misc
la
ssificatio
n
error
rate, rate of the nu
mber
of b
a
ckgrou
nd p
i
xel
s
, and the r
a
tio
of the nu
mber
of
w
r
ongly class
i
fi
ed foregr
ou
nd
pixels.
Ke
y
w
ords
: Image Se
g
m
e
n
tation, Graph C
u
t, Energy functi
o
n
, Pixel
Copy
right
©
2015 Un
ive
r
sita
s Ah
mad
Dah
l
an
. All rig
h
t
s r
ese
rved
.
1. Introduc
tion
Image e
ngin
eerin
g is a ri
ch
re
sea
r
ch
area
that ha
s been
explored for m
any
years [1].
Re
cently, re
searche
r
s hav
e cla
ssifie
d
image e
ngine
ering i
n
to three dom
ain
s
, namely, imag
e
pro
c
e
ssi
ng, i
m
age a
nalysi
s
, and im
age
comp
re
hen
si
on. Gen
e
rally
, image seg
m
entation, which
has
bee
n wi
dely used in
many different appli
c
ati
ons, i
s
a
ke
y issu
e in th
e field of im
age
pro
c
e
ssi
ng. Image segme
n
tation refe
rs to the divi
sion of an image
into several
sep
a
rate
regi
ons
to represent variou
s types
of object
s
[2, 3].
Image proce
ssi
ng spe
c
ifically aims to impl
eme
n
t the pattern
recognition p
r
o
c
ess, and
image
segm
entation is
a basi
c
work in pattern
recognitio
n
and compu
t
er vision. I
m
age
segm
entation
sepa
rate
s a
n
image into
several re
gio
n
s an
d then
provide
s
d
e
scriptio
ns to th
ese
regio
n
s [4]. Using
ce
rtain similar criteri
a
of low-le
vel vi
sual featu
r
e
s
,
such as
col
o
r, texture, an
d
sha
pe, imag
e segm
entati
on is defin
ed
as the
co
nversi
on of a d
i
gital image to several no
n-
overlap
p
ing region
s, whi
c
h
result in obj
e
c
ts in imag
es
[5, 6].
Unfortu
nately
,
the visual content
s of t
he image
s are characterize
d by d
i
versity,
compl
e
xity, and ra
ndom
ne
ss; a
n
in-dep
th unde
rsta
n
d
ing of the i
n
ternal vi
sion
mech
ani
sm
of
peopl
e al
so
remai
n
s l
a
cking [7]. To t
he be
st of
our
kn
owle
d
ge, no m
a
tu
re
segm
enta
t
ion
approa
ch exi
s
ts to sati
sfy all the requi
re
ments for a
p
p
lication e
n
viro
nments.
Becau
s
e
of t
he la
ck of p
r
i
o
r info
rmatio
n ab
out
obj
e
c
ts i
n
a
n
ima
ge, providing
accu
rate
image
se
gme
n
tation results is difficult
with the u
s
e of
existing m
e
thod
s. The
de
velopment
of an
effective and
highly accu
ra
te segme
n
tation app
roa
c
h i
s
theref
ore im
portant.
2. State o
f
th
e Art
Image se
gm
entation is a
high-level a
pplication tha
t
has bee
n widely u
s
ed
in many
fields,
such as remote comm
uni
cation, military
,
remote
sensi
ng, met
eorol
ogy, im
age
processi
ng, and intelligent
transportation, to nam
e a few. In
thi
s
section,
we discuss rel
a
ted
works abo
ut image segm
e
n
tation.
Wan
g
et
al. pre
s
ente
d
a
robu
st an
d eff
i
ci
ent
app
roa
c
h to
segme
n
t image
s
wit
h
limited
and intuitive use
r
interacti
on; the pro
p
o
s
ed al
gor
ith
m
integrate
d
g
eode
si
c dista
n
ce info
rmati
on
with the flexibility of level set met
hod
s in energy minimi
zation,
whi
c
h depends
on
compl
e
me
nt
a
r
y
st
ren
g
t
h
s [
8
]
.
