Internati
o
nal
Journal of Ele
c
trical
and Computer
Engineering
(IJE
CE)
V
o
l.
6, N
o
. 4
,
A
ugu
st
2016
, pp
. 16
17
~
1
626
I
S
SN
: 208
8-8
7
0
8
,
D
O
I
:
10.115
91
/ij
ece.v6
i
4.9
688
1
617
Jo
urn
a
l
h
o
me
pa
ge
: h
ttp
://iaesjo
u
r
na
l.com/
o
n
lin
e/ind
e
x.ph
p
/
IJECE
Pre-processing Technique for Wireless Capsule Endoscopy
Image Enhancement
Ros
d
ian
a
Sh
ahril,
Sabari
ah
Bah
a
run
,
AK
M Muz
a
hidul Islam
Malay
s
i
a
-Japan
International In
s
titut
e
of
T
echnol
og
y
(MJIIT), Un
iversiti
T
e
knolo
g
i Mal
a
y
s
ia (UT
M
)
Kuala
Lumpur C
a
mpus, Jalan
Sultan Yah
y
a Petra, 54100, Kuala Lumpur, Malay
s
ia
Article Info
A
B
STRAC
T
Article histo
r
y:
Received Dec 11, 2015
Rev
i
sed
Jun
3
,
2
016
Accepted
Jun 18, 2016
Wireless capsule endoscop
y
(WCE) is us
ed to
examine human dig
e
stive
tract
in order to detect abnormal areas
. Howeve
r, it has been a challen
g
ing task to
detect an abnor
mal area such as
bleedi
ng due to
poor quality
and
dark images
of WCE. In this paper, pre-pr
ocessing techn
i
que is introduced to ease
classification of
the bleeding
ar
ea. Anis
otrop
i
c
contrast d
i
ffusio
n
method is
emplo
y
ed in our pre-processing techn
i
que
as a contrast enhan
c
ement of the
images. Th
ere is a drawb
ack
to
the method pr
oposed b
y
B
.
Li ear
lier
,
in
which th
e qua
lit
y of
W
C
E im
ag
e is
d
e
graded
w
h
en th
e num
ber
of it
erat
ion
increases. To solve this problem, va
riance is emplo
y
ed in our proposed
method. To fur
t
her enh
a
nce WCE image,
Discrete Cosine
Transf
orm is used
with an
isotropic contr
a
st diffusion.
Experimen
t
al r
e
sults show that both
proposed contr
a
st enhan
cement algor
ithm and
sharpening
WCE imag
e
algorithm
provid
e
bet
t
er p
e
rform
ance
com
p
ared
with B.
Li’s
a
l
g
o
rithm
s
i
nc
e
S
D
M
E
and EBCM
values
are
s
t
able
whenev
er the number
of iterations
incre
a
s
e
. M
o
reo
v
er, th
e s
h
arpne
s
s
m
eas
urem
ent
s
us
ing gradient
and P
S
N
R
are bo
th
improved b
y
31
.5%
and
20.3%
, r
e
spectively
.
Keyword:
Ani
s
ot
r
opi
c di
f
f
usi
o
n
Co
lo
r con
t
rast
Discrete c
o
sine
trans
f
orm
Enhancem
ent image
Hessian m
a
trix
W
i
r
e
less cap
s
ule en
do
scop
y
Copyright ©
201
6 Institut
e
o
f
Ad
vanced
Engin
eer
ing and S
c
i
e
nce.
All rights re
se
rve
d
.
Co
rresp
ond
i
ng
Autho
r
:
Ro
sd
ian
a
Sh
ahril,
Malaysia-Jap
an
In tern
ation
a
l
In
stitu
te of Tech
no
log
y
(MJIIT),
Un
i
v
ersiti Tekn
o
l
o
g
i
Malaysia (UTM),
5
410
0 Ku
ala Lu
m
p
u
r
.
Em
a
il: ro
sd
ianash
ah
ril@g
m
ai
l.co
m
1.
INTRODUCTION
W
i
reless capsu
le en
do
scop
y (W
CE) is a p
ill-sh
ap
ed
d
e
v
i
ce, wh
ich
cap
t
ures th
e imag
es wh
ile
m
o
v
i
n
g
aro
und
th
e d
i
g
e
stiv
e tract to
d
e
tect fo
r an
y
ab
norm
a
l
ities. W
C
E co
n
s
ists
o
f
a lig
h
t
so
urce, ca
m
e
ra,
rad
i
o
tran
sm
itt
er an
d
b
a
ttery. Th
e p
a
tien
t
swallo
ws th
e cap
su
le lik
e a p
ill, an
d
th
en
it cap
tures th
e i
m
ag
es in
in
tern
al organ
an
d
sen
t
ou
t wirelessly
to a special recorder whic
h is att
ach
ed
to
th
e
patien
t
’s waist. Th
is
p
r
o
cess con
tinu
e
s un
til th
e en
d
o
f
b
a
ttery
’s life. Fin
a
lly,
a
ll i
m
ag
es in
th
e reco
rd
er are d
o
wn
lo
ad
ed
to in
sp
ect
an
d
co
m
p
u
t
e by a p
h
y
sician
.
WCE im
ag
es h
a
v
e
v
e
ry po
or qu
ality d
u
e
of low transm
is
sio
n
po
wer of
WCE
devi
ce a
n
d
ba
n
d
wi
dt
h c
o
nst
r
a
i
nt
s.
C
o
m
put
er ai
ded det
ect
i
on
(C
AD
) sy
st
em
aim
s
t
o
reduce t
h
e b
u
r
d
en a
nd
ove
rsi
g
ht
s o
f
o
b
ser
v
at
i
o
ns
.
The t
e
rm
of “
c
om
put
er ai
de
d det
ect
i
o
n sy
st
em
” i
s
defi
n
e
d as a sy
st
em
or so
ft
wa
re
t
h
at
i
s
assi
gn
ed t
o
recognize sus
p
icious c
h
aracte
r
istics of t
h
e
medical im
age
s
before a
phy
sician ins
p
ects these im
ages. The
p
u
rp
o
s
e is to
redu
ce th
e m
i
ss-in
terpretatio
n
of th
e an
al
y
s
i
s
of t
h
e i
m
ages. P
r
e-
pr
oc
essi
ng t
e
c
hni
q
u
es i
s
requ
ired
in C
A
D system
as first step to enh
a
n
c
e th
e
qual
ity of
W
C
E i
m
ages in
or
de
r to
detect the
bleeding
easily. Recently, resea
r
che
r
s i
n
troduce
d
vari
ous
pre
-
proce
s
sing technique
s
to im
prove C
A
D system
.
The m
a
jor c
o
nt
ri
b
u
t
i
o
n o
f
t
h
e st
u
d
y
i
s
t
o
desi
g
n
a
pre
-
pr
ocessi
ng t
e
c
hni
que
f
o
r c
o
m
put
er ai
ded
detection (C
AD) system
. The aim
is to de
tect the a
b
normalities such
that cla
ssification proce
ss
be
ca
m
e
easi
e
r. T
o
det
ect
t
h
e
W
C
E
im
ages ani
s
ot
r
opi
c
di
ff
usi
o
n
pr
o
pose
d
by
[1]
was
f
u
rt
he
r en
ha
nced
by
usi
n
g
Evaluation Warning : The document was created with Spire.PDF for Python.
