TELKOM
NIKA
, Vol.11, No
.1, Janua
ry 2013, pp. 19
~27
ISSN: 2302-4
046
19
Re
cei
v
ed Se
ptem
ber 27, 2012; Revi
se
d No
vem
ber
19, 2012; Accepted Novem
ber 27, 20
12
Digital Medical Image Enhanced by wavelet
Illumination-Reflection Model
Xiong Jie
Schoo
l of Com
puter Scie
nce
& Engin
eer
i
ng,
XI’AN T
e
chnol
ogic
a
l Un
iversit
y
xi
’an cit
y
sh
aa
nxi prov
ince C
h
in
a, telep
hon
e:152
29
084
30
7
e-mail: xio
n
g
jie
69@
126.com
A
b
st
r
a
ct
W
hen a di
git
a
l medic
a
l i
m
age is e
nha
n
c
ed, t
he usef
ul deta
ils of the imag
e sh
oul
d be
strengthe
ne
d
，
but th
e
det
ails c
a
n
not
b
e
stren
g
the
n
e
d
by
thes
e a
l
gorit
hms
b
a
s
ed
on tra
d
iti
o
nal
illu
min
a
tion-r
e
fl
ection
mo
del.
Accordin
g to the ima
g
i
n
g
princip
l
e a
n
d
med
i
cal re
qu
ire
m
ent, w
a
ve
let
illu
min
a
tion-r
e
fl
ection mod
e
l and a
n
e
w
al
gorith
m
bas
ed
on th
e
mo
de
l are
pro
pose
d
. The i
m
age
i
s
deco
m
pose
d
i
n
to ill
u
m
in
atio
n and r
e
flectio
n
by w
a
vele
t il
lu
min
a
tio
n
-refl
ection
mo
de
l. T
he deta
ils of
th
e
reflectio
n
are
strengthe
ne
d. T
he
dyn
a
m
ic r
ang
e of the i
l
l
u
min
a
tion
is reduc
ed i
n
ord
e
r to en
hanc
e
th
e
imag
e. Experi
m
e
n
ts and
an
alysis sh
ow
that the met
hod
is obvi
ously
be
tter than Histo
gra
m
Equ
a
li
z
a
t
i
on,
Ho
mo
mor
p
h
i
c
F
iltering and multi-scal
e
s
Reti
nex.
Key
w
ords
:
d
i
gital m
edical
im
age, illumination-refle
c
t
i
on m
odel, st
ationary
wa
velet tran
sform,
hom
om
orphic filtering, m
u
lti-scale
s
Retin
e
x
Copy
right
©
2013 Un
ive
r
sita
s Ah
mad
Dah
l
an
. All rig
h
t
s r
ese
rved
.
1. Introduc
tion
The The ima
ges
with poor contra
st and
low brightn
e
ss b
r
ing g
r
ea
t obstacl
es to
doctor’
diagn
osi
s
. F
o
r such ima
ges, do
cu
me
nt [1] has consi
dered th
at three mai
n
method
s
are
Histo
g
ra
m E
quali
z
ation,
Homo
morphi
c Filte
r
ing
an
d multi-scale
s
Retinex,
which
are u
s
e
d
to
enha
nce the
s
e imag
es.
Homo
morphi
c Filterin
g
a
nd multi-scal
es Retinex are ba
se
d on
traditional illu
mination
-refle
c
tion mo
del. Becau
s
e
of
the limitation
s
of Homom
o
rphic Filte
r
ing
and
multi-scal
es
Retinex b
a
sed on
the i
m
aging
pr
i
n
ciple, the d
e
tails of th
e i
m
age
s
can’t
b
e
stren
g
then
ed
by them. T
herefo
r
e,
we
sho
u
ld p
r
o
pose an
effective digital
medical im
age
enha
ncement
method. These image
s enhan
ced
by the method have sui
t
able co
ntra
st,
brightn
e
ss
an
d dynami
c
ra
nge. Be
side
s, the detai
l
s
of these ima
ges en
han
ce
d by the
met
hod
can b
e
stre
ng
thened.