Evaluation Warning : The document was created with Spire.PDF for Python.
ISSN: 16
93-6
930
TELKOM
NIKA
Vol. 13, No
. 4, Decem
b
e
r
2015 : 133
7 – 1342
1338
Han
et al. propo
sed
a ne
w mo
del to d
eal with th
e i
m
age
seg
m
e
n
tation p
r
obl
em. Thi
s
model was
constructe
d th
roug
h analy
s
i
s
of the ch
ar
acteri
stic
of textile/fabric i
m
age
s. The
main
innovation
in
this
wo
rk is it
s in
co
rp
oratio
n of
a
ca
rtoo
n-an
d-texture
de
com
p
o
s
ition p
r
o
c
e
s
s in
to
the mod
e
l, a
s
well
a
s
its bi
as fiel
d fun
c
ti
on d
e
si
gne
d t
o
e
s
timate th
e deviatio
n
d
egre
e
b
e
twe
e
n
the cart
oon i
m
age an
d the
piece
w
i
s
e co
nstant ap
prox
imation [9].
Miao et
al. p
r
opo
sed
a
ne
w al
gorith
m
t
o
segm
e
n
t la
rge top
ograph
ic ma
ps ba
se
d on
the
ideas of fuzzy theory, randomiz
ed
sam
p
ling, and m
u
ltilevel imag
e fusi
on. A l
a
rge topographic
map
wa
s ran
domly sampl
ed first. Then
, the optimal
clu
s
terin
g
ce
nters were
a
c
quire
d
with fu
zzy
c-m
ean
s
(F
CM)
clu
s
teri
ng
. Multilevel i
m
age
fusi
on
wa
s develo
ped to
fu
se
the segm
ent
ed
image
s into the final se
gm
entation map
s
[10].
To enh
an
ce
the quality of image se
gmentat
ion,
Yu et al. propo
sed a m
e
thod to
con
s
tru
c
t a
gene
rali
zed f
u
zzy co
mple
ment; the au
thors al
so d
e
velope
d a
gene
rali
zed f
u
zzy
compl
e
me
nt operator, whi
c
h ha
s a go
o
d
prop
erty fo
r param
eter o
p
timization in
real ap
plicati
ons
[10].
Aside f
r
om t
he afo
r
em
en
tioned
wo
rks, other
meth
ods have
be
en u
s
e
d
for image
segm
entation
,
such a
s
in
corpo
r
ating
adaptive lo
cal inform
atio
n into fuzzy
clu
s
terin
g
[
12],
arbitrary
noi
se mo
del
s vi
a solution
of
minimal
surface
problem
s [13],
clu
s
t
e
ring
te
chniq
ue
optimize
d
by cu
ckoo se
arch [14], modified Gau
ssi
an
mixture mode
ls inco
rp
orati
ng local
spati
a
l
informatio
n [15], conditio
n
a
l rand
om fie
l
d learni
ng with convolutio
nal neu
ral n
e
twork featu
r
es
[16], dynami
c
incorp
oratio
n of
wavelet
filter in
F
C
M
[17,18], proliferation
ind
e
x evaluatio
n [
19],
and fuzzy acti
ve contou
r m
odel with
kernel metri
c
.
3. Methodol
og
y
for Image Segmenta
tion Base
d on Graph Cut
Optimiza
tion
From the ab
o
v
e analysi
s
, we can see that
image se
gmentation i
s
a fundament
al task in
comp
uter vi
sion. In this
se
ction, we
discu
ss th
e i
m
age
se
gme
n
tation p
r
obl
em, and
so
me
example
s
of image segm
e
n
tation are d
e
s
cribe
d
as foll
ows.
Figure 1. Examples of ima
ge se
gmentat
ion.