I
S
SN
:
2
088
-87
08
I
J
ECE
Vo
l. 6
,
N
o
. 4
,
Au
gu
st 2
016
:
16
17
–
1
626
1
618
Hessi
an
m
a
t
r
ix,
w
h
i
c
h
get
s
a co
nt
rast
de
scri
pt
i
o
n
of
o
n
e
poi
nt
i
n
a
n
im
age. H
o
we
ver
,
t
h
i
s
m
e
t
hod
has
d
r
awb
ack
s
where
W
C
E im
ag
es still h
a
s a n
o
i
se and
no
t sh
arp. An
o
t
h
e
r
drawb
a
ck
is qu
ality o
f
th
e i
m
a
g
e will
be de
gra
d
e i
f
n
u
m
b
er of i
t
e
rat
i
on
goes
u
p
. T
h
en
, we s
o
l
v
e
d
t
h
i
s
pr
obl
em
by
i
n
t
r
o
d
u
ci
n
g
vari
ance
fo
rm
ul
a i
n
their schem
e
[
2
]. To enhanc
e the W
C
E image, Disc
ret
e
Co
sin
e
Transform
(DCT) is i
m
p
l
e
m
en
ted
with
ani
s
ot
r
o
pi
c co
nt
rast
di
ff
usi
o
n i
n
o
r
der t
o
get
hi
gh
q
u
al
i
t
y
W
C
E
i
m
age w
h
i
c
h i
s
m
o
re
sha
r
pnes
s
.
Im
age
shar
pe
ni
n
g
i
s
u
s
ed t
o
m
a
ke fi
ne
det
a
i
l
s
m
o
re cl
ear
or
hi
g
h
l
i
ght
e
d
by
am
pl
ify
t
h
e
hi
gh
f
r
e
que
ncy
of
t
h
e
i
m
age.
More
ove
r, R
G
B is use
d
in
our propose
d
algorithm
as
color space t
o
re
pre
s
ent the
details of
W
C
E im
age. The
rest
o
f
t
h
e pa
p
e
r
i
s
or
ga
ni
zed
as fol
l
o
ws.
Se
ct
i
on 2 pre
s
ent
s
t
h
e rel
a
t
e
d w
o
r
k
s
i
n
pre
-
pr
o
cessi
ng
t
ech
ni
que
f
o
r
C
AD sy
st
em
and al
s
o
m
a
t
h
em
at
i
cal back
gr
o
u
n
d
o
f
t
h
i
s
pape
r are
bri
e
fl
y
expl
ai
ne
d.
Pro
p
o
sed m
e
tho
d
i
s
descri
bed i
n
S
ect
i
on 3 a
nd f
o
l
l
o
we
d by
e
x
peri
m
e
nt
al
re
sul
t
s
dem
onst
r
a
t
ed i
n
Sect
i
on
4. Fi
nal
l
y
, co
n
c
l
udi
n
g
rem
a
rks o
f
t
h
e
pape
r i
s
gi
ve
n i
n
Sect
i
o
n
5.
2.
RELATED WORKS
B
l
eedi
ng i
s
us
ed as an i
ndi
c
a
t
i
on o
f
s
o
m
e
severe c
o
ndi
t
i
ons a
n
d di
seas
es suc
h
as va
s
c
ul
ar l
e
si
o
n
s
,
tum
o
rs and C
r
ohn’s
disease.
Comm
on co
lor sp
ace u
s
ed is
RGB wh
ich is
c
o
n
t
ain
e
d th
r
e
e ch
an
nels; r
e
d, gr
een
and bl
ue c
h
annel. Anot
her c
o
l
o
r
spaces
use
d
are
HIS,
HSV, YC
bCr a
n
d C
I
E La
b.
HIS a
n
d HSV a
r
e t
h
e two
m
o
st co
mm
on visions a
n
d perception
repre
s
entation
of
points in RGB
color s
p
ace.
R
e
searche
r
s i
n
troduce
d
m
a
ny
pre
-
p
r
oc
essi
ng t
e
c
h
ni
q
u
es.
In
[
3
]
,
w
e
i
ght
fact
or i
s
use
d
t
o
i
d
ent
i
fy
bl
eedi
ng
by
deri
vi
n
g
feat
u
r
es o
f
b
l
eed
ing
reg
i
on
co
lor. Six
statistical
p
a
ram
e
ters are in
tr
oduced in [4] to ex
tract the features from
bleeding in
spatial cha
r
acteristics in
HI
S color
s
p
ace from
the
im
ages. In [5],
c
o
lor s
p
aces suc
h
a
s
R
G
B,
HSB a
n
d
YUV
are investigate
d
to see whic
h color sp
ac
e i
s
abl
e
t
o
di
scl
o
se l
e
si
on st
ruct
ure an
d ge
om
et
ry
, col
o
r a
nd t
e
xt
u
r
e
to
d
e
term
in
e an
d to
b
e
con
s
idered in
t
h
eir analysis.
Hi
st
o
g
ram
based i
nde
x i
m
ag
e i
s
used i
n
[6]
t
o
ext
r
act
col
o
r t
e
xt
ure
of bl
e
e
di
n
g
. I
n
[
7
]
,
r
a
t
i
o
of R
t
o
G p
i
x
e
l in
ten
s
ity an
d
d
i
fferent statis
tical p
a
ra
m
e
ters are
us
ed to extract the bleeding.
In
[8
],
RG
B and
HI
S ar
e
use
d
as color
spaces in e
x
tra
c
tion of blee
ding
features
. T
h
ey rem
ove da
rk
pixels from
W
C
E im
ages since
they
are difficult
to be r
ecognized by hum
a
n
eyes.
In [9], CIE
La
b
c
o
lor
space is
use
d
a
s
col
o
r
space
s i
n
thei
r
pr
o
pose
d
t
ech
ni
q
u
es. F
o
u
r
fu
nct
i
o
ns are use
d
i
n
t
h
ei
r schem
e
i
n
cl
ud
i
ng pa
ram
e
t
e
r
s
whi
c
h
depe
nd
o
n
eig
e
nv
alu
e
s of
Hessian an
d Lap
l
acian
t
o
d
e
t
ect th
e
b
l
eed
i
n
g
reg
i
o
n
.
In [10
]
,
h
i
stog
ram
p
r
ob
ab
ility is used
b
y
calculating m
e
an and standa
rd
devi
at
i
o
n t
o
ext
r
act
c
o
l
o
r feat
u
r
es o
f
bl
eedi
n
g.
In t
h
i
s
p
r
o
p
o
sed
m
e
t
hod
,
ani
s
ot
r
o
pi
c di
f
f
u
si
o
n
al
so
k
n
o
w
n as
Per
o
na-
M
al
i
k
di
ff
usi
o
n, a
very
p
o
pul
ar t
ech
ni
q
u
e ai
m
i
ng t
o
re
du
ce
im
age
noi
se
i
s
em
pl
oy
ed.
It
i
s
use
d
t
o
e
n
hance
W
C
E i
m
ages by
re
duci
n
g
noi
s
e
wi
t
h
o
u
t
rem
ovi
ng
t
h
e
si
g
n
i
fi
cant
p
a
rts
o
f
t
h
e i
m
ag
e con
t
en
t. In
[1
1
]
, artificial fish
-warm
algorithm
was use
d
to e
nhance m
e
dical digital
radi
og
rap
h
y
(
D
R
)
i
m
age whi
c
h i
s
effect
i
v
e i
n
el
im
i
n
at
i
ng n
o
i
s
e an
d en
hanci
ng t
h
e det
a
i
l
wi
t
h
hi
gh
effectiv
en
ess an
d
rob
u
stn
e
ss. Mu
lti-wav
e
let tran
sform
an
d
med
i
an
filter is u
s
ed
in
[12
]
to
rem
o
v
e
th
e i
m
p
u
l
se
noi
se
vi
e
w
ed
a
s
ra
nd
om
noi
se
fr
om
t
h
e bl
u
r
r
e
d
un
de
rwat
er
im
age.
2.
1.
Aniso
t
r
o
pic D
i
ffusi
on
Perona
-Malik
m
odel has introduce
d
a ne
w defi
n
ition
of
scale-space a
n
d class algorit
h
m
s
using a
di
ff
usi
on
pr
oce
ss t
o
av
oi
d t
h
e
bl
u
rri
n
g
a
nd al
so l
o
cal
i
zat
i
on
pr
o
b
l
e
m
s
of l
i
n
ear di
f
f
u
si
o
n
fi
l
t
e
ri
ng.