In this
re
gard
,
acco
rding
t
o
the
digital
medi
cal im
ag
e featu
r
e
s
a
n
d
the
sh
ort
c
o
m
ings of
traditional
illu
mination
-refle
c
tion m
odel,
a digital
medi
cal im
age
en
han
ceme
nt m
e
thod
ba
sed
on
wavelet illumi
nation-refle
c
ti
on mod
e
l is
prop
osed.
In
the method, t
he imag
es
are explaine
d
b
y
wavelet ill
u
m
ination
-refle
c
tion
mod
e
l, the im
age
s are d
e
com
posed i
n
to i
llumination
a
nd
reflectio
n
by
stationa
ry wa
velet tran
sform, t
he detail
s
of
refle
c
tio
n
are
stren
g
thene
d an
d t
he
dynamic
range of illumination is compressed.
2. Image Enhancement
Method based on Tr
aditional Illuminati
on-
reflection Model
Explaining P
h
ysical thou
gh: Whe
n
a
n
image
)
,
(
y
x
f
is gene
rated,
its values i
s
prop
ortio
nal to the radi
atio
n ene
rgy of the phy
si
cal
source
s an
d the ra
diation energy
must be
non-ze
ro a
n
d
limitation. It is explain
ed b
y
Eq(1).
)
,
(
)
,
(
)
,
(
0
0
y
x
r
y
x
i
y
x
f
(1)
Whe
r
e :
)
,
(
0
y
x
i
---illu
mination de
ci
de by
the physical so
urce
s and
)
,
(
0
0
y
x
i
.
)
,
(
0
y
x
r
---re
f
lection de
cid
e
by the imaged obje
c
ts a
n
d
1
)
,
(
0
0
y
x
r
.
Evaluation Warning : The document was created with Spire.PDF for Python.
ISSN: 23
02-4
046
TELKOM
NIKA
Vol. 11, No
. 1, Janua
ry 2013: 19 – 27
20
Figure 1. Tra
d
itional illumi
nation-refle
c
ti
on model
Traditional ill
umination-reflection model
thin
ks: The values of illumi
nation change
slowly
in the
spatial
domai
n. On
the co
ntra
ry, dram
at
ic
cha
nge
s of the i
m
age val
u
e
s
are
de
cide
d
by
reflectio
n
, especi
a
lly the e
dge
s of obje
c
ts in t
he ima
ge. In the me
antime, the low fre
que
ncy
of
image
ha
s
so
mething to
d
o
with ill
umin
ation an
d the
high f
r
eq
uen
cy ha
s
som
e
thing to
do
wi
th
reflectio
n
i
n
freque
ncy
do
main. Th
e
reflection
is
deci
ded
by t
he
refle
c
tivity of the
obj
e
c
ts
surfaces in scene. The details
are decided by the reflection.
T
he illumination is deci
ded by
ambient light.
The dynami
c
range i
s
de
ci
ded by the illumination.
Homo
morphi
c Filte
r
ing i
s
a meth
od
whi
c
h e
nha
n
c
e
s
ima
g
e
s
i
n
freq
uen
cy
domain.
(
F
ig
ur
e
2)
Figure2
Hom
o
morphi
c Filtering
The key of Homomo
rphi
c
Filtering is th
at the multiplication b
e
twe
en the illumin
a
tion and
the reflectio
n
of images i
s
become into the additio
n
b
e
twee
n them.
)
,
(
ln
)
,
(
ln
))
,
(
)
,
(
ln(
)
,
(
ln
)
,
(
y
x
r
y
x
i
y
x
r
y
x
i
y
x
f
y
x
z
o
o
o
o
(2)
It thinks that the high dyna
mic ra
nge
of image
s is cau
s
ed by illumin
a
tion
)
,
(
y
x
i
o
. The
image
s are filtered by high
-pass freq
uen
cy filter
)
,
(
v
u
H
in order to enh
an
ce these ima
g
e
s.
Eq (2) i
s
tran
sform
ed by F
FT2.
