Indonesi
an
Journa
l
of El
ect
ri
cal Engineer
ing
an
d
Comp
ut
er
Scie
nce
Vo
l.
12
,
No.
3
,
Decem
ber
201
8
, p
p.
1
1
06
~
1
1
10
IS
S
N: 25
02
-
4752, DO
I: 10
.11
591/ijeecs
.v1
2
.i
3
.pp
1
1
06
-
1
1
10
1106
Journ
al h
om
e
page
:
http:
//
ia
es
core.c
om/j
ourn
als/i
ndex.
ph
p/ij
eecs
The
Eff
ects
of Se
gment
ation Te
ch
niq
ues in Digital
Image B
ased
Identific
atio
n
of
Ethiopi
an
Paper Cu
rrency
So
lo
mo
n W
onda
ya Gu
angul
Depa
rtment
o
f
S
ta
ti
st
ic
s
,
Coll
ege of
Sci
ence
,
Bah
ir
Dar
Univer
sit
y
Art
ic
le
In
f
o
ABSTR
A
CT
Art
ic
le
history:
Re
cei
ved
Dec
1
3
, 201
7
Re
vised
Ma
r 3
,
2018
Accepte
d
Aug
2
1
, 201
8
Paper
and
coi
n
a
re
the
two
m
ost
comm
on
cur
ren
ci
es
in
al
l
over
th
e
world.
In
Et
hiopia
al
so
p
ape
r
and
co
in
c
urre
nc
y
ar
e
use
d
for
m
edi
um
o
f
exc
hang
e.
Thi
s
pap
er
pr
e
sents
the
compara
t
ive
stud
y
o
f
segm
ent
ation
technique
s
towar
ds
Et
hiopian
pape
r
cur
r
en
c
y
cl
assi
fi
ca
t
ion
.
Otsu,
FC
M
an
d
K
-
m
ea
ns
segm
ent
at
ion
te
c
hnique
s
are
conside
red
for
thi
s
st
ud
y
and
BP
NN
i
s
used
for
cl
assifi
ca
t
ion
of
cur
ren
c
ie
s.
For
t
he
class
ifi
cation
,
images
are
col
l
ec
t
ed
from
comm
erc
ia
l
ban
k
of
Et
hiopi
a
an
d
Dashen
Bank;
for
our
dat
a
set
,
a
tot
a
l
of
500
images
sampl
es
were
col
l
ecte
d.
From
the
se
images,
91.
2%
ac
cur
acy
is
ac
hi
eve
d
when
Otsu
segm
ent
at
i
on
is
used
on
BP
NN
with
TANH
le
arn
ing
func
ti
on
.
Ke
yw
or
ds:
BPNN
FCM
K
-
m
eans
Otsu
Copyright
©
201
8
Instit
ut
e
o
f Ad
vanc
ed
Engi
n
ee
r
ing
and
S
cienc
e
.
Al
l
rights re
serv
ed.
Corres
pond
in
g
Aut
h
or
:
So
lom
on
Won
daya G
ua
ngul
,
Dep
a
rtm
ent o
f St
at
ist
ic
s
,
Coll
ege
of
Sci
ence
,
Ba
hir Da
r
U
ni
ver
sit
y,
Em
a
il
:
so
lom
o
nw@
bdu.
e
du.e
t
1.
INTROD
U
CTION
Currency
is
a
m
edium
of
ex
change
t
hat
is
us
e
d
in
e
ver
y
act
ivit
y
of
hu
m
an
li
fe.
Pape
r
an
d
c
oin
a
re
the
two
m
os
t
com
m
on
cur
re
nc
ie
s
in
al
l
ov
e
r
the
w
or
l
d.
I
n
Ethio
pia
al
so
pa
per
a
nd
coi
n
currency
is
us
e
d
f
or
m
edium
of
ex
change
,
as
s
ho
wn
in
Fig
ur
e
1
.
C
urre
ntly
Ethiop
ia
n
c
urre
ncy
inclu
des
c
oin
s
su
c
h
a
s
one
ce
nt,
five
ce
nt,
te
n
c
ent,
twe
nty
-
fi
ve
cent,
fi
fty
cent
an
d
one
Bi
r
r
cent
a
nd
pa
pe
r
cu
rr
e
ncy
on
e
Bi
rr
,
fi
ve
Bi
r
r,
te
n
Bi
rr
,
fifty
Bi
rr
and
hundre
d
B
irr
[
1].