A
n
i
s
ot
r
opi
c
di
ff
usi
on m
e
t
hod
pr
op
ose
d
by
Pero
na
-M
al
i
k
i
s
m
a
t
h
em
at
i
cal
l
y
form
ul
at
ed as a di
ffusi
on
pr
ocess
,
and
enco
u
r
ages i
n
t
r
a-
regi
on sm
oot
hi
n
g
i
n
pre
f
e
r
ence t
o
sm
oot
hi
n
g
acr
oss t
h
e bo
u
nda
ri
es.
It
i
s
a very
p
o
p
u
l
a
r
t
echni
q
u
e t
o
r
e
duce
i
m
age noi
se,
whi
c
h i
s
al
so use
d
t
o
e
nha
nce
WC
E i
m
ages, by
red
u
ci
n
g
noi
se
wi
t
h
o
u
t
rem
o
v
i
n
g
t
h
e sig
n
i
fican
t
p
a
rts o
f
th
e im
ag
e co
n
t
en
t.
In filterin
g
m
e
th
od
, th
e estim
a
tio
n
o
f
th
e l
o
cal imag
e
stru
cture is dri
v
en
b
y
th
e
k
n
o
w
led
g
e
on
the statistics
o
f
th
e no
ise degr
ad
atio
n
and
edge str
e
ng
th
s
[
13],[
14
].
Per
ona
-M
al
i
k
m
odel
i
s
out
pe
rf
orm
e
d com
p
ared t
o
C
a
nny
edge
det
ect
or
whi
c
h i
s
edge
s
persi
s
t
e
d st
a
b
l
e
ove
r
a very
l
o
n
g
t
i
m
e even
wi
t
h
o
u
t
usi
n
g
no
n-m
a
xi
m
a
suppr
essi
on a
n
d
hy
st
ere
s
i
s
t
h
res
hol
di
n
g
.
Gi
ve
n t
h
e
o
r
i
g
i
n
al
im
age
u
0
(
x
,y,t)
and t
h
e sm
oot
he
d ver
s
i
o
ns
com
p
ri
se a f
a
m
i
ly
im
ages
u(x,y,t)
where
t
is
th
e a
m
o
u
n
t
of
sm
oot
hi
ng
.
An
i
s
ot
ro
pi
c
di
ff
us
i
on
o
f
Pe
ro
na
-
M
al
i
k
i
s
de
fi
ne
d as
[
13]
:
(1)
whe
r
e
div is
the dive
rgence
operat
or,
is th
e sp
atial grad
ient,
is th
e grad
ien
t
m
a
gni
t
ude
, an
d g
(
x
,
y
,
t
)
i
s
a di
ff
usi
on c
o
ef
f
i
ci
ent
.
g(
x,y
,
t
)
cont
r
o
l
s
t
h
e r
a
t
e
of di
f
f
u
s
i
o
n. The i
s
ot
ro
p
i
c heat
di
ff
usi
o
n
eq
uat
i
on,
is redu
ced if
g
(
x
,
y,t) is a
co
nstan
t
.
(
2
)
Evaluation Warning : The document was created with Spire.PDF for Python.
I
J
ECE
I
S
SN
:
208
8-8
7
0
8
Pre-
proce
ssi
n
g
Tech
ni
q
u
e f
o
r
Wi
rel
e
ss C
a
ps
ul
e E
n
d
o
sc
opy
Im
ag
e E
n
h
a
n
c
e
ment
(
R
os
di
a
n
a
S
h
ahri
l
)
1
619
Di
scret
i
ze t
h
e
equat
i
o
n
(1
) an
d 4
neare
s
t
-
nei
g
h
b
o
rs
di
scret
i
zat
i
on was
p
r
o
pos
ed
by
Per
o
na-M
al
i
k
as
fo
llows:
(3
)
whe
r
e,
fo
r t
h
e
num
eri
cal
sche
m
e
t
o
be
st
abl
e
an
d t
h
e
sy
m
bol
i
ndi
cat
es nea
r
est
-
nei
g
hb
or
s di
ffe
re
nces,
(4)
Each i
t
e
rat
i
on
up
dat
e
s t
h
e co
nd
uct
i
o
n coe
f
f
i
ci
ent
s
as a funct
i
on
of t
h
e
m
a
gni
t
ude
of t
h
e bri
ght
ness
gra
d
i
e
nt
:
(
5
)
whe
r
e,
g
has
t
w
o
f
u
n
c
t
i
o
n
di
ffe
rences
an
d
d
e
fi
ne
d as:
(
6
)
The
pa
ram
e
ter of consta
nt K is a c
o
nductanc
e
pa
ra
m
e
ter th
at in
flu
e
n
ces th
e
d
i
ffu
s
i
o
n pro
cess
wh
ich
is sepa
rates f
o
r
w
ar
d
(lo
w
c
ont
rast)
fr
om
backward (hi
g
h
contrast) diffusion
areas.
2.
2.
Hessian
m
a
tri
x
Hessi
an m
a
t
r
ix
i
s
a
s
qua
re m
a
t
r
i
x
of sec
o
nd
-
o
r
d
er
pa
rt
i
a
l
deri
vat
i
v
es
o
f
a
f
unct
i
o
n a
nd i
t
used a
s
ap
pro
ach in th
i
s
wo
rk
t
o
g
e
t a con
t
rast
d
e
scrip
tio
n
o
f
on
e
po
in
t i
n
an
im
ag
e. Hessian m
a
t
r
ix
o
f
o
n
e
po
int in
a
g
r
ay im
ag
e which
is
d
e
fin
e
d as fo
llo
ws [1
]:
(
7
)
whe
r
e,
are
the second-order derivative of
th
e i
m
ag
e alon
g dir
ectio
n
o
f
x, y, x
y
respectively a
n
d
[
1
]
ha
s
pr
op
ose
d
a
ne
w c
once
p
t
of
co
nt
r
a
st
as f
o
l
l
o
ws:
(
8
)
whe
r
e,
and
i
s
a t
w
o
ei
ge
n
val
u
e o
f
Hessi
an
m
a
t
r
i
x
a
n
d
de
n
o
t
e
d
by
[
1
5]
.
(9)
Em
ploying this
contrast s
p
ace, pr
oposed by
[1], c
h
ange t
h
e
ori
g
in
al a
n
isotropic
diffusion i
n
to:
(10)
Evaluation Warning : The document was created with Spire.PDF for Python.
I
S
SN
:
2
088
-87
08
I
J
ECE
Vo
l. 6
,
N
o
. 4
,
Au
gu
st 2
016
:
16
17
–
1
626
1
620
2.
3.
Ga
ussia
n
filter
Gau
s
sian
filterin
g
is u
s
ed
in th
e
sm
o
o
t
h
i
ng o
f
an
i
m
ag
e t
o
rem
o
v
e
n
o
i
se in
W
C
E im
a
g
es. In
ou
r
p
r
op
o
s
ed
m
e
th
o
d
, two
d
i
m
e
n
s
io
n
a
l of Gau
s
sian
filter is
u
s
ed
as a k
e
rn
el to
co
n
v
o
l
ve with
an
im
ag
e sin
c
e
w
o
r
k
in
g
w
i
th
ima
g
e
.
Out
put
=
f
*
z, w
here
f
is
an image and
g
is
a k
e
rn
el. Gau
s
sian
filter is
d
e
fin
e
d as:
(
1
1)
whe
r
e,
i
s
a st
anda
r
d
de
vi
at
i
o
n
o
f
t
h
e
Ga
us
si
an di
st
ri
but
i
o
n.
It
i
s
acc
om
pl
i
s
he
d by
c
o
n
vol
vi
n
g
bet
w
e
e
n a
kernel and image. The
output i
m
age pixel is cal
culated by
m
u
ltiplying each
kernel value by the
co
rresp
ond
ing
in
pu
t im
ag
e p
i
x
e
l.
3.
THE PROPOSED
METHOD
Pr
e-pr
o
cessing tech
n
i
qu
e is
pr
opo
sed
i
n
th
is p
a
p
e
r a
nd
pre
s
ents the ste
p
-by-step
for
WCE im
ages
enha
ncem
ent
.