Suppo
se th
at
R
me
an
s the
whol
e regio
n
s
in
an
imag
e, and
then
segmentatio
n
results
are represent
ed by sepa
ra
ting
R
to several non-overla
pping sub
s
et
s, that is,
12
,,
,
N
R
RR
, a
nd
the followin
g
equatio
ns
sh
ould be
satisfi
ed.
1
N
i
i
R
R
(1)
,
ij
R
Rf
o
r
i
j
(2)
Let
,,
GV
E
W
be a g
r
a
ph, in whi
c
h
V
refers
to a
s
e
t of vertices
,
E
is a set of edge
s,
and
W
den
otes weig
hts of
graph edg
es. Grap
h cut
of
G
is to divid
e
the g
r
ap
h to
two
different
disjoi
nt sets, that is,
GX
Y
. After
w
ards,
cost f
unctio
n
of
gra
ph cut is d
e
fined a
s
follows.
,
mn
G
mX
n
Y
C
(3)
Evaluation Warning : The document was created with Spire.PDF for Python.
TELKOM
NIKA
ISSN:
1693-6
930
A Novel Im
ag
e Segm
entation Algorithm
Based o
n
Graph Cut Optim
i
zation…
(Z. Guang
-hu
a
)
1339
w
h
er
e
pa
ra
me
te
r
mn
refers t
o
wei
ght of e
dge
mn
E
. Assum
e
that graph
cut refers to
a
clo
s
ed
conto
u
r and di
screte formula i
s
e
x
ploited to cal
c
ulate the
co
ntour len
g
th.
Grap
h cut opt
imization p
r
o
b
lem ca
n be i
llustrate
d as f
o
llows.
12
1
,
C
i
i
M
inc
u
t
C
C
w
(4)
whe
r
e
12
,
CC
C
refers to a partition and sym
b
o
l
i
w
denotes th
e weig
ht betwee
n
the
edge
s which con
n
e
c
t
the
set
1
C
and the
se
t
2
C
. Furtherm
o
re, we
supp
ose that
12
,,
,
x
A
AA
A
is a
binary ve
cto
r
of se
gme
n
tation label
, and
i
A
den
otes a la
bel
. Thus,
A
repre
s
ent
s a
segm
entation
sch
eme an
d
it can descri
be both
fore
g
r
oun
d and ba
ckgro
und of the image to be
segm
ented. Hen
c
e,
an en
ergy
fun
c
tion
EA
for the imag
e seg
m
entati
on process is descri
bed
as
follows
.
E
AR
A
B
A
(5)
whe
r
e the foll
owin
g co
nditions
sho
u
ld b
e
satisfie
d.
1)
xx
xX
R
AR
A
(6)
2)
,
,
,
xy
xy
xy
N
B
AB
A
A
(7)
3)
0,
,
1,
xy
xy
xy
A
A
AA
A
A
(8)
w
h
er
e s
y
mbol
R
A
d
enote
s
a
regio
nal
prop
erties rep
r
e
s
entation, a
n
d
pa
ramete
r
is
use
d
to defin
e
the wei
ght of
R
A
. Afterward
s
, weight
bet
wee
n
two va
rious pixel
s
i
s
co
mpute
d
as
follows
.
1
,,,
1
,
2
ij
xy
ij
x
y
P
i
j
P
v
F
P
v
F
(9)
whe
r
e
,
P
ij
mea
n
s the
edge
prob
ability of the pixel
,
ij
, and
ij
P
vF
and
xy
P
vF
refers to the prob
abilitie
s whi
c
h are be
longe
d to pixel
,
ij
and
,
x
y
respectively. Usi
ng the
para
m
eter
,,
,
ij
x
y
, the ene
rgy fun
c
tion can be
comp
uted by the followin
g
equatio
n.