))
,
(
(ln
))
,
(
(ln
))
,
(
(
y
x
r
F
y
x
i
F
y
x
z
F
o
o
(3)
)
,
(
)
,
(
)
,
(
v
u
F
v
u
F
v
u
Z
o
o
r
i
(4)
)
,
(
)
,
(
)
,
(
)
,
(
)
,
(
)
,
(
)
,
(
v
u
F
v
u
H
v
u
F
v
u
H
v
u
Z
v
u
H
v
u
S
o
o
r
i
(5)
The impa
ct of
)
,
(
y
x
F
o
i
is eliminated in Eq (5).
)
,
(
v
u
S
is inverse tra
n
sformed by IFFT2.
Evaluation Warning : The document was created with Spire.PDF for Python.
TELKOM
NIKA
ISSN:
2302-4
046
Digital Medi
cal Im
age Enhanced b
y
wa
velet Illu
m
i
nation-Refle
c
tion
Model (Xion
g
Jie)
21
))
,
(
)
,
(
(
))
,
(
)
,
(
(
))
,
(
(
)
,
(
1
1
1
v
u
F
v
u
H
F
v
u
F
v
u
H
F
v
u
S
F
y
x
s
o
o
r
i
(6)
Suppo
se:
))
,
(
)
,
(
(
)
,
(
1
'
v
u
F
v
u
H
F
y
x
i
o
i
))
,
(
)
,
(
(
)
,
(
1
'
v
u
F
v
u
H
F
y
x
r
o
r
We can get
)
,
(
)
,
(
)
,
(
'
'
y
x
r
y
x
i
y
x
s
(7)
)
,
(
)
,
(
)
,
(
'
'
)
,
(
)
,
(
)
,
(
'
'
y
x
r
y
x
i
e
e
e
y
x
g
o
o
y
x
r
y
x
i
y
x
s
(8)
In document
[2], image
s a
r
e d
e
co
mposed into
illumination
and
reflect
i
on by
Homo
morphi
c Filterin
g. Experime
n
ts s
how that: If the cut
-
off fre
quen
cy of
)
,
(
v
u
H
is highe
r,
the dynamic
rang
e co
mpression an
d the details
lo
ss a
r
e hig
h
e
r
. If the cut-off frequen
cy of
)
,
(
v
u
H
is lo
wer,
the
dynami
c
ra
n
ge
com
p
re
ssi
on a
n
d
the
d
e
tails l
o
ss are lo
we
r. Th
e
details
of images e
n
han
ced by Homomo
rphi
c
Filtering a
r
e d
a
mage
d.
Retinex i
s
b
a
se
d on ill
u
m
ination
-refle
c
tion mo
del,
too. It
[3]
thinks that: All details in
scene
a
r
e i
n
clud
e in
refle
c
tion
)
,
(
y
x
r
o
. Illumination
)
,
(
y
x
i
o
co
nsi
s
t
s
of
all li
ght
sou
r
ces in
scene.
The hi
gh dynamic range of im
ages i
s
de
cided by illumination.If illuminat
ion i
s
separat
ed
from images,
the influence
of
illumination can effectiv
ely be e
liminated and the
images dynamic
rang
e ca
n be
comp
re
ssed.
Figure 3 Reti
nex
K
k
k
k
y
x
f
y
x
F
y
x
f
W
y
x
g
1
)))
,
(
)
,
(
log(
)
,
(
(log
exp(
)
,
(
(9)
Whe
r
e:
k
--
-s
c
a
le,
)
,
(
y
x
F
k
---the
surrou
nd fun
c
tion in scal
e
k
,
k
W
---the
weightin
g co
efficients
corresp
ond to
)
,
(
y
x
F
k
.
Evaluation Warning : The document was created with Spire.PDF for Python.
ISSN: 23
02-4
046
TELKOM
NIKA
Vol. 11, No
. 1, Janua
ry 2013: 19 – 27
22
In Eq (9), it is difficult that i
llumination
)
,
(
y
x
i
o
and reflectio
n
)
,
(
y
x
r
o
are
sepa
rate
d
in
)
,
(
y
x
f
.
In do
cume
nt
[4], digital me
dical i
m
ag
es
are
enh
an
ce
d by multi-scales
Retin
e
x. It thinks
that: The illumination
)
,
(
0
y
x
i
is equivalent to
X-ray intensi
t
y through the human b
o
d
y
to th
e
imaging
device and d
e
ci
de
s the ima
g
e
s
dynamic
ran
g
e
. The reflecti
on
)
,
(
0
y
x
r
is equival
ent to
the cha
nge
o
f
X-ray inten
s
ity abso
r
b
e
d
by t
he hum
an body a
n
d
represents t
he detail
s
of the
imaging of th
e human b
o
d
y
.