In t
h
e p
r
op
ose
d
m
e
t
hod
, co
nt
rast
enha
ncem
ent
m
e
t
hod i
s
em
pl
oy
ed t
o
m
a
ke m
o
re cont
r
a
st
o
n
the
W
C
E im
ages com
p
ared t
o
these
usin
g
t
h
e orig
i
n
al concep
t in
tro
d
u
c
ed
b
y
B. Li
[1
]. B. Li’s algo
ri
th
m
is
base
d on a
n
i
s
o
t
ro
pi
c cont
ra
st
di
ff
usi
on an
d
Hessi
an m
a
t
r
i
x
. Here,
vari
a
n
c
e
form
ul
a i
s
i
n
tro
d
u
ced t
o
o
v
e
r
com
e
the B. Li’s wea
kne
sses. In
order to m
a
ke the
details of
each
im
age to be more
visible, s
h
a
r
pe
ning algorit
h
m
is
pr
o
pose
d
t
o
ea
se t
h
e cl
assi
fi
c
a
t
i
on
pr
ocess
o
f
t
h
e
a
bnorm
alities such as
ble
e
ding in
W
C
E
im
ages [16].
3.
1.
Contr
a
st Im
age Enhancem
ent
In
ou
r
pr
op
ose
d
m
e
t
hod, a
n
i
s
ot
r
opi
c c
ont
r
a
s
t
di
ff
usi
o
n i
s
e
m
pl
oy
ed t
o
co
nt
rast
t
h
e i
m
ages
due t
o
lo
w
d
a
rk
q
u
a
lity o
f
th
e
W
C
E
i
m
ag
es. It also ab
le t
o
m
a
k
e
ch
aracteristics in
a
WCE im
a
g
e is m
o
re v
i
si
b
l
e
b
y
hum
an eyes als
o
by com
puter
machine.
Due
t
o
t
h
e
w
eakne
ss
of
B
.
Li
’s a
s
e
xpl
a
i
ned i
n
Sect
i
o
n
1,
t
h
us
vari
ance i
s
i
n
t
r
od
uced
i
n
t
h
i
s
pr
o
pose
d
m
e
t
hod t
o
o
v
erc
o
m
e
t
h
e weak
ness
. The v
a
ri
ance
(Va
r
2
) is a m
e
asure
of the spread
of
pixel values
aroun
d th
e im
a
g
e m
ean
. By usin
g th
e
v
a
riance, it esti
m
a
tes
th
e co
n
t
rast
p
r
o
b
a
b
ility d
i
strib
u
tion
of th
e imag
e.
It will g
i
v
e
an
id
ea h
o
w th
e pix
e
l v
a
lu
e spread
in
i
m
ag
e. A s
m
all
v
a
rian
ce
m
ean
s th
at th
e d
a
ta p
o
i
n
t
s t
e
n
d
to
be ve
ry close t
o
the m
ean while high va
ria
n
ce m
eans
th
at th
e d
a
ta po
ints sp
read
ou
t aroun
d
the m
e
a
n
and
from
each othe
r. The
r
efore
,
it
doe
s
not
de
gra
d
e th
e
quality of the im
age. The
varia
n
ce
form
ula (Var
2
) is:
(
1
2)
whe
r
e,
is a m
e
an
of
4
nearest
-
nei
g
hbors
, N
= size of im
ag
e m
× n and
i
s
a pi
xel
of
i
m
age
i
.Th
e
fo
llowing
form
u
l
a is u
s
ed
to calcu
late t
h
e m
ean of
4-nearest-neighbors,
:
(
1
3)
The E
q
.
(
9
)
i
s
r
e
vi
sed
by
usi
n
g st
a
nda
rd
va
ri
ance,
(
V
ar
2
) as fo
llo
ws:
(
1
4)
In
o
r
d
e
r to
g
e
t
m
o
re h
i
gh
qu
ality W
C
E i
m
ag
e wh
ich
i
s
sh
arp
e
r,
DC
T is u
s
ed
wit
h
an
iso
t
ro
p
i
c
cont
rast
di
f
f
u
s
i
on
whe
r
e DC
T i
s
enabl
i
n
g
t
o
sha
r
p
WC
E
im
age and t
h
e det
a
i
l
s
i
s
expl
ai
ned i
n
t
h
e
nex
t
sect
i
on.
3.
2.
Shar
p
ening I
m
age
Using Discrete Cosine
Trans
f
orm
(DCT
)
A DCT can
b
e
u
s
ed
in
im
ag
e p
r
o
cessi
n
g
such
as
p
a
ttern
reco
gn
itio
n and filterin
g
b
y
com
p
u
tin
g
th
e
fast
Fo
uri
e
r t
r
a
n
sf
orm
i
s
deve
l
ope
d.
In
WC
E
im
age shar
pen
i
ng, t
h
e i
d
ea i
s
t
o
use
hi
g
h
f
r
e
que
ncy
com
p
o
n
ent
s
th
at are
ex
tracted
fro
m
th
e
o
r
i
g
in
al im
ag
es to
h
i
gh
ligh
t
th
e inv
i
sib
l
e
details. High
freq
u
e
n
c
y is
u
s
ed
t
o
det
e
rm
i
n
e or
d
e
t
ect
t
h
e ed
ge i
n
f
o
rm
at
i
on. E
x
am
pl
es of
suc
h
m
e
t
hods
are
S
obel
,
C
a
n
n
y
,
a
n
d
u
n
s
h
ar
p m
a
ski
n
g
but these m
e
thods
are m
o
re s
e
nsitive to
noi
se. S
h
arpeni
ng im
age is done
, by re
stori
ng
high freque
ncy
into
o
r
i
g
in
al i
m
ag
e. O
n
th
e o
t
h
e
r
han
d
, b
l
ur
r
e
d
imag
e is p
r
od
uced
b
y
lo
ss of
h
i
gh
f
r
e
qu
en
cy
. I
n
our
m
e
th
od
, hi
gh
Evaluation Warning : The document was created with Spire.PDF for Python.
I
J
ECE
I
S
SN
:
208
8-8
7
0
8
Pre-
proce
ssi
n
g
Tech
ni
q
u
e f
o
r
Wi
rel
e
ss C
a
ps
ul
e E
n
d
o
sc
opy
Im
ag
e E
n
h
a
n
c
e
ment
(
R
os
di
a
n
a
S
h
ahri
l
)
1
621
freq
u
e
n
c
y con
t
ain
i
n
g
t
h
e fin
e
d
e
tails o
f
th
e orig
in
al im
ag
e i
s
ex
tracted
b
y
u
s
ing
DCT th
en
it is co
m
b
in
ed
wit
h
th
e
W
C
E o
r
i
g
i
n
al i
m
ag
e to
p
r
odu
ce a n
e
w imag
e with
i
m
p
r
ov
ed
q
u
a
lity in
v
i
su
al app
e
aran
ce. Th
e fo
llowing
math
e
m
atica
l
m
o
d
e
l is assumed
fo
r i
n
pu
t
RGB im
ag
e:
whe
r
e
u
,
z
and
n
are
the i
n
put
noisy im
age, the
noise
free
imag
e and the
noise com
p
one
n
t resp
ectiv
ely i
n
red,
g
r
een
an
d
b
l
u
e
ch
ann
e
l.
In
human
eye, it is
m
o
re sen
s
itiv
e to
flick
e
ri
ng
of h
i
g
h
sp
atial frequ
e
n
c
ies th
an
low
sp
atial frequ
encies. By u
s
ing
th
is con
c
ep
t,
DCT is propo
sed
in ou
r sch
e
me to
am
p
lify th
e i
m
ag
e frequ
en
cy to
sharp t
h
e im
age. T
h
e
DCT
of
a data se
quenc
e
u(x,y)
, x=
0,
1, …,
(M-1
)
,
y=
0,
1,
…,
(N-1
)
i
s
de
fi
ne
d as
[
17]
(1
5)
whe
r
e
(x,
y
)
is
t
h
e
d
-th DC
T coefficient,
u(x,y)
represe
n
t
s
t
h
e i
m
age dat
a
, and M
,
N i
s
t
h
e wi
dt
h a
nd l
e
ngt
h
of
u(x,y)
. In
th
is step
, DCT is u
s
ed
t
o
ex
tract th
e h
i
g
h
fre
que
ncy
i
n
W
C
E im
age. Then
, hi
g
h
f
r
eq
ue
n
c
y
i
s
ap
p
lied to
t
h
e
o
r
i
g
in
al im
ag
e to
sh
arp
e
n the
WCE im
age as shown i
n
Fi
gure
1.