D
ue
to
highly
so
phist
ic
at
ed
de
vices,
currency
no
te
s
counter
fe
it
ing
is
the
m
ajo
r
pr
ob
le
m
ar
ound
t
he
w
or
l
d
a
nd
it
is
ver
y
dif
ficult
t
o
i
den
ti
fy
t
he
forg
e
d
note
s
from
the
act
ual
no
te
s
.
W
it
h
the
help
of
c
om
pu
te
r
vi
sion,
it
is
bette
r
to
ide
ntify
forg
e
d
note
s
f
r
om
act
ual
cur
re
ncy
note
s
as
c
om
par
ed
to hum
an
ey
es
[2
]
.
(a)
(b)
Figure
1.
Act
ua
l Et
hiopian
pa
per cu
rr
e
ncy (a
)
&
Fa
ke
Ethi
opia
n pa
per cu
rrency
(b)
Evaluation Warning : The document was created with Spire.PDF for Python.
Ind
on
esi
a
n
J
E
le
c Eng &
Co
m
p
Sci
IS
S
N:
25
02
-
4752
Th
e
Eff
ect
s o
f
Segme
nta
ti
on
Tech
niques i
n Digit
al
Ima
ge Base
d…
(
Solo
mon
W
onday
a Gua
ngul
)
1107
Currentl
y,
co
m
pu
te
r
vision
is
gr
ad
ually
fin
ding
ap
plica
ti
on
s
in
dif
fe
ren
t
pro
blem
do
m
ai
ns
[3
]
.
A
te
ch
nolo
gy
of
C
om
pu
te
r
visio
n
ap
plica
ti
on
s
re
qu
ire
s
diff
e
re
nt
im
ag
e
pr
e
-
processi
ng
te
ch
niques
a
m
ong
these;
segm
entat
ion
te
c
hn
i
que
is
the
bac
k
bone
in
im
age
proces
sin
g
fiel
d.
Se
gm
entat
ion
te
chn
i
que
is
use
d
t
o
extract
the
re
pr
ese
ntati
ve
re
gions
of
the
im
age.
T
o
ide
ntify
these
fe
at
ur
es
t
he
give
n
im
age
has
to
be
par
ti
ti
on
e
d
t
h
a
t
are
vis
ually
dif
fer
e
nt
[
4].
The
re
is
no
par
ti
cula
r
im
a
ge
se
gm
entat
i
on
te
ch
nique
that
is
appr
opriat
e to
al
l im
age p
r
oc
essing
a
reas [5
]
. Th
ere
fore,
th
is researc
h
pa
pe
r
f
ocu
se
d on
t
he
com
par
at
ive
stud
y
of
segm
entat
ion
te
ch
niques
towa
r
ds
Et
hi
op
ia
n
c
urrenc
y
recog
niti
on
syst
em
.
Th
e
f
ollow
i
ng
im
age
segm
entat
ion
techn
i
qu
e
s ar
e
pr
ese
nted
in
t
hi
s p
a
per
:
1
.
1.
O
tsu
Im
ag
e
Se
gmen
tation
Otsu
Im
age
Segm
entat
ion
the
m
os
t
co
m
m
on
ly
us
ed
te
chn
i
qu
e
to
par
ti
ti
on
the
i
m
age
as
fo
re
gro
un
d
and
bac
kgr
ound.
Otsu
'
s
m
et
h
od
in
vo
l
ves
it
erati
ng
t
hro
ugh
al
l
the
possible
thres
hold
valu
es
an
d
cal
culat
ing
a
m
easur
e
of
s
pread
for
t
he
pixe
l
l
evels
eac
h
s
ide
of
t
he
t
hr
es
ho
l
d,
i.e
.
t
he
pix
el
s
that
ei
t
her
fall
s
in
f
or
e
gro
un
d
or
bac
kgr
ound.
The
aim
is
to
fin
d
the
thres
hold
val
ue
w
here
the
su
m
of
fo
re
gro
und
an
d
backgroun
d
sp
rea
ds
is at
it
s
m
ini
m
um
[
6].
Algorithm
steps:
1)
Com
pu
te
h
ist
ogram
an
d p
rob
abi
li
ti
es o
f
eac
h
inte
ns
it
y l
eve
l.