3. Digital Medical Image Enhancement ba
sed on
Wav
e
let Illu
mination-refl
ection Model
3.1. Wav
e
let Illumina
tion-reflection Model
The key di
sa
dvantage of tradition
al illuminatio
n
-refle
c
tion mod
e
l is that the illumination
and th
e
refle
c
tion
of the
i
m
age
can
no
t be
effect
ivel
y sep
a
rated.
Becau
s
e
of t
he di
sa
dvant
age,
the details
of image
s en
ha
nce
d
mu
st be
lost
wh
en im
age
s are enh
anced by the
model. Fo
r th
is
rea
s
on,
wav
e
let illumin
a
tion-refle
c
tion
model i
s
p
r
o
posed. In th
e ne
w m
ode
l, image
s a
r
e
decomposed
into high f
r
eq
uency part
and low f
r
equen
cy part. In other
word
s, the illumination is
repla
c
e
d
by the low fre
que
ncy part an
d the refle
c
tion i
s
repl
aced by
the high freq
uen
cy part.
An image
)
,
(
y
x
f
is decompo
se
d by 2D multi-scale d
e
comp
osition.
)
,
(
)
,
(
)
,
(
)
,
(
)
,
(
)
,
(
1
2
1
y
x
W
y
x
W
y
x
W
y
x
W
y
x
V
y
x
f
j
j
j
(10
)
)
,
(
)
,
(
)
,
(
1
1
y
x
W
y
x
V
y
x
V
j
j
j
(11
)
)
,
(
)
,
(
)
,
(
1
1
y
x
V
y
x
V
y
x
W
j
j
j
(12
)
)
,
(
)
,
(
)
,
(
)
,
(
))
,
(
)
,
(
(
)
,
(
1
1
1
1
1
y
x
W
y
x
W
y
x
V
y
x
W
y
x
W
y
x
V
y
x
V
j
j
j
j
j
j
j
(13
)
s
u
pp
os
e
：
)
,
(
)
,
(
)
,
(
)
,
(
1
1
y
x
W
y
x
W
y
x
W
y
x
W
j
j
(14
)
We can get
)
,
(
)
,
(
)
,
(
y
x
W
y
x
V
y
x
f
j
(15
)
Bec
a
us
e
)
,
(
y
x
V
j
is the low fre
que
ncy part of
)
,
(
y
x
f
,
)
,
(
)
,
(
0
y
x
i
y
x
V
j
.
Bec
a
us
e
)
,
(
y
x
W
is the high fre
q
u
ency pa
rt of
)
,
(
y
x
f
,
)
,
(
)
,
(
0
y
x
r
y
x
W
.
Figure 4 is an
example that the illumination
and refle
c
t
i
on of a gray image are se
p
a
rated
by wavelet illumination-reflection model.
In su
mma
ry, tradition
al i
llumination
-
re
flection
mod
e
l can
be
repla
c
ed
by
wavelet
illumination-reflection model.
Evaluation Warning : The document was created with Spire.PDF for Python.
TELKOM
NIKA
ISSN:
2302-4
046
Digital Medi
cal Im
age Enhanced b
y
wa
velet Illu
m
i
nation-Refle
c
tion
Model (Xion
g
Jie)
23
Figure 4. a is an origi
nal im
age. Figu
re 4
.
b is
the illumination of Figu
re 4.a. figure
4.c is
the reflection of Figure 4.a.
We can get the fact
s: it is easy that illumi
nation and reflection are
sep
a
rate
d by wavelet illumi
nation-refle
c
ti
on model.
3.2. Digital Medical Image
Enhanced b
y
Wa
v
e
let Ill
umination-reflection Model
The advantage of
wavelet
illumination-reflec
tion
model is that the
illumination
and the
reflectio
n
of i
m
age
s
ca
n b
e
effectively
sep
a
rate
d. T
he dyn
a
mic range
of im
ag
es i
s
de
cide
d
by
the illumination
whi
c
h i
s
t
he low f
r
equency
part
of i
m
ages decomposed
by
wavelet transform.