Fi
gu
re
1.
C
o
m
p
o
n
e
n
t
o
f
ori
g
i
n
al
, s
h
ar
pe
ned
and
bl
ur
red
i
m
age i
n
f
r
eq
ue
n
c
y
dom
ai
n
Let
X
be the s
h
arp im
age,
I
be
t
h
e ori
g
i
n
al
i
m
age t
h
at
i
s
t
o
be rec
o
vere
d a
nd
Y
be t
h
e hi
g
h
f
r
eq
ue
ncy
,
th
en
t
h
eir
relatio
n is represen
t
e
d
u
s
ing
t
h
e follo
wing
eq
u
a
ti
o
n
:
Thi
s
im
age i
s
trans
f
orm
e
d i
n
t
o
DC
T bl
oc
ks
i
n
t
h
e fre
que
nc
y
dom
ai
n.
The
fi
rst
DC
T coe
ffi
ci
ent
t
op
left in
b
l
o
c
k
DCT is F(0
,
0),
wh
ere it represen
ts th
e av
erag
e in
ten
s
ity o
f
t
h
e b
l
o
c
k. It is also
kn
own
as
th
e DC
com
pone
nt
o
r
ener
gy
bl
oc
ks.
Ot
he
r bl
oc
ks of DC
T
c
o
e
ffi
ci
ent
s
are
cal
l
e
d AC
coe
ffi
ci
e
n
t
s
. Gen
e
ral
l
y
,
sha
r
p
spaces s
u
c
h
as
edges c
o
rres
pond t
o
hi
gh fre
que
ncy
re
gi
on
while long
unc
hanging s
p
ace
s corres
p
ond t
o
low
fre
que
ncy
re
g
i
on
[1
7]
,[
1
8
]
.
Aft
e
r com
b
i
n
i
n
g t
h
e
hi
g
h
fre
quency with
ori
g
inal
i
m
age, the image is
t
r
ans
f
o
r
m
e
d back i
n
t
o
t
h
e s
p
at
i
a
l
dom
ai
n using t
h
e i
nve
rse
cosi
ne
di
scret
e
t
r
ansf
orm
(IC
DT)
whi
c
h i
s
d
e
fi
ne
d
by
:
(1
6)
w
h
er
e m
=
0
,
1,…,
(
M
-1)
an
d n=0
,
1
,
…,
(N
-1
).
Evaluation Warning : The document was created with Spire.PDF for Python.
I
S
SN
:
2
088
-87
08
I
J
ECE
Vo
l. 6
,
N
o
. 4
,
Au
gu
st 2
016
:
16
17
–
1
626
1
622
The
fol
l
o
wi
n
g
i
s
t
h
e al
g
o
ri
t
h
m
for t
h
e
p
r
o
p
o
se
d m
e
t
hod i
m
pl
em
ent
a
t
i
on:
Step 1:
Read
im
age
Step 2:
C
a
l
c
ul
at
e Hess
i
a
n m
a
t
r
i
x
and
get
ei
ge
nv
al
ue
s,
v
1
and
v
2
Step 3:
Calculate
Step 4:
Start nu
m
b
er of iteration
for each
ch
ann
e
l, RGB
Step 4.
1:
Red
u
ce
no
ise u
s
ing
Gau
s
sian
filter,
G = Ga
ussi
an
(
c
,
)
Step 4.
2:
C
o
nvo
lv
e t
h
e im
ag
e b
e
tween
k
e
rn
el
(Gau
ssian
filter),
I =
G*
c
Step 4.
3:
Appl
y the standard
varia
n
ce,
Step 4.
4:
C
a
l
c
ul
at
e t
h
e
di
f
f
us
ed i
m
age by
us
i
ng
a
n
i
s
ot
ro
p
i
c Per
ona
-M
al
i
k
,
u=d
i
v (g*I
)
Step 5:
Get
C
min
and
C
max
Step 6:
The
di
f
f
use
d
re
sul
t
i
s
t
r
a
n
sf
or
m
e
d bac
k
t
o
i
m
age space
,
O
u
t
p
ut
=
[
(
c – c
min
)/
(
c
max
- sc
min
)
]
*25
5
Step 7:
Ap
ply
the DC
T,
C
d
In
th
is algorithm
,
C
min
and
C
max
is
obt
ai
ne
d
fr
om
whe
r
e
C
min
and
C
max
is th
e m
i
n
i
m
u
m
a
n
d
ma
x
i
mu
m
v
a
l
u
e
o
f
c(x,y
), res
p
ectively.
4.
EX
PER
I
M
E
NTA
L
R
E
SU
LTS AN
D DISC
USSION
M
a
t
l
a
b R
201
1
b
i
s
used i
n
t
h
i
s
expe
ri
m
e
nt. Fo
ur i
m
ages are used a
nd
t
h
e expect
e
d
r
e
sul
t
of t
h
i
s
p
a
p
e
r is to
pro
d
u
ce
h
i
gh
qu
ality W
C
E imag
es th
at are sh
arp
e
r an
d co
n
t
rast i
m
a
g
es. In
co
lo
r
i
m
ag
e
enha
ncem
ent techni
que
, it is very
hard to a
n
alyze the
qua
nt
itative effect, a
s
there is
no s
p
ecific
m
easure
m
en
t
to validate the
quality of the
color im
age.
Here
, c
o
m
p
arison
betwee
n B
.
Li’s
m
e
thod and
propose
d
m
e
thod
usi
n
g vari
a
n
ce (wi
t
h
o
u
t
sha
r
p
e
ni
n
g
usi
n
g D
C
T)
i
s
prese
n
t
e
d
i
n
Fi
g
u
r
e 2.
Fi
gu
re
2.
The
ori
g
i
n
al
i
m
ages com
p
ared
t
o
B
.
Li
’s a
n
d
pr
o
pos
ed
m
e
t
hod
Ori
g
in
al im
ag
es in
th
is fig
u
re h
a
v
e
poo
r
qu
ality an
d
are v
a
gu
e.
However, th
e enh
a
n
c
ed
i
m
ag
e b
y
t
h
e pr
o
p
o
s
ed
m
e
t
hod
becom
e
m
u
ch bet
t
e
r
and m
o
re co
nt
rast
, an
d
not
d
e
gra
d
i
n
g t
h
e i
m
age. Thi
s
ca
n be see
n
th
at ev
en
thou
gh
B. Li’s m
e
th
o
d
d
o
m
a
k
e
some en
h
a
n
cem
e
n
t to th
e orig
i
n
al bu
t du
e to
it
s li
m
itat
i
o
n
and
so
m
e
dra
w
back as d
i
scusse
d i
n
pre
v
i
o
use s
ectio
n. Th
e in
trodu
ctio
n
o
f
v
a
rian
ce in
B. Li’s
m
e
th
od
prov
es to g
i
ve
bet
t
e
r e
nha
nce
m
ent
of
t
h
e i
m
ages.
Qu
ality m
easu
r
e an
alysis on
WCE im
ag
es u
s
ing
Second
Deri
v
a
tiv
e lik
e Measurem
en
t (SDME) and
Edg
e
Based
C
o
n
t
rast (EBC
M) as th
e p
e
rfo
r
m
a
n
ce esti
ma
tio
n
stand
a
rd
is co
n
s
i
d
ered
.
EBCM is v
e
ry sen
s
itive
t
o
co
nt
o
u
r
s
or
edge
s
due
t
o
b
a
sed
h
u
m
a
n pe
rcept
i
o
n.