,
,,,
,,,
ln
,
ij
x
y
ij
ij
ij
ij
x
y
ij
x
y
B
i
j
x
y
f
f
Ef
P
v
f
Dv
v
(10
)
w
h
er
e
is d
e
fined
as a
co
nstant,
,,
,
ij
x
y
d
e
n
o
te
s
th
e pa
r
a
me
te
r
w
h
ic
h
is
us
ed
to
descri
be th
e i
m
porta
nce of
neig
hbo
rho
o
d
pixel
s
,
ij
and
,
x
y
. Then, im
ag
e segme
n
tation ta
sk
is
s
o
lved by optimiz
e this
energy func
tion.
4. Experiment and Resul
t
s An
aly
s
is
In ord
e
r to te
st the effe
ctivene
ss
of ou
r
prop
osed
al
g
o
rithm,
expe
ri
ments are de
sign
ed
to con
d
u
c
t p
e
rform
a
n
c
e
e
v
aluation. M
o
reove
r
, several im
age
segmentatio
n
approa
che
s
are
utilized, in
clu
d
ing: 1
)
Parzen
-wi
ndo
w
based th
re
sh
olding
(PWT), 2) Doubl
e-t
h
re
shol
d ima
g
e
binari
z
atio
n (DTIB),
3
)
Lo
cal gray
leve
l
differe
n
c
e
(LGL
D), and
4) No
rm
aliz
ed Cut (Nc
u
t).
In
particula
r, da
taset utilized
in this
experi
m
ent is
th
e
same a
s
p
ape
r, and
we
ch
oose fou
r
types
Evaluation Warning : The document was created with Spire.PDF for Python.
ISSN: 16
93-6
930
TELKOM
NIKA
Vol. 13, No
. 4, Decem
b
e
r
2015 : 133
7 – 1342
1340
infrared im
ag
es i
n
o
u
r exp
e
rime
nt, that i
s
, 1
)
Cro
w
, 2
)
Eagle,
3) Ai
rplane
an
d 4
)
Sailboat. On
t
h
e
other
hand,
mis-cla
s
sifica
tion erro
r rat
e
(ME),
ra
te
of numb
e
r of
backg
rou
nd
pixels
(FPR), an
d
ratio of n
u
m
ber
of wron
g
l
y classified
foreg
r
o
und pi
xels
(F
NR) a
r
e
expl
oited as
p
e
rfo
r
ma
n
c
e
evaluation cri
t
eria
Mis
-
c
l
ass
i
fic
a
tion error rate (ME) is
defined as
follows
.
1
TS
T
S
TT
B
BF
F
ME
BF
(11
)
whe
r
e
T
B
and
T
F
denote
ba
ckgroun
d a
n
d
foreg
r
ou
nd
for the g
r
o
u
nd truth i
m
a
ge,
more
over,
S
B
and
S
F
refer to b
a
ckgroun
d an
d foreg
r
ou
nd
in segm
entati
on re
sults. Fu
rtherm
o
re,
lowe
r value o
f
ME metric demon
strate
s
highe
r quality
of segmentat
ion re
sults.
FPR
rep
r
e
s
e
n
ts the
ratio
of numb
e
r of
backg
rou
nd
p
i
xels that
mis-cla
s
sified to
the total
numbe
r of
ba
ckgro
und
pix
e
ls, mo
re
over, FNR de
not
e
s
the
ratio
of
numbe
r of
wrongly
classifi
ed
foreg
r
ou
nd pi
xels to the total numbe
r of foreg
r
o
und pix
e
ls.
TS
T
B
F
FPR
B
(12
)
TS
T
FB
FN
R
F
(13
)
Afterwa
r
ds,
e
x
perime
n
tal result
s for th
e
above th
ree
perfo
rma
n
ce
evaluation
metrics
with different
method
s are provide
d
as f
o
llows.
(a) Cro
w
(b) Eagl
e
(c
) Airpla
ne
(d) Sailb
oat
Figure 2. Experime
n
tal re
sults for differe
nt image type
0
0.1
0.2
0.3
0.4
0.5
0.6
PW
T
D
T
I
B
L
GL
D
N
cut
O
ur
me
t
h
o
d
ME
FP
R
FN
R
0
0.
