TELKOM
NIKA Indonesia
n
Journal of
Electrical En
gineering
Vol. 16, No. 3, Dece
mbe
r
2
015, pp. 539
~ 545
DOI: 10.115
9
1
/telkomni
ka.
v
16i3.937
3
539
Re
cei
v
ed
Jul
y
24, 201
5; Revi
sed O
c
tob
e
r 17, 201
5; Acce
pted No
vem
ber 8, 20
15
A Hybrid the Nonsubsampled Contourlet Transform
and Homomorphic Filtering for Enhancing
Mammograms
Khad
douj Ta
ifi*, Rachid Ahdid, Moha
med Fakir, Said Safi
F
a
cult
y
of Scie
nce an
d T
e
chnolo
g
y
B
eni-M
el
lal, Morocc
o
Lab
orator
y of Informatio
n
Pro
c
essin
g
& Deci
sion Su
pp
ort (T
IAD)
F
a
cult
y
of Scie
nce an
d T
e
chnics,
Universit
y
Sultan
Mo
ul
a
y
Sliman
e
*Corres
p
o
ndi
n
g
author, e-ma
i
l
: taif_kha@
hot
mail.fr
A
b
st
r
a
ct
Mammogra
m
i
s
important fo
r early br
east
cancer
d
e
tec
t
ion. But du
e to the low
co
ntrast of
micr
ocalc
i
ficati
ons an
d nois
e
,
it is difficult to detec
t microc
alcificati
on. T
h
i
s
paper pr
ese
n
ts a comp
arat
ive
study in d
i
gita
l
ma
mmo
g
ra
ph
y ima
ge
enh
a
n
ce
me
nt
base
d
on thre
e diff
erent al
gor
ith
m
s: ho
mo
mor
phic
filterin
g, uns
ha
rp
maski
ng
a
nd
our
prop
os
ed methods. T
h
is
latter us
es
a hybr
id meth
od
Co
mbi
n
in
g
contour
let an
d
ho
mo
morph
i
c filterin
g.
Perfor
ma
nce of the
g
i
ven tec
hni
qu
e has b
een
mea
s
ured i
n
ter
m
s of
distrib
u
tion se
parati
on
meas
ure (DSM), target-t
oback
g
ro
u
nd en
ha
nce
m
e
n
t me
asure b
a
s
ed on stan
da
r
d
deviation (TBES) and ta
rget-to-backgr
o
und enh
anc
ement
m
e
asure
based on entropy (TBEE). The
prop
osed
met
hods w
e
r
e
tes
t
ed w
i
th the r
e
ferent
s
ma
mmo
g
ra
phy
dat
a Base
Min
i
M
I
AS. Experi
m
e
n
tal
results show
that the prop
ose
d
met
hod i
m
pr
oves the visi
bil
i
t
y of microcalc
i
fication.
Ke
y
w
ord:
micr
ocalcific
a
tio
n
, contour
let, enh
a
n
ce
me
nt ho
mo
mor
phic filt
erin
g
Copy
right
©
2015 In
stitu
t
e o
f
Ad
van
ced
En
g
i
n
eerin
g and
Scien
ce. All
rig
h
t
s reser
ve
d
.
1. Introduc
tion
Brea
st ca
nce
r
is th
e mo
st commo
n
cancer i
n
wo
men a
nd ra
nks first in t
he world
contin
ue
s to
be the le
adin
g
ca
use
of d
eath over 40
years [1]. In Moro
cco, b
r
e
a
st can
c
e
r
is
also
the first
wom
an an
d a
c
cording to
data
from t
he
re
gi
ster
of the
G
r
eate
r
Casab
l
anca 20
04 t
he
incid
e
n
c
e
sta
ndardized
on
the worl
d p
opulatio
n is
about
35
ca
ses /
100,00
0
wom
en /yea
r it
rep
r
e
s
ent
s 3
6
% of all
women'
s
ca
ncers. A
c
cordi
ng to
ho
spital data
and
200
8 data,
the
incid
e
n
c
e wa
s 36.5 cases
/ 100,000 wo
men / year [2].
Variou
s
studi
es
have
co
nfirmed
this is the d
e
tectio
n
of ea
rly sta
g
e
b
r
ea
st
can
c
er may
improve
pro
g
nosi
s
. Mamm
ogra
phy tech
nique
remai
n
s the e
s
senti
a
l dete
c
ting
brea
st, the m
o
st
efficient in
monitori
ng a
nd ea
rly det
ection
of bre
a
st can
c
e
r
. It helps to
hi
ghlight p
o
ten
t
ial
radiol
ogi
cal si
gns
su
ch a
s
suspi
c
io
us o
p
a
c
ities
whi
c
h can tran
slate from malign
ant
lesion
s.