EB
C
M
i
s
de
fi
ne
d
b
y
[1
9]
,[
2
0
]
.
Ori
g
inal
Im
ages
En
hance
d
Im
ages using
Pr
opo
sed
Meth
od
En
hance
d
Im
ages using
Bi-Li’s
Meth
o
d
Patien
t
-A
Patien
t
-B
Patien
t
-C
Patien
t
-D
Evaluation Warning : The document was created with Spire.PDF for Python.
I
J
ECE
I
S
SN
:
208
8-8
7
0
8
Pre-
proce
ssi
n
g
Tech
ni
q
u
e f
o
r
Wi
rel
e
ss C
a
ps
ul
e E
n
d
o
sc
opy
Im
ag
e E
n
h
a
n
c
e
ment
(
R
os
di
a
n
a
S
h
ahri
l
)
1
623
(17)
C
ont
ra
st
c(i,j)
for a
pixel
of i
m
age X l
o
cate
d
at
(i,
j
)
i
s
t
hus de
fi
ne
d as
(18)
whe
r
e,
g(
k,l
)
i
s
t
h
e e
dge
val
u
e
at
pi
xel
(k,
l
)
and
is th
e
n
e
ighb
oring
p
i
x
e
ls at
(i,
j
)
.
On
t
h
e
ot
he
r
han
d
,
S
D
M
E
di
vi
de
an
i
m
age i
n
t
o
k
1
× k
2
bloc
ks
and t
h
en ave
r
a
g
e
values
of t
h
e
measure
result
s of all bl
ock is
calculated i
n
t
h
e e
n
t
i
r
e i
m
age. S
D
M
E
i
s
de
fi
ne
d
by
[
21]
(19)
B
y
usi
n
g b
o
t
h
SDM
E
a
n
d EB
C
M
, i
t
i
s
pr
ove
d t
h
at
t
h
e p
r
op
ose
d
m
e
t
hod
(
b
l
ack l
i
n
e
)
i
s
b
e
t
t
e
r t
h
an B
.
Li
’s m
e
t
hod (y
el
l
o
w l
i
n
e) as
sho
w
n i
n
Fi
g
u
r
e 3 a
n
d Fi
g
u
r
e
4. T
h
e
hi
g
h
e
r
val
u
e o
f
S
D
M
E
and
EB
C
M
m
eans
th
at th
e im
ag
e
s
h
a
v
e
a b
e
tter q
u
a
lity. It sh
ows th
at
v
a
riance is ab
le to
solv
e th
e weakn
e
ss o
f
B
.
Li’s m
e
th
od
wh
ereb
y th
e qu
ality o
f
i
m
ag
e will n
o
t
d
e
g
r
ad
e wh
en
th
e
nu
m
b
er o
f
iterat
i
o
n
is in
creased
.
Here th
e
num
b
e
r o
f
iteratio
n
is d
e
fi
n
e
d
as th
e nu
m
b
er of rep
e
titin
g
th
e p
r
o
ce
ss
or iteratin
g
op
eratio
n
s
yield
i
n
g
resu
lt clo
s
er to
th
e
desi
re
d one
.
Fi
gu
re
3.
S
D
M
E
m
easure pl
ot
f
o
r
WC
E i
m
age
A-
D
Fi
gu
re
4.
EB
C
M
m
easure pl
o
t
fo
r
WC
E i
m
age
A-
D
Hi
st
o
g
ram
i
s
use
d
t
o
see t
h
e di
st
ri
but
i
o
n o
f
col
o
r i
n
a
n
im
age i
n
re
d, g
r
e
e
n an
d bl
ue c
h
annel
.
T
w
o
cri
t
e
ri
a can be
i
d
ent
i
f
i
e
d i
n
hi
st
o
g
ram
ei
t
h
er t
h
e i
m
age
i
s
un
dere
x
pos
ure
or l
a
c
k
o
f
cont
r
a
st
. B
y
us
i
n
g
histogram
,
the characteristics
of an im
age can be
detected such as eithe
r
an
i
m
ag
e is satu
ration
,
sp
ikes an
d
gaps a
n
d also
provide t
h
e inform
ation
ab
o
u
t
co
nt
rast
,
sha
r
p
an
d
ot
he
rs t
o
i
n
terpret t
h
e
visual of im
age. Figure
5 s
h
o
w
s
t
h
e
hi
st
og
ram
for i
m
age
A-
D i
n
re
d,
gr
een
an
d
b
l
ue cha
n
nel
.
D
i
st
ri
but
i
o
n
nu
m
b
er of
pi
xel
can
b
e
seen
from
these histogram
s
at each c
h
a
n
nel level.
Evaluation Warning : The document was created with Spire.PDF for Python.
I
S
SN
:
2
088
-87
08
I
J
ECE
Vo
l. 6
,
N
o
. 4
,
Au
gu
st 2
016
:
16
17
–
1
626
1
624
Fi
gu
re 5.
The
gra
p
h dem
onst
r
at
es
t
h
e hi
st
o
g
r
am
col
o
r di
st
r
i
but
i
o
n
i
n
re
d,
gree
n
a
n
d bl
ue chan
nel
i
n
W
C
E
im
ages A-D
As ca
n see
f
r
o
m
Fi
gure
5
,
t
h
e hi
st
og
ram
s
f
o
r
o
r
ig
in
al i
m
ag
e are
no
t
d
i
stribu
ted
over th
e en
tire
in
ten
s
ity rang
e. Thu
s
, orig
i
n
al i
m
ag
e h
a
s l
o
w con
t
ra
st a
n
d va
gue. C
o
m
p
are to histogram
s
for enhanced
im
ages usi
n
g
B
.
Li
’s m
e
t
hod are
wi
de
r i
n
t
e
nsi
t
y
ran
g
e a
nd t
h
e co
nt
rast
i
n
t
h
e i
m
age is i
n
crease
d
. T
h
e g
o
o
d
q
u
a
lity of th
e i
m
ag
e shou
ld
has wi
d
e
r in
tensity ran
g
e
wh
i
c
h
is t
h
e
p
i
x
e
l
s
shou
ld
b
e
d
i
strib
u
t
ed
ev
en
ly o
v
e
r
th
e who
l
e in
ten
s
ity rrang
e and
p
eak of i
n
ten
s
ity rang
e is
higher that m
e
ans t
h
e im
age is m
o
re sharp. Thus
,
th
is ch
aracteritics is sho
w
ed
i
n
h
i
sto
g
ram
for enh
a
n
c
ed
im
a
g
e
u
s
ing
o
u
r
p
r
o
p
o
s
ed
m
e
th
o
d
wh
ich
is
b
e
tter th
an
hi
st
o
g
ram
for
e
nha
nce
d
i
m
ages usi
n
g
B
.
Li
’s
m
e
t
hod.
PSNR
of
ori
g
i
n
al
and e
n
ha
nced i
m
age o
b
t
a
i
n
ed
by
B
.
Li
’s m
e
t
hod
and
pr
o
pos
e
d
m
e
t
hod i
s
prese
n
t
e
d
i
n
T
a
bl
e 1
.
A
s
we
can see
fr
om
thi
s
t
a
bl
e,
PS
N
R
val
u
e
of
o
u
r
pr
o
pos
ed m
e
tho
d
i
s
bet
t
e
r t
h
an
B
.
Li
’s
m
e
t
hod. T
a
bl
e 2 sh
ows t
h
e m
easurem
ent
of sha
r
pnes
s
i
n
W
C
E i
m
ag
e usi
ng
gra
d
i
e
nt
whi
c
h i
s
m
e
asuri
n
g
ed
g
e
inform
ati
o
n. Let say
I(x
,
y) is an im
ag
e, t
h
en t
h
e
gra
d
i
e
nt
vect
or
i
s
d
e
f
i
ned a
s
(20)
wh
ere t
h
e
p
a
rti
a
l d
e
ri
v
a
tiv
e
with
resp
ect to
x
and
y
i
s
defi
ne
d as
(2
1)
Th
e resu
lt
prov
ed
th
at bo
th
p
r
op
o
s
ed
con
t
rast
enhancem
ent algorithm
an
d
s
h
ar
peni
n
g
WC
E i
m
age
al
go
ri
t
h
m
pro
v
i
de bet
t
e
r per
f
o
rm
ance
com
p
ared with B.