1
0.
2
0.
3
0.
4
0.
5
0.
6
PW
T
D
T
I
B
L
G
L
D
N
c
u
t
O
u
r
me
t
h
o
d
ME
FP
R
FN
R
0
0.
1
0.
2
0.
3
0.
4
PW
T
D
T
I
B
L
G
L
D
N
c
u
t
O
u
r
me
t
h
o
d
ME
FP
R
FN
R
0
0.
1
0.
2
0.
3
0.
4
0.
5
0.
6
0.
7
0.
8
PW
T
D
T
I
B
L
G
L
D
N
c
u
t
O
u
r
me
t
h
o
d
ME
FPR
FN
R
Evaluation Warning : The document was created with Spire.PDF for Python.
TELKOM
NIKA
ISSN:
1693-6
930
A Novel Im
ag
e Segm
entation Algorithm
Based o
n
Graph Cut Optim
i
zation…
(Z. Guang
-hu
a
)
1341
Integrating all
the above experim
ental re
sult
s, we cal
c
ulate the average pe
rform
a
nce
evaluation re
sults from Fig
u
re 2 to Figu
re 5 as follo
ws.
Table 1. Overall experim
en
tal resu
lt
s for
all the four im
age types
PWT
DTIB
LGLD
Ncut
Our
m
e
thod
ME
0.218
0.374
0.041
0.519
0.004
FPR
0.219
0.366
0.041
0.515
0.003
FNR
0 0 0
0.025
0.040
It can be se
e
n
from the ab
ove experim
e
n
tal
results th
at our propo
sed gra
ph cut based
image
se
gm
entation
algo
rithm i
s
a
b
le
to effect
ively
segm
ent im
a
ges with
hig
h
accu
ra
cy. T
he
rea
s
on
s li
e in
that 1) th
e g
r
aph
mod
e
l i
s
able to
de
scribe
relatio
n
ships
between
different im
a
ge
pixels an
d th
en co
nvert
s
the imag
e se
g
m
ent pr
o
b
lem
to graph
cut
optimizatio
n, and 2
)
graph
cut
optimizatio
n has the a
b
ility to deeply mine intern
al co
rrel
a
tion
s bet
wee
n
variou
s graph n
ode
s.
5. Conclusio
n
In this study, we presente
d
a novel gra
ph-cut-ba
sed
image segm
entation ap
proach by
conve
r
ting th
e imag
e
seg
m
entation
pro
b
lem to
a
gr
a
ph-cut
optimi
z
ation
proble
m
. Inspi
r
ed
b
y
the
facts that an image can be
sepa
rated in
to seve
ral re
gion
s, and image se
gme
n
tation re
sults
are
rega
rd
ed as
several non
-overlap
ping region
s, we convert the image se
gme
n
tation task into a
grap
h cut op
timization p
r
o
b
lem. The m
a
in innovatio
ns of thi
s
stu
d
y are its
de
velopment of
a
binary
vecto
r
of the
seg
m
entation
la
bel, which
can d
e
scribe
both the
foregro
und
an
d
the
backg
rou
nd o
f
an image, a
s
well a
s
the
segm
ent
ation
of an image
via minimizati
on of an ene
rgy
function. Part
icula
r
ly, the energy
functi
on is
solved
by estimati
on
of the importance b
e
twe
e
n
neigh
bor
pixe
ls. To evalu
a
t
e the perfo
rmance
of the
prop
osed al
gorithm, ME,
FPR, and
F
NR
were u
s
e
d
a
s
the
pe
rformance eval
u
a
tion
crite
r
ia.
Fou
r
type
s of ima
ges
were utili
zed
to
con
s
tru
c
t a
d
a
ta set, na
m
e
ly, the crow, eagle,
airpl
ane, an
d sail
boat data
se
ts. Finally, the
experim
ental results sho
w
that
our
prop
ose
d
al
go
rith
m can effe
ctively produ
ce
accurate ima
g
e
segm
entation
result
s.