The detail
s
o
f
images
are
deci
ded by th
e refle
c
tion
which i
s
the hi
gh freq
uen
cy
part of imag
es
decomposed by wavelet transfor
m. For this reason,
wavelet
illumination-reflection model can
synchronousl
y do both the
com
p
ression of the high dynamic
range deci
ded by the illumination
and th
e
stre
ngtheni
ng
of the u
s
eful
d
e
tails
de
cide
d by the
reflection. It i
s
benefi
c
ial to
the
requi
rem
ent
that the useful det
ails sho
u
ld
be
stren
g
then
e
d
in digital
medical i
m
age
enha
ncement
.
Figure 5. Images e
nha
nce
d
by wavelet illumination
-
re
flection mod
e
l
Whe
n
a
si
gn
al is de
com
p
ose
d
by
stati
onary
wa
vel
e
t tran
sform, t
he
sign
al le
n
g
th is not
cha
nge
d. It is benefi
c
ial to
the de
com
p
o
s
ition
a
nd
co
mpositio
n of i
m
age
s. Wavelet illuminati
on-
reflectio
n
mo
del ca
n be e
a
sily expressed by stat
ion
a
ry wavel
e
t tran
sform. T
h
e wavelet
scales
and fun
c
tion
are d
e
termin
ed in acco
rd
a
n
ce
with the
experim
ent a
nd the actu
al
situation. In this
pape
r, the wa
velet scal
e
s i
s
three
and th
e wavelet fun
c
tion is ‘
s
ym4
’
.
3.3. Lo
w
-
pa
s
s
Filter
Do
cume
nt [5] thinks that th
e high
dynam
ic r
ang
e of im
age
s is de
cid
ed by the
en
ergy in
every frequ
e
n
cy ban
d of the illuminatio
n. We c
an attenuate the e
nergy in eve
r
y frequen
cy b
and
of the illumination in order to compress the hi
gh dynamic range of image
s. First, the illumination
is tran
sfo
r
me
d by FFT2. S
e
co
nd, the ill
umination
tra
n
sformed i
s
filtered by G
a
ussian lo
w-p
a
ss
filter. At last, the illumination is rest
ructured by IFFT2. See Eq( 16, 17 ,18).
)
)
,
(
(
2
0
2
)
,
(
D
v
u
D
c
e
rH
v
u
H
(16
)
Fi
gure
4.
a
Fi
gure 4.
b
Fi
gure 4.c
S
t
a
t
i
ona
ry
wa
ve
le
t
de
c
o
m
positi
on
illum
i
na
tio
n
re
fle
c
tion
L
o
w-pa
ss fi
lte
r
Bay
e
s
sof
t
-t
hre
s
h
o
l
d
e
s
tim
a
tion
De
ta
ils
s
t
r
e
ng
t
h
en
ed
S
t
a
t
i
ona
ry
wav
e
let
c
o
m
positio
n
Br
i
g
ht
n
e
s
s
ad
j
u
s
t
m
e
n
t
Evaluation Warning : The document was created with Spire.PDF for Python.
ISSN: 23
02-4
046
TELKOM
NIKA
Vol. 11, No
. 1, Janua
ry 2013: 19 – 27
24
2
2
)
2
(
)
2
(
)
,
(
N
v
M
u
v
u
D
(17
)
Whe
r
e
M
,
N
is
the si
ze
of image,
C is a
con
s
tant
whi
c
h d
e
termi
n
e
s
the i
n
clin
ed
plane of filter,
0
D
is cut-off fre
quen
cy,
r
H
is ra
tio coefficie
n
t and
7
.
0
r
H
in this paper.
)))
,
(
(
(
5
.
0
0
v
u
D
median
median
D
(18
)
3.4. Ba
y
es Soft-th
resh
old Estimation
and De
tails Streng
theni
ng
Images
cont
ain noi
se or unwa
n
ted d
e
tails in varying degree
s. The corre
s
pondi
ng
threshold
s
sh
ould
be
set f
o
r the
reflecti
on. The
detai
ls a
bove the
corre
s
p
ondin
g
thre
sh
old
s
are
stren
g
then
ed.