Li
’s algorith
m sin
c
e SDME
an
d EBCM
v
a
lu
e is
stable whe
n
e
v
er num
b
er of i
t
erations inc
r
e
a
ses,
and s
h
arpne
ss m
easurement us
ing
gradient and PSNR are
bot
h i
m
pro
v
ed
by
3
1
.
5
%
an
d
20
.3
%,
res
p
ect
i
v
el
y
.
Tabl
e
1.
PS
NR
val
u
e
f
o
r
en
ha
nced
i
m
age usi
n
g
B
.
Li
’s m
e
tho
d
a
n
d
pr
o
p
o
s
ed m
e
t
hod
(i
n
deci
bel
)
I
m
age B.
Li’
s
P
r
oposed
Patient A
15.
595
04
16.
182
49
Patient B
16.
580
4
17.
457
08
Patient C
14.
951
92
17.
985
42
Patient D
22.
479
94
22.
575
74
Original
Im
ages
E
nha
nce
d
Im
ages
us
i
ng Pr
o
p
o
s
e
d
M
e
thod
E
nhanced Im
ages
using Bi-
L
i
’
s
M
e
thod
Patient-A
Patient-B
Patient-C
Patient-D
Evaluation Warning : The document was created with Spire.PDF for Python.
I
J
ECE
I
S
SN
:
208
8-8
7
0
8
Pre-
proce
ssi
n
g
Tech
ni
q
u
e f
o
r
Wi
rel
e
ss C
a
ps
ul
e E
n
d
o
sc
opy
Im
ag
e E
n
h
a
n
c
e
ment
(
R
os
di
a
n
a
S
h
ahri
l
)
1
625
Tabl
e
2.
Sha
r
p
n
ess m
easurem
ent
s
f
o
r e
nha
nc
ed i
m
age usi
n
g
B
.
Li
’s
m
e
t
hod, a
n
d
pr
o
pose
d
m
e
t
hod
w
h
e
n
num
ber
of
i
t
e
r
a
t
i
on i
s
3
I
m
age B.
Li’
s
P
r
oposed
Patient A
5.
1743
9
6.
1405
2
Patient B
5.
8915
8
7.
7044
1
Patient C
3.
5732
3
3.
7039
9
Patient D
2.
3822
7
2.
4267
3
5.
CO
NCL
USI
O
N
Pr
e-pr
o
cessing tech
n
i
qu
e is
a v
e
r
y
im
p
o
r
tan
t
pr
o
cess in CA
D
system b
e
fo
r
e
d
i
agnosin
g
a
W
C
E
i
m
ag
es in
o
r
d
e
r to
en
h
a
n
ce low qu
ality W
C
E i
m
ag
e, wh
ich
is no
isy, d
a
rk an
d
v
a
g
u
e
or
u
n
c
ertain
tex
t
ure in
an
im
age. The ai
m
of pre
-
proc
essing technique is to
sim
p
lify a classificatio
n
task
in CAD system.
W
e
introduced
va
riance val
u
e in
an
i
s
ot
r
o
pi
c co
nt
rast
di
ff
usi
o
n t
o
o
v
erc
o
m
e
t
h
e d
r
aw
bac
k
of
ori
g
i
n
al
al
g
o
ri
t
h
m
.
Th
en
, in
ord
e
r
to
g
e
t m
o
re sharpn
ess i
n
th
e i
m
ag
es, DCT is u
s
ed
wh
ich
is
sh
own
to
b
e
effectiv
e in
en
h
a
n
c
ing
the
W
C
E im
ages. The
res
u
lts proved that
t
h
e
pr
op
ose
d
m
e
tho
d
i
s
bet
t
e
r t
h
an B
.
Li
’s m
e
tho
d
.
Fut
u
re
res
earch
work shou
ld
focu
s
o
n
how to
n
o
rm
alize th
e
co
lor
W
C
E im
age, si
nce a
b
norm
alities su
ch as blee
ding c
o
lor
has
vari
ous
re
d c
o
l
o
rs
i
n
o
r
de
r t
o
ease cl
assi
fi
cat
i
on t
a
s
k
i
n
C
A
D sy
st
em
.
ACKNOWLE
DGE
M
ENTS
Th
is research
i
s
p
a
rtially supp
or
ted
by Uni
v
ersity Research Resear
ch
Gran
t of Un
i
v
ersi
ti Tek
n
o
l
og
i
Malaysia G
U
P
Tier
1
w
ith
Vo
te No
. 05H
61 of
Min
i
str
y
o
f
H
i
gh
er Edu
cati
o
n (
M
O
H
E
)
year
2
014
to 2016
.
REFERE
NC
ES
[1]
B. Li and M. Q. H. Meng,
“Wir
e
less Capsule Endoscop
y
Imag
es Enhanc
ement
Via Adaptiv
e C
ontrast Diffusio
n
,”
Journal of Visua
l
Communicati
o
n
and. Image
Representation
,
vol/issue: 23(1)
, pp
. 222–228, 2012.
[2]
R. Shahril,
et al.
, “Anisotropic Contrast Diffusion Enhancement
Using Variance
For Wireless Ca
psule Endosco
p
y
Im
ages
,”
In
terna
tional Con
f
eren
ce on
in In
formatics, Electronics
&
Vi
sion (
I
CIEV)
,
pp. 1-6
,
2014
.
[3]
Y. G. Lee and
G. Yoon, “Bleeding De
tection
Algorithm For
Capsule Endoscop
y
,”
World Academy of Science
Engineering and
Technolog
y,
vol/issue: 5(9)
, pp
.
544-549, 2011
.
[4]
L. Cui,
et al.
, “Bleeding Detection In Wirele
ss Capsule Endoscop
y
Images B
y
Support Vector Classifier
,”
IEEE
International Co
nference on
Info
rmation and Au
tomation
, pp
. 174
6–1751, 2010
.
[5]
P.
Sz
c
z
y
p
i
ń
ski,
et al.
, “
T
extur
e
And Color Bas
e
d Im
age S
e
gm
enta
tion And Patholog
y
Detection In Capsule
Endoscop
y
Videos,”
Computer
Methods and
Programs in Biom
edicine
, vol/issue: 113(1)
, pp
.
3
96-411, 2014
.
[6]
T. Ghosh,
et a
l
.
, “An Auto
matic
Bleeding D
e
tection Scheme In
Wireless Capsule
Endoscop
y
B
a
sed On Histogram
Of An RGB-In
dexed Image,”
36t
h A
nnual
Inte
rnat
i
onal
Confe
r
e
n
e
of
t
h
e
IEE
E
E
ngi
ne
e
r
i
n
g i
n
Me
dic
i
ne
and
Biology Society
(
E
MBC)
, pp. 46
83-4686, 2014
.
[7]
T. Ghosh,
et al.
, “Statistical Feature Based No
vel Method To
De
tect Bleeding
In Wire
less Capsule Endoscop
y
Im
ages
,”
In
terna
tional Con
f
eren
ce on In
fo
rmatics, Electronics
&
V
i
sion
, pp
. 1-4
,
20
14.
[8]
G. Pan,
et al.
,
“
B
leeding De
te
ction In
W
i
rel
e
ss Capsule End
o
scop
y
B
a
sed
On Probabilisti
c
Neural N
e
twor
k.
Journal of Med
i
cal S
y
stems
, vo
l/issue: 35(6), pp.
1477–1484, 201
1.
[9]
I. N. Figu
eiredo
,
et a
l
.
, “Computer-Assisted Bleeding Detection I
n
Wireless Caps
ule Endoscop
y
I
m
ages,”
Journal
Computer Meth
ods in Biomech
anics and Biomed
ica
l
Engineering: Im
aging &
Visualization
, v
o
l/is
s
u
e: 1(4), p
p
.