2)
Set u
p
init
ia
l cl
ass proba
bili
ty
an
d i
niti
al
class m
eans.
3)
Step th
r
ough al
l possi
ble th
res
ho
l
ds
m
axi
m
um
intensit
y.
4)
Update
qi and
μi.
5)
Com
pu
te
betw
een class
va
riance.
6)
Desire
d
th
res
hold c
orres
pond
s to
t
he
m
axi
m
um
v
al
ue
of
be
twee
n
cl
ass
v
a
r
ia
nce.
1
.
2
.
F
C
M
im
ag
e
se
gmen
tat
ion
In
FCM
,
it
is
possible
f
or
a
data
sam
p
le
to
belo
ng
to
m
ult
iple
clu
ste
rs
at
the
sam
e
tim
e
.
The
sim
il
arity
is
ind
ic
at
ed
by
the
m
e
m
ber
shi
p
val
ue.
I
n
F
CM
a
data
sa
m
ple
is
assigne
d
with
a
m
e
m
ber
s
hip
value
based
on
it
s
si
m
il
arity
with
the
cl
us
te
r
center
.
The
m
e
m
ber
sh
ip
va
lues
are
betw
een
0
to
1
a
nd
m
or
e
the
si
m
il
arity,
higher
the
m
e
m
ber
sh
i
p
value.
Defuzzi
ficat
io
n
is
a
ppli
ed
at
the
e
nd
of
t
he
cl
us
te
rin
g
pr
ocess
t
o
decide
the
cl
ust
ering.
FCM
is
a
rep
et
it
ive
alg
ori
th
m
and
the
so
luti
on
is
achieve
d
by
re
pe
ti
ti
vely
up
da
ti
ng
the
cl
us
te
r
ce
nter
a
nd m
e
m
ber
sh
ip
value [
7].
1
.
3
.
K
-
me
an
s
ima
ge
se
gme
nt
ati
on
K
-
Me
a
ns
is
le
ast
-
square
par
t
it
ion
ing
m
et
ho
ds
that
div
ide
a
colle
ct
ion
of
obj
ect
s
into
K
gr
ou
ps
.
T
he
al
gorithm
it
era
te
s o
ver
t
wo st
eps:
1)
Com
pu
te
the
m
ean of
eac
h
c
luster.
2)
Com
pu
te
the
di
sta
nce
of
eac
h
po
int
f
r
om
eac
h
cl
us
te
r
by
co
m
pu
ti
ng
it
s
distance
from
the
corres
pondin
g
cl
us
te
r
m
ean. Assig
n
eac
h p
oi
nt to
the
cluste
r
it
is n
ea
rest t
o.
Iterate
over
t
he
above
tw
o
st
eps
ti
ll
th
e
sum
of
square
d
within
group
e
rror
s
ca
nnot
be
lowered
a
ny
m
or
e.
The
init
ia
l
assignm
ent
of
po
i
nts
to
cl
us
te
rs
ca
n
be
done
rand
om
l
y.
In
the
it
erati
on
s,
the
al
gorith
m
trie
s
to m
ini
m
iz
e
the su
m
, o
ver
all
g
r
oups
, of
t
he
sq
ua
re
d
withi
n group errors,
wh
ic
h
are t
he dist
ances
of
th
e p
oi
nts
to
the
re
sp
ect
iv
e
gro
up
m
eans.
Conver
ge
nce
i
s
reache
d
w
he
n
the
obj
ect
ive
functi
on
(i.e
.,
the
resid
ual
s
um
-
of
-
sq
ua
res
)
ca
nnot
b
e lo
we
red an
y
m
or
e [
8].
Yao
Y
u
&
Xueso
ng
Suo
co
nducte
d
a
stu
dy
to
detect
sm
oo
th
ness
of
bo
t
tl
e
cap.
In
t
his
pa
per
the
auth
or
hav
e
use
d
MATL
AB
and
C
la
ng
uage
as
a
too
l
for
i
m
age
segm
e
ntati
on
,
e
nhan
ce
m
ent,
filt
ering
a
nd
oth
e
r processi
ng
of sm
oo
thn
e
s
s of ca
p [9
]
.
On
researc
h
[
10
]
,
the
a
utho
rs
pro
posed
le
tt
uce
i
m
age
segm
entat
ion
.