The detail
s
unde
r the correspon
di
ng
threshold
s
are atten
uate
d
. Becau
s
e t
he
wavelet
coef
ficients in
reflection
obe
y gene
ral
G
aussia
n
di
stribution, the
threshold
s
a
r
e
determi
ned b
y
Bayes soft-thre
shol
d
[8]
. See Eq(1
9 ,20,21 ).
x
n
r
Thr
/
2
(19
)
Whe
r
e :
Th
r
--- the
cor
r
e
s
pondi
ng thre
shold;
r
---coeffi
cient,
2
r
;
n
---noise stan
dard d
e
viatio
n;
x
---sign
al stan
dard d
e
viatio
n.
6745
.
0
/
)
)
,
(
(
j
i
y
median
n
n
(20
)
Whe
r
e:
)
,
(
j
i
y
is
the wavelet c
oeffic
i
ent of the reflec
tion.
2
)
var(
n
x
y
(21
)
Whe
r
e:
)
var(
y
---wavelet coeffic
i
ents
mat
r
ix varianc
e
.
Acco
rdi
ng to
the co
rrespo
nding th
re
sh
old, t
he
wave
let coeffici
ent
s of the
refle
c
tion a
r
e
stren
g
then
ed
or attenuate
d
.
See Eq (22 ).
Thr
j
i
y
j
i
y
Thr
j
i
y
j
i
y
j
i
y
)
,
(
,
2
/
)
,
(
)
,
(
,
2
)
,
(
)
,
(
ˆ
(22
)
Whe
r
e
)
,
(
ˆ
j
i
y
is the wavelet co
efficient of refle
c
tion st
rength
ened o
r
atten
uated.
3.5. Brightn
e
ss Adjus
t
me
nt
Becau
s
e the
energy of the illuminatio
n is
attenuat
ed by low-pa
ss filter fo
r the hig
h
dynamic ra
ng
e comp
resse
d
, the ima
g
e
s
re
stru
ctur
ed
by stationa
ry
wavelet tran
sform a
r
e
darker
than befo
r
e. The imag
es
brightn
e
ss
sh
ould be a
d
ju
sted by Gam
m
a adju
s
tme
n
t in orde
r to
the
human eye
s
comfo
r
table
o
b
se
rve these image
s.
Evaluation Warning : The document was created with Spire.PDF for Python.
TELKOM
NIKA
ISSN:
2302-4
046
Digital Medi
cal Im
age Enhanced b
y
wa
velet Illu
m
i
nation-Refle
c
tion
Model (Xion
g
Jie)
25
t
y
x
g
y
x
g
)
,
(
)
,
(
(23)
Whe
r
e
)
,
(
y
x
g
is th
e imag
e
re
structured
by
st
ationary wav
e
let
tran
sfo
r
m,
t
is the
Gamma coeff
i
cient,
7
.
0
t
in this pape
r.
3.6. Experiments
、
Subje
c
tiv
e
and Objectiv
e Ev
alu
a
tion
The co
mpa
r
a
t
ive experime
n
t is that a digi
tal medical image is en
h
anced by Histogram
Equalization
in Photosh
o
p
, Homom
o
rphic Filt
e
r
ing
,
multi-scale
s
Retin
e
x and the meth
od
prop
osed in this pa
per.
Figure 6.a is
an origi
nal im
age, Figu
re 6
.
b
is enha
nce
d
by Histog
ra
m Equalizatio
n in
Photosh
op, F
i
gure 6.
c is e
nhan
ce
d by Homo
morphi
c Filterin
g, Figure 6.d i
s
en
han
ced by m
u
lti-
scale
s
Retin
e
x
(The scale
s
are 80,15
0,2
50.)
and Fi
gu
re 6.e is en
ha
nce
d
by the method
prop
osed in this pa
per.