198-210, 2013
.
[10]
S.
Sa
inju,
et a
l
.
, “Bleeding Detection In Wir
e
les
s
Capsule Endos
cop
y
Based On
Color Featur
es From Histogram
Probabilit
y,
”
26th IEEE Canadia
n
Conferen
ce of
El
ectrica
l
and
C
o
mputer Engin
e
ering,
pp
. 1-4
,
2
013.
[11]
B. Juntao, “An
Improved Medical
DR Image Enhancement
Method,”
TELK
OMNIKA Indonesian Journal o
f
Ele
c
trica
l
Eng
i
n
eering
, vol/issue: 12(4), pp. 2718
-2723, 2014
.
[12]
M. Sheng,
et al.
, “
U
nderwat
e
r
Im
ages Enha
ncem
ent using
Mu
lti-W
a
vel
e
t
Transform
and Median Filt
er,
”
TELKOMNIKA Indonesian Journ
a
l
of Electrical
Engineering
, vol/issue: 12(3)
, pp
. 2306-2313, 201
4.
[13]
P
.
P
e
rona, and J
.
M
a
lik, “
S
cal
e-
S
p
ace And Edge Detec
tion Us
ing Anis
otropic Diffus
i
on,”
IEEE Transactions
on
PAMI 12
, pp
. 62
9–639, 1990
.
[14]
G.
Gerig,
et al.
,
“
N
onlinear Anis
otropic
Filte
ring
of MRI Data
,”
I
EEE Transactions on Medical I
m
aging
, vol/issue:
1(2), 1992
.
[15]
R. A. Carmona and S. F. Z
hong, “Adaptive smo
o
thing resp
ectin
g feature dir
ections,”
IEEE T
r
ansactions on Image
Proc
e
ssing
, vo
l/issue: 7(3), pp. 3
53–358, 1998
.
[16]
R. Shahril,
et a
l
.
, “Pre-Processing Technique Based
On Disc
re
te
Cosine
Tra
n
sform
(DCT) and Anis
otropic Con
t
rast
Diffusion for Wireless Capsul
e Endoscop
y
Images”,
IEEE Con
f
er
ence on
in Biomedica
l
Engin
eering and Sciences
(
I
ECBES,
pp
. 92
2-927, 2014
.
[17]
N.
Ahmed,
et a
l
.
, “
D
is
cret
e Cos
i
ne Tr
ans
f
orm
,
”
I
EEE Transactio
ns on Computers
, pp
. 90-93
, 197
3.
[18]
A. K. Jain, “Fundamentals of
dig
ital image processing,”
Pr
en
ti
ce Hall
, 1989
.
Evaluation Warning : The document was created with Spire.PDF for Python.
I
S
SN
:
2
088
-87
08
I
J
ECE
Vo
l. 6
,
N
o
. 4
,
Au
gu
st 2
016
:
16
17
–
1
626
1
626
[19]
A. Beghdadi
an
d L. Negrate, “Cont
rast enh
a
ncement technique base
d on local
detection of
edg
e
s,”
Comput. Vis.
Graphics Image
Process
, vol. 46
, pp. 162–174, 19
89.
[20]
T. Celik and T
.
Tjahjad
i
“
A
utom
atic Im
age Equali
zat
ion and Contrast Enhan
cemen
t Using Gaussian Mixture
Modeling,”
I
E
EE Transactions
on Image
Proces
sing
, vol/issue: 2
1
(1), 2012
.
[21]
K. Panetta,
et al.
, “Nonlinear Unsharp Masking
for M
a
m
m
ogram
Enhancem
ent,
”
IEEE Transactions on
Information Technology on
Bio
m
edicin
e
, vol/issue: 15(6)
, 2011
.
BIOGRAP
HI
ES OF
AUTH
ORS
Ros
d
iana S
h
ahri
l has
rece
ived he
r Bachelor
in
Computer Engineer
ing from
F
acult
y of Elec
tric
al
Engineering, U
n
iversiti Teknol
ogi Malay
s
ia
in 2008. She h
a
s completed h
e
r Masters
in
Mathem
ati
c
s (Num
erical m
e
thod
s and m
u
ltigrid)
from
Facult
y
of Science, Univer
siti Teknolog
i
M
a
la
y
s
ia in 201
2. S
h
e worked a
s
a Res
earch As
s
i
s
t
ant and Anal
ys
t Deve
loper f
o
r four
y
e
ars
.
Her curren
t
r
e
s
earch
inter
e
res
t
inc
l
ude
m
a
th
em
atic
al m
e
tho
d
s
,
ca
lculus
,
m
e
dical
im
age
processing and
n
e
ural network
.
Sabariah Bah
a
ru
n receiv
e
d her Bache
l
or in Mathem
ati
c
s in Au
gust 1982 from Indiana State
Univers
i
t
y
, Indi
ana, US
A. S
h
e has
com
p
leted h
e
r M
a
s
t
ers
in M
a
them
at
ics
in 19
83 from
Ohio
Universit
y
, Ohi
o
, USA. She has obtained h
e
r
PhD in Applied Mathem
atics
from
Universiti
Teknologi M
a
lay
s
ia, 2005.
Ear
l
ier, Dr. Sab
a
riah
work
ed as a Sen
i
or Lecturer
in the Department
of Mathem
atics
under Facult
y
of
Science at Univ
ersiti T
e
knologi
Malay
s
ia. She has also served
as the Deput
y
Dean (Academ
i
c
)
in Malay
s
ia-
J
apan Intern
atio
nal Institut
e
of
Technolog
y
,
Universiti T
e
kn
ologi Mala
ysi
a
for over two
y
e
ar
s. Currentl
y
,
Dr. Sabariah is
the Associate
Proffesor at Mala
ysi
a
-Japan I
n
terna
tiona
l In
stitut
e
of Te
ch
nolog
y, Univer
siti Tekno
logi
M
a
la
y
s
ia
. Her res
earch ar
ea in
Graph and F
u
zzy Graph M
odeli
ng, M
a
them
ati
c
a
l
Thinking, and
wireless com
m
unica
tion.
She h
a
s over 50
Int’
l Pu
blic
ation
.
A.
K.
M.
M
u
z
a
h
i
dul Islam
has
rece
ived M
.
S
c
.
in Com
puter S
c
ien
ce
and Eng
i
neer
ing from
Kharkiv Nation
a
l University
of
Radio Electron
i
cs, Ukrain
e, in
1999 and D.Eng
.
in Computer
Science and Eng
i
neer
ing from
Nago
y
a
Institu
te o
f
Technolog
y
,
Japan in 2007.
Dr
. Muzahid
h
a
s
recieved
Japanes
e
Government
Monbusho Sc
holarship (October
2002 - Mar
c
h
2006). He h
a
s
worked in various
industries in Bangladesh and
currently
serving as a Sen
i
or Lecturer
at
Malay
s
ia-Jap
an International
In
s
titut
e
of
Technol
og
y
(MJIIT)
of
Universiti
Tekn
ologi Mal
a
y
s
ia
(UTM
), Kuala
Lum
pur. His
res
earch int
e
res
t
s
includ
e Network Achitecture, Communication
Protocol, Cognitive Radio Networ
k, Wireless Sensor Networ
k,
and Network Security
. Dr.
Muzahid has pu
blished over 60
intern
ation
a
l r
e
s
earch pub
li
cat
io
ns
. He has
s
e
rv
ed in th
e 7TH
AUN/SEED-Net 2014 Int’l Conf
erence on
EEE,
ICBAPS2015, ICIEV15 and IC
AEE 2015 Int’
l
Conferences. Cu
rrently
is servin
g as th
e Progr
am Chair of ICAI
CT 2016 and
Secretariat for
the
ICaTAS 2016.
Dr. Muzahid
is
a Senior I
E
E
E
Me
m
b
er (SMIEEE) and
is an
I
ET Mem
b
er wit
h
Designator
y
Le
tt
ers
(MIET)
.
Evaluation Warning : The document was created with Spire.PDF for Python.