In
this
pa
pe
r
the
auth
or
s
des
cri
bed
the
tradit
ion
al
2
-
D
m
axi
m
u
m
entr
op
y
al
gori
thm
has
so
m
e
fau
lt
s,
su
c
h
as
low
accu
ra
cy
of
segm
entat
ion
,
slow
sp
e
ed
,
a
nd
poor
a
nti
-
noi
se
abili
ty
.
All
t
he
st
ud
ie
s
s
howed
that
Im
age
segm
entat
ion
is
the
basic
pa
rts
in
com
pu
te
r
visi
on
processin
g.
So
t
his
re
sea
rc
h
pap
e
r
f
oc
us
e
d
on
t
he
perform
ance
analy
sis
of
segm
entat
ion
techn
i
qu
e
s to
w
ard
s
to
cl
assi
ficat
ion
of Ethi
opia
n pa
per cu
rrency
.
2.
RESEA
R
CH MET
HO
D
In
this
st
ud
y,
scan
ner
HP
sc
an
j
et
pro
4500
a
nd
cam
era
canon
EO
S
600d
are
us
e
d
to
captu
re
t
he
i
m
age.
To
m
ini
m
iz
e
no
ise
s
li
ke
li
ghti
ng
a
nd
blurre
d
im
age
effe
ct
s
w
hile
capt
ur
in
g
t
he
i
m
ages
scan
ne
rs
a
re
eff
ect
ive
.
T
herefo
re,
scan
ne
r
is
us
e
d
i
n
this
stud
y.
I
n
orde
r
to
ha
ve
a
go
od
dataset
f
orm
al
l
pe
rsp
ect
iv
e
ne
w,
old
a
nd
a
per
t
ure
pa
per
cu
rr
e
nc
ie
s
for
both
a
ct
ual
an
d
fak
e
pap
e
r
c
urre
ncies
are
c
onside
r
ed.
A
t
otal
of
5
pa
per
currency
no
te
s
ty
pe
each
ha
ving
10
0
are
consi
der
e
d
f
or
this
stud
y.
O
nce
the
im
ages
are
col
le
ct
ed
,
pr
e
-
processi
ng step
s ar
e
pe
rfor
m
ed
to
achie
ve
t
he
goal o
f
t
he
st
ud
y t
hro
ugh
M
ATL
AB, 2
014.
Evaluation Warning : The document was created with Spire.PDF for Python.
IS
S
N
:
250
2
-
4752
Ind
on
esi
a
n
J
E
le
c Eng &
Co
m
p
Sci,
Vo
l.
12
, N
o.
3
,
Dece
m
ber
2
01
8
:
1
1
0
6
–
1
1
10
1108
3.
IMAGE P
ROCESSI
NG S
Y
STE
M
Pape
r
curre
nc
ie
s
are
scanned
with
a
sui
ta
ble
reso
luti
on
a
nd
are
s
tore
d
as
a
bin
ary
im
age.
Be
fore
bei
ng
analy
zed
i
m
age
unde
r
goes
s
om
e
pre
-
proce
ssin
g,
includi
ng
re
ha
bili
ta
ti
on
,
s
m
oo
thing
and
norm
al
iz
a
ti
on
.
T
he
pr
e
-
processi
ng
is
carried
out
in
order
to
im
pr
ove
the
qual
it
y
of
im
age
t
o
be
processe
d
[
11
]
.
Im
age
pr
oce
ssing
a
nd
patte
rn
rec
ogniti
on
pe
rfor
m
ed
by
analy
zi
ng
th
e
i
m
age
of
Ethio
pian
pap
e
r
cu
rr
e
ncy
[12].
The
ph
a
se
of
act
ivit
ie
s
fo
r
i
den
ti
ficat
ion
of
pa
pe
r
currency
c
on
sis
ts
of
the
f
ollo
wing
com
po
ne
nts:
The
im
age
is
a
cqu
i
red
from
t
he
scan
ne
r
w
hich
is
us
e
d
as
in
pu
t
im
age.
The
i
m
age
is
us
ually
stored
i
n
j
pg,
ti
ff
or
pn
g
file
fo
rm
at
.
In
t
his
research
pap
e
r
80
X
80,
360
X
36
0
an
d
512
X51
2
i
m
a
ges
are
te
ste
d.