In the future, we will extend our work in t
he followi
ng asp
e
ct
s: 1) we will att
e
mpt to
utilize the hierarchi
c
al graph cut algorithm in t
he image segmentation task
, 2)
we will introduce
other
optimi
z
ation te
chnol
ogie
s
to
seg
m
ent ima
g
e
s
(e.g., pa
rticl
e
swa
r
m o
p
timization
), a
n
d
3)
we will te
st the perfo
rman
ce of our pr
op
ose
d
method
by using oth
e
r
data sets.
Referen
ces
[1]
Colom
bo MG, D’Ad
da D, Pirelli L
H
.
T
he particip
a
tion of ne
w
tech
nol
og
y-b
a
se
d firms in EU-fund
e
d
R&D partn
ersh
ips: T
he role of venture cap
i
tal
.
Research Policy
. 2016; 45(
2)
: 361-37
5.
[2]
Lush
n
ikov PM,
Vladimir
o
va
N. Modeli
ng o
f
nonl
i
near co
mbini
ng of mu
ltiple l
a
ser b
e
a
ms in Kerr
medi
um.
Optics Express
. 201
5; 23(24): 3
112
0-31
125.
[3]
Z
e
w
e
n
Li
u, D
o
ng
D, C
hun
w
e
n LI. C
h
i
nese
w
o
rd
segm
ent
ation
meth
od
for sh
ort C
h
in
e
s
e te
xt
base
d
on co
nd
ition
a
l
rand
om fie
l
ds.
Journ
a
l
of T
s
in
ghu
a U
n
ivers
i
ty (Scienc
e a
n
d
T
e
chn
o
lo
gy)
. 201
5;
55(
8):
906-
910, 9
15.
[4]
Pike T
W
. Modelli
ng e
ggs
hel
l macul
a
tion.
Avi
an Bio
l
ogy R
e
s
earch.
20
15; 8(
4): 237-2
43.
[5]
Xu
an
S, H
a
n
Y. Improved
e
x
treme
va
lue
w
e
ig
hted
sp
ar
se re
prese
n
tati
ona
l im
ag
e d
e
noisi
ng
w
i
t
h
rand
om pertur
batio
n.
Journ
a
l
of Electronic Ima
g
i
ng.
20
15; 24(6): 06
30
04-
063
00
4.
[6]
Ni T
,
Gu X,
W
ang H. F
u
zz
y c-M
eans
an
d Mea
n
S
h
ift Algorit
hm for
3DPo
int Cl
ou
d
s
Den
o
isi
n
g
.
T
E
LKOMNIKA Indon
esi
an Jou
r
nal of Electric
al Eng
i
ne
eri
n
g
.
2014; 1
2
(7): 5
546-
555
1.
[7]
Basterrech
S,
Rubi
no
G. Ra
n
dom
Neur
al
N
e
t
w
o
r
k M
ode
l f
o
r Su
pervis
e
d
Lear
nin
g
Pr
obl
ems.
Ne
ural
Network World
. 2015; 25(
5): 457.
[8]
Costall
i
S, Ru
gger
i A. Indig
n
a
tion, Ide
o
lo
gi
es
, and Arme
d
Mobil
i
zatio
n
: Civil W
a
r i
n
Ital
y
,
19
43–
45.
International Security.
201
5; 40(2): 11
9-1
5
7
.
[9]
McGlenn
en M
S. "B
y
M
y
Hea
r
t": Gerald Vize
nor'
s
Almost Ashore a
nd Be
a
r
Island: T
he W
a
r at Sugar
Point.
T
r
ans
mo
tion
. 201
5; 1(2)
: 1.
[10]
Kumar S, T
o
shni
w
a
l D. A dat
a mini
ng
frame
w
o
r
k to an
al
yz
e roa
d
acci
den
t data.
Journ
a
l
of Big Data
.
201
5; 2(1): 26.