Acco
rdi
ng to
t
he
subj
ective
judgme
n
t, the
effect
of Hi
st
ogra
m
e
quali
z
ation
en
han
ceme
nt
is
the wo
rst. The
ima
ge co
ntrast of
multi-scale
s
Retin
e
x enha
ncem
ent is the
be
st, but the loss of
the image
de
tails is
more
and ma
ny of
details
ca
n n
o
t be ob
se
rv
ed by hu
man
eyes. Th
e im
age
whi
c
h i
s
en
h
anced by
Ho
momorphi
c F
iltering a
nd t
he metho
d
p
r
opo
sed i
n
thi
s
pa
per
co
ntains
the most d
e
tails. The
mo
st abu
nda
nt details
ca
n b
e
ob
serve
d
by human
e
y
es. The im
age
contrast
enh
anced by th
e
method
pro
posed in
th
is pape
r i
s
mu
ch b
e
tter tha
n
Ho
momo
rp
hic
Filtering.
Le
si
on of
tumo
rs and
hyp
e
rpl
a
sia
ca
n
be
easily
ob
se
rved in
the
ima
ge e
nha
nced
by
the metho
d
prop
osed i
n
this p
ape
r. F
r
om
huma
n
eyes
ob
servi
ng, the
effect of the met
hod
prop
osed in this pa
per i
s
the be
st.
The chan
ge
s in image b
r
ig
htness
an
d contra
st are
a
nalyze
d
by the method in d
o
cum
ent
[6] and Eq ( 24, 25).
Figure 6.a
Figure 6.b
Figure 6.c
Figure 6.d
Figure 6.e
Evaluation Warning : The document was created with Spire.PDF for Python.
ISSN: 23
02-4
046
TELKOM
NIKA
Vol. 11, No
. 1, Janua
ry 2013: 19 – 27
26
)
(
)
(
)
(
f
Var
f
Var
g
Var
C
(24
)
)
(
)
(
)
(
f
Mean
f
Mean
g
Mean
L
(25
)
Whe
r
e C i
s
th
e rate of co
ntrast
chan
ge,
L is the rate o
f
brightne
ss
chang
e.
The inform
ation entro
pie
s
of Figure 6 a
r
e analyzed b
y
the method in document [7].
n
i
i
i
p
p
E
1
lg
(26
)
Whe
r
e E i
s
t
he ima
ge info
rmation
entro
py,
i
p
is the
nu
mber of the i
m
age
pixels
whe
n
gray value is i
,
n is the image gray level.
Table1 Perf
orma
nce Parameter
Figure6.a
Figure6.b
Figure6.c
Figure6.d
Figure6.e
Brightness(L)
0
0.7489
7.7680
6.2934
4.2593
Contrast(
C
)
0
0.7341
3.3802
1.2883
2.6276
Entrop
y
(E)
3.3914
3.7435
2.5110
3.4987
4.8268
As we
can
se
e from th
e Ta
ble 1, the i
n
fo
rm
ation
entro
py of the ima
ge en
han
ce
d
by the
method in thi
s
pape
r is ma
ximum and the contrast of
Figure 6.e is better than figure 6.d. Fig
u
re
6.e brig
htne
ss is m
ode
rat
e
and
suitabl
e observe
d b
y
human eye
s
. According
to the evalua
tion
method in d
o
c
ume
n
t [5], we invited ten docto
rs to
ev
aluate from F
i
gure
6.a to Figure
6.e. Th
ey
a
ll co
ns
id
er
ed
th
a
t
the
de
ta
ils
,
the
contrast, the
bri
g
h
t
ness of
figure 6.e i
s
the
b
e
st.
We
ca
n
get
the co
ncl
u
si
o
n
that the q
u
a
lity of image
enha
nced
b
y
the method
pro
p
o
s
ed i
n
this pa
pe
r is
the
best in the
s
e
method
s.
4. Conclusio
n
Wavelet illum
i
nation-refle
c
t
i
on model h
a
s
an
intuitive physical me
aning an
d the strict
mathemati
c
al
sen
s
e. It can
effectively separate t
he ill
umination
an
d refle
c
tion.
This i
s
be
nefi
c
ial
to enh
an
ce
digital me
dical imag
es which
the
high
dynami
c
ra
nge
of the i
m
age
s
sho
u
l
d
be
comp
re
ssed
and the
detai
ls of the i
m
a
ges
sh
ould
b
e
strength
e
n
ed. Experim
e
n
ts an
d a
nal
ysis
sho
w
th
at the
mod
e
l i
s
o
b
v
iously b
e
tter than
traditio
nal illumi
nati
on-refle
c
tion
model.