Eve
n
though,
pre
-
pr
ocessin
g
is
slo
w,
w
hen
the
siz
e
of
im
age
in
creases,
segm
entat
ion
bec
ome
eff
ect
i
ve.
For
that
reason,
the im
ages a
re
resized
to 51
2 X51
2 [13].
Pr
e
-
proces
sin
g
are
th
os
e
o
pe
rati
on
s
that
a
r
e
norm
al
ly
req
uire
d
pri
or
to
the
m
ai
n
data
a
naly
sis
an
d
extracti
on
of
inf
or
m
at
ion
.
T
he
aim
of
i
m
a
ge
pre
-
process
ing
is
to
sup
press
undesire
d
distor
ti
ons
or
enh
a
nce
so
m
e i
m
age f
eat
ur
es t
hat are
im
po
rtant for
fu
rther p
r
ocessin
g or analy
sis [
1
4].
Im
age
segm
entat
ion
is
a
n
im
portant
c
om
ponen
t
of
im
age
processi
ng
te
chn
i
qu
e
that
det
erm
ines
the
accuracy
of
th
e
syst
e
m
.
I
m
age
segm
entat
ion
is
def
ine
d
as
the
pa
rtit
ion
in
g
of
an
im
age
into
none
ov
e
rla
pp
i
ng,
const
it
uen
t
re
gi
on
s
t
hat
are
hom
og
en
ous
wi
th
re
s
pect
to
s
om
e
char
act
erist
ic
su
ch
a
s
inte
ns
it
y
or
te
xtu
re
[15].
FCM
, K
-
Me
an
s and
Otsu se
gm
entat
ion
(
Fig
ur
e
2)
te
ch
niqu
es are c
onside
r
ed
in
this
pa
per.
Figure
2
.
Ots
u Segm
entat
ion
4.
RESU
LT
S
The
w
ord
network
i
n
the
te
r
m
'
arti
fici
al
ne
ur
al
net
wor
k'
ref
er
s
to
t
he
inter
–
c
onnecti
ons
betwee
n
th
e
neur
on
s
in
the
diff
e
re
nt
la
ye
rs
of
each
syst
em
.
A
syst
e
m
has
three
la
ye
rs
.
The
first
la
ye
r
has
i
nput
ne
ur
on
s
,
wh
ic
h
se
nd
data
via
sy
nap
se
s
to
the
seco
nd
l
ay
er
of
neur
ons,
a
nd
the
n
via
m
or
e
synapses
to
the
thir
d
la
ye
r
of
ou
t
pu
t
ne
uro
ns.
More
com
plex
syst
e
m
s
will
hav
e
m
or
e
la
ye
rs
of
neur
on
s
with
so
m
e
havi
ng
inc
reased
l
ay
ers
of
in
put
ne
uro
ns
an
d
ou
t
pu
t neur
on
s
.
The
s
ynapses
st
or
e pa
ram
et
ers
cal
led
"wei
gh
ts"
th
at
m
anipu
la
te
the
data
in the calc
ulati
on
s
[
16]
.
The
m
os
t
popula
r
neural
net
w
ork
m
od
el
is
t
he
r
nu
lt
il
ay
er
per
ce
ptr
on
(M
LP),
w
hich
is
an
exte
ns
io
n
of
t
he
si
ng
le
l
ay
er
pe
rcep
t
ron
propose
d
by
Rose
nb
la
tt
.
Mult
il
ay
er
perce
ptr
on
s
,
in
ge
ner
al
,
are
fee
dfo
rw
a
r
d
netw
ork, ha
ving
disti
nct in
pu
t
, output, a
nd
hid
de
n
la
ye
rs
.
In
t
his
pap
e
r,
MLP
ne
ural
ne
twork
with
ba
ck
pro
pag
at
io
n
is
us
e
d
f
or
c
la
ssifyi
ng
Et
hio
pia
n
pape
r
currencies
t
o
t
heir
c
orres
pondin
g
cl
ass.
I
n
this
ex
per
im
ent,
the
ne
ural
ne
twork
is
te
ste
d
by
SIGMO
I
D
a
nd
TANH
act
ivat
ion
f
unct
ion
s
.