Ho
wever, de
spite si
gnifica
nt prog
re
ss i
n
term
s of eq
uipment, all radiolo
g
ist
s
re
cog
n
ize
the diffic
u
lty
of interpreting mammograms
whic
h fu
rt
her in
cre
a
se
d by the type
of breast tissue
examined.
Mammog
r
a
p
h
ic i
m
ag
es show a
contrast b
e
tw
e
e
n
the t
w
o
ma
in con
s
tituen
ts of th
e
brea
st fatty tissue an
d co
nne
ctive-
fi
bro
u
s matrix. In
gene
ral, it is extremely
difficult to de
fi
ne
norm
a
lity of mammog
r
a
phic im
age
s:
Indeed, the
appe
ara
n
ce
of the ma
mmary gla
n
d
is
extremely variable de
pen
di
ng on th
e pati
ent’s a
ge an
d
the peri
od d
u
r
ing
whi
c
h th
e mammo
gram
is don
e.
Contrast
enh
ancement
ha
s a
n
imp
o
rta
n
t role
in i
m
age
pro
c
e
s
si
ng a
s
it
extract the
useful info
rm
ation from th
e disto
r
ted i
m
age.
Image
enhan
ce
me
nt is used fo
r improving the
visual qu
ality of an image
. Objective o
f
Image
enh
ancement i
s
to pro
c
e
ss
a
n
image
so t
hat
result is mo
re suitabl
e th
an ori
g
inal i
m
age fo
r
sp
e
c
ific a
pplication. Digital im
age en
han
ce
ment
techni
que
s provide a multitude of ch
oice
s for
imp
r
ovin
g the visual q
uality of images.
The fund
ame
n
tal enha
nce
m
ent nee
ded
in ma
mmo
g
r
aphy i
s
an i
n
crea
se in
contra
st,
esp
e
ci
ally for den
se
brea
sts. Co
ntra
st b
e
twee
n
mali
g
nant tissu
e
a
nd n
o
rm
al de
nse
tissue
m
a
y
be pre
s
e
n
t on
a mammog
r
am, but belo
w
the thre
sho
l
d of human p
e
rception.
Evaluation Warning : The document was created with Spire.PDF for Python.
ISSN: 23
02-4
046
TELKOM
NI
KA
Vol. 16, No. 3, Dece
mb
er 201
5 : 539 – 545
540
To illustrate e
dge
s and
sm
all details in
a mammog
r
a
m
image, Un
sha
r
p ma
skin
g filter is
very useful [3
].The un
sha
r
p ma
ski
ng m
e
thod
red
u
ce
s lo
w-frequ
en
cy detail
s
b
u
t amplifie
s hig
h
-
freque
ncy.
This p
ape
r wi
ll provide a
n
overview
of cont
ra
st enh
a
n
cem
ent tech
nique
s it is o
r
gani
ze
as follo
ws: S
e
ction 2, d
e
a
ls with
co
ntrast e
nha
nce
m
ent. Sectio
n 3, descri
b
es evalu
a
tion
o
f
contrast en
ha
ncem
ent tech
nique
s for ma
mmogr
aphi
c. Section 4, experim
ental re
sults.
2. Con
t
ras
t
En
hanceme
n
t
2.2. Homomorphic Filteri
n
g
Cla
ssi
cal te
chniqu
e of hist
ogra
m
equ
ali
z
ation d
o
e
s
n
o
t produ
ce
an
effective result in th
e
pre
s
en
ce
of n
on–u
niform
illumination
pat
tern in
an
ima
ge. The
r
efo
r
e
it is
ne
ce
ssary to develo
p
a
freque
ncy
do
main a
pproa
ch that imp
r
ov
es the
app
ea
ran
c
e
of an i
m
age
by sim
u
ltaneo
us
gra
y
–
level ran
ge
compressio
n a
nd contra
st e
nhan
cem
ent. A homomo
r
p
h
ic fre
que
ncy
domain filte
r
i
ng
techni
que is
investigate
d
in the ca
se o
f
various different illumin
a
tion pattern
s in grey scale
image a
s
well
as col
o
r ima
ge.
The ho
momo
rphi
c filter fun
c
tion d
e
crea
ses the e
n
e
r
gy
of low fre
que
ncie
s a
nd in
crea
se
s
those of hig
h
freque
nci
e
s i
n
the image [4].
We
ca
n m
o
d
e
l an
ima
ge
(matrix of l