Ho
we
ver,
the details
of images e
nhan
ce
d by Wavelet
illumination
-
reflection mo
del
have a certain
relations
h
ip
with the
wav
e
let func
tion
us
ed in
it. From now on,
we
s
h
ould
res
e
arch the type of
wavelet function. Moreov
er, the
wav
e
let trans
f
o
r
m theo
ry is co
nsta
ntly evolving. Wa
velet
illumination-reflection m
o
del usi
ng curv
elet tran
sform instead of
wavelet transform m
a
y have
better re
sult
s.
Ackn
o
w
l
e
dg
ements
This wo
rk was
fina
nci
a
lly
sup
porte
d by
the
Prin
cipal Fou
ndati
on of XI’AN Technolo
g
ical
University (XAGDXJJ1
119
).
Referen
ces
[1]
Xi
on
g Jie, Ha
n
Li-na, Geng
g
uo-h
ua a
nd Z
h
ou Min
g
-qu
a
n
.
Digita
l
med
i
cal
ima
ge e
nha
nc
ed by si
mil
a
r
meth
od of Reti
nex
. Comp
uter
Engin
eeri
ng a
nd App
licat
i
ons
. 2009; 45-
24: 14-1
6
(in Ch
in
ese)
Evaluation Warning : The document was created with Spire.PDF for Python.
TELKOM
NIKA
ISSN:
2302-4
046
Digital Medi
cal Im
age Enhanced b
y
wa
velet Illu
m
i
nation-Refle
c
tion
Model (Xion
g
Jie)
27
[2]
Xi
on
g Ji
e, Ha
n Li-
na, Ge
ng
guo-
hu
a a
n
d
Z
hou M
i
ng-
q
uan.
Us
in
g H
SV spac
e re
a
l
-color
i
m
a
g
e
enh
anc
ed by h
o
mo
mor
phic filt
er in tw
o chann
el
. Comp
uter Engi
neer
in
g and
Applic
ations. 2
009; 45-
27:
18-2
0
(in Ch
in
ese)
[3]
Jobso
n
DJ,
R
ahmam
Z
,
W
oode
ll GA.
Diffusio
n
Pro
perti
es a
n
d
p
e
rfor
mance
of
a c
enter/surro
un
d
retinex
.
IEEE Trans on I
m
age
Processing: Sp
ecial Issue on
Color Pr
ocessing
. 199
7; 6-7: 451-
462.
[4]
W
ang, Yan-ch
en, LI. Shu-ji
e, Huan
g. Lia
n
-qi
ng
Enh
anc
ement of radi
ogra
p
h
y
bas
e
d
multiscal
e
Retinex
.
Optics and Precis
ion
Engi
neer
in
g
. 2006; 14-
1: 70-7
7
(in Ch
ines
e)
[5]
Xi
on
g, Jie.
R
e
al co
lor
imag
e
enh
anc
ement
base
d
o
n
freq
u
ency d
o
m
a
i
n,
w
a
velet transfo
rm a
n
d
ne
ura
l
netw
o
rk
.
Northw
est University
doctoral thes
is
. 2010 (in C
h
i
n
ese)
[6]
Jobso
n
DJ, R
ahmam Z
,
W
oode
ll GA.
T
he statistics of visual r
epr
es
e
n
tation Pr
ocee
din
g
s of SPIE
Visual Informa
tion Proc
essin
g
. W
a
shington:
SPIE Press, 2002; 25-3
5
.
[7]
Fu
, Zu
-y
u
n
.
Informatio
n
T
heor
y
, Beijin
g: Elec
tronics Industr
y Press, 2001 (i
n Chi
nese)
[8]
Yan, Jin-
w
e
n.
Digital I
m
a
g
e
Processin
g
Beiji
ng
. Nati
on
a
l
Defens
e Ind
u
str
y
Press, 2
007; 95 (i
n
Chin
ese)
Evaluation Warning : The document was created with Spire.PDF for Python.