The
perform
ances
of
the
cl
assi
fier
wer
e
te
ste
d
by
AN
N
(
A
rtific
ia
l
Neu
ral
Netw
ork) u
si
ng th
ree
diff
e
re
nt tech
niques
of se
gm
entat
ion
. In order
to
tra
in the classi
fie
r
s,
70%
we
re
use
d
f
or
m
od
el
trai
nin
g
and
30%
wer
e
us
e
d
f
or
pe
rfo
rm
ance
te
sti
ng
.
As
s
how
n
in
Fi
gure
3
a
nd
4
,
the
res
ult
in
d
ic
at
ed
that
ther
e
was
91.
2%
achie
ve
d
us
in
g
BPN
N
with
ots
u
s
egm
entat
ion
.
The
ai
m
of
th
e
resea
rch
paper
is
t
o
identify
the
effe
ct
s
of
segm
entat
ion
te
chn
i
ques
in
cl
assifi
cat
ion
of
Ethio
pian
pa
per
c
urre
ncies.
I
n
this
pa
per,
BPNN
are
use
d
a
nd
the
accu
racy
of
t
he
sys
tem
is
pr
ese
nted,
a
nd
the
re
sul
ts
of
B
PNN
w
it
h
TA
N
H
act
ivati
on
functi
on
wer
e
discusse
d an
d hope
fu
l
res
ults we
re
ob
ta
ine
d.
Evaluation Warning : The document was created with Spire.PDF for Python.
Ind
on
esi
a
n
J
E
le
c Eng &
Co
m
p
Sci
IS
S
N:
25
02
-
4752
Th
e
Eff
ect
s o
f
Segme
nta
ti
on
Tech
niques i
n Digit
al
Ima
ge Base
d…
(
Solo
mon
W
onday
a Gua
ngul
)
1109
Figure
3.
Re
su
l
ts of se
gm
entation
tec
hn
i
ques
Figure
4.
Proto
ty
pe
ACKN
OWLE
DGE
MENTS
We
gr
eat
ly
ack
nowled
ge
Ba
hi
r
D
ar
U
niv
e
rsity
.
REFERE
NCE
S
[1]
Jegna
w Fent
ahu
n,
“
Autom
at
ic recognit
ion
of Eth
i
opia
n
p
ape
r
cur
r
ency
”
,
The
sis,
Addis Ababa
Un
iv
ersity
,
2014.
[2]
Ze
wde
Dinku
a
nd
Kum
udha
Rai
m
ond,
”
Counte
r
fei
t
Curr
ency
Id
ent
ifica
ti
on
S
y
s
t
em
-
A
Case
Stud
y
on
Et
hiop
ian
Birr
Note
”,
Addi
s Ababa
Univ
ersity
,
2009.
[3]
W
il
li
am K.
Pratt:
Digital
image
p
roc
essing,
PIK
S
Scie
nti
f
ic i
nside
,
John W
il
e
y
,
4th
Edition, 2007.
[4]
Arti
T
ane
j
a, Pri
ya
Ran
ja
n
,
Am
it
Ujjl
a
y
a
n
“
A Perf
orm
anc
e
Stud
y
o
f
Im
age
Segm
en
ta
ti
on
T
ec
hniqu
e
s”,
IE
EE,
2015
.
[5]
Abrham
Deba
su
Men
gistu,
Seff
i
Gebe
y
e
hu
Me
ngistu
,
Dagna
c
hew
Mele
sew
Alema
y
ehu,
“
Im
age
Anal
y
sis
for
Et
hiopian
Coff
e
e
Plan
t
Dise
ase
s
Ide
nti
f
ic
a
ti
on”
,
I
JBB
,
2016.
[6]
Ta
Yang
Goh,
Shafriz
a
NishaB
asa
h,
Hani
za
Ya
zi
d,
Muham
m
adJuha
iri
Azi
z
Safa
r,
Fathi
nu
l
S
y
a
hirAhm
ad
Saad
,
“
Perform
anc
e
an
aly
s
is of
image t
hre
sholding: Otsu t
e
chni
que
”
Sc
i
enc
eDir
ect, 201
7.
[7]
Mahipa
l
Singh
Choudhr
y
,
Rajiv
Kapoor,
“
Perfo
rm
anc
e
Anal
y
sis
of
Fuzz
y
C
-
Me
ans
Cluste
ring
Methods
for
MRI
Im
age
Segm
ent
a
ti
on”
,
Sci
enceDi
rec
t
,
2016
.