Evaluation Warning : The document was created with Spire.PDF for Python.
ISSN: 16
93-6
930
TELKOM
NIKA
Vol. 13, No
. 4, Decem
b
e
r
2015 : 133
7 – 1342
1342
[11]
Heg
de V
N
, R
a
o KN. A
n
Outc
ome B
a
sed
C
u
rricul
u
m D
e
si
gn
in E
n
g
i
ne
er
ing-C
a
se
Stud
y A
ppr
oach.
Journ
a
l of Eng
i
neer
ing E
ducat
ion T
r
ansfor
m
a
t
ions
. 201
5; 29
(2): 40-46.
[12]
Pahl
avan
i P, Nahr ST
, Karimi
R. Build
ing
det
ection us
in
g ae
rial ima
ges a
n
d
LiDA
R data
via ad
aptiv
e
neur
o-fuzz
y
s
ystems.
Journal
of Geomatics Scienc
e an
d T
e
chn
o
lo
gy
. 201
5; 5(1): 109-1
2
5.
[13]
Okazaki T
,
Aibara
U, Seti
ya
ni L. A S
i
mul
a
ti
on Stu
d
y
on
T
he Behavi
o
r
Anal
ys
is of T
he De
gree
of
Members
h
ip i
n
F
u
zz
y
c-mean
s Method.
IEIE T
r
ansactions
on Smart Processin
g
& Computin
g
. 201
5;
4(4): 209-
21
5.
[14]
Sing
h S, T
u
li R, Sharma V.
F
u
zz
y
bi-criteri
a
pro
b
lem
of c
l
usteri
ng rati
on
shops t
o
w
a
r
e
hous
e sites
.
Internatio
na
l Journ
a
l of Know
led
ge a
nd
Res
earch i
n
Mana
ge
me
nt and E-
Co
mmerce
. 2
0
15; 5(3): 1-7.
[15]
Venkates
h
R. Improvin
g
Data
Accurac
y
Usin
g
Pro
a
ctive Co
rrelate
d
Fuzz
y S
y
stem in
Wire
less
S
ens
o
r
Net
w
orks.
KSII T
r
ansactions
on Internet a
n
d
Informati
on Sy
stems.
20
15; 9
(
9): 3515-
35
38
.
[16]
Josep
h
B, R
a
mach
andr
an
B, Muthukri
shna
n
P. Intelli
ge
nt Detec
t
ion a
nd C
l
a
ssificatio
n
Of
Microcalc
i
ficati
on In
C
o
mpre
ssed M
a
mmog
r
am Imag
e.
Image
An
alysis
& Stereo
lo
gy.
201
5; 3
4
(3)
:
183-
198.
[17]
Harikir
an J,
Lakshmi PV, K
u
mar RK. Multip
le F
e
ature
F
u
zz
y
c-m
ean
s Cluster
ing
A
l
gorit
hm for
Segme
n
tatio
n
of Microarr
a
y Images.
Inter
n
a
t
iona
l Jour
na
l
of Electrica
l
a
nd C
o
mput
er
Engi
neer
in
g
,
201
5; 5(5): 104
5-10
53.
[18]
Srisae
ng P, Ba
xter GS, W
ild
G. An ada
ptive
neur
o-fuzz
y i
n
ference s
y
ste
m
for forecasti
ng Austra
lia'
s
domestic l
o
w
c
o
st carrier pas
seng
er dem
an
d.
Aviation
. 2
0
15; 19(3): 1
50-
163.
[19]
Prabh
avath
y
P
,
T
r
ipath
y
BK.
An integr
ated
coveri
ng-b
a
se
d roug
h fuzz
y
set clusterin
g
appr
oach fo
r
sequ
enti
a
l data
.
Internation
a
l
Journ
a
l of Re
a
s
oni
ng-b
a
se
d Intelli
ge
nt Systems
. 20
15; 7(3-
4): 296-3
04.
Evaluation Warning : The document was created with Spire.PDF for Python.