[8]
Nam
ei
rak
pamD
hana
ch
andr
a
,
Khum
ant
hemM
angl
em,
Yam
bem
JinaCha
nu,
“
Im
age
Segm
ent
a
ti
o
n
Us
ing
K
-
m
ea
ns
Cluste
ring
Algor
it
hm
and
Subtr
a
ct
iv
e
Clust
eri
ng
Algorit
hm
”,
Sc
i
enc
eDir
ect, 201
6.
[9]
Yao
Yu,
Xuesong
Suo,
“
Det
ec
t
ion
S
y
stem
of
Bott
le
Ca
p
Sm
oothne
ss
Based
on
Im
age
Proce
ss
ing”
,
TE
LKOM
NIK
A
,
T
elec
om
m
unic
at
ion
Com
puti
ng
Elec
tron
ic
s
and Control
,
2016.
[10]
Jun
Sun
Jun
Su
n,
Yan
W
ang
Yan
W
ang,
Xiaoh
ong
W
u
Xiaohong
W
u,
Xiaodong
Zha
ng,
Hong
yan
Gao,
“
A
New
Im
age
Segm
entati
on
Algo
rit
h
m
and
Its
Ap
pli
c
at
ion
in
L
et
tu
ce
Obj
ec
t
Segm
ent
at
ion
”
,
TE
LKOM
NIK
A,
Te
l
ec
om
m
unic
ation
Com
puti
ng
El
e
ct
roni
cs
and
Control
,
2016.
86
87
88
89
90
91
92
FCM
K-m
eans
Ot
su
Se
r
i
es1
Evaluation Warning : The document was created with Spire.PDF for Python.
IS
S
N
:
250
2
-
4752
Ind
on
esi
a
n
J
E
le
c Eng &
Co
m
p
Sci,
Vo
l.
12
, N
o.
3
,
Dece
m
ber
2
01
8
:
1
1
0
6
–
1
1
10
1110
[11]
Li
li
k
Sum
ar
y
an
ti ,
Aina
Mus
dholifah
,
Sri Ha
rt
ati,
“
Digit
al
Im
age
Based
Ide
nt
ifi
c
ation
of
Rice
Vari
ety
Us
ing
Im
age
Proce
ss
ing
and
Neura
l
N
et
work
”,
TELKO
MN
I
KA
Indone
sian
J
ourna
l
of
E
le
c
trica
l
Engi
n
ee
r
ing,
2015.
[12]
Kom
al
Vora,
Am
i
Shah,
Ja
y
Mehta,
“
A
Revi
ew
Paper
on
Curre
nc
y
Rec
ogn
it
ion
S
y
stem”,
Inte
rn
at
ion
al
Journal
o
f
Com
pute
r
Appli
ca
t
ions,
2015
.
[13]
Megha
Th
akur
,
Am
rit
Kaur,
“
Vari
ous
fak
e
cur
r
e
nc
y
detec
ti
on
t
e
chni
ques”
,
Int
er
nat
ion
al
Journa
l
for
Technol
ogi
c
al
Resea
rc
h
in
Eng
ine
er
ing, 2014.
[14]
Ehsani
AR,
Sh
afr
y
MR,
Noro
uzi
A.
“
Digital
Denta
l
X
-
Ra
y
Im
age
Segm
ent
at
ion
and
Fe
a
ture
Ex
tracti
on
”
,
Indone
sian
Jour
nal
of
E
le
c
tri
c
al
Engi
ne
eri
ng
and
Com
pute
r
Sci
en
ce
,
2013.
[15]
Muham
m
adSar
f
raz
,
”
An
Intelli
g
ent
Pap
er
Curr
en
c
y
Re
cogni
t
ion
S
y
stem”,
Proce
d
ia
Com
pute
r
Sci
enc
e
,
2015
[16]
Subhadip
Basu
,
Nibar
an
Das,
Ram
Sarka
r,
”
An
MLP
base
d
App
roa
ch
for
Rec
og
nit
ion
of
Handwrit
te
n
‘B
angla
’
Num
era
ls”,
In
te
r
nat
ion
al
Conf
erence
on
Artif
ic
i
al
Inte
lligen
ce,
200
5.
Evaluation Warning : The document was created with Spire.PDF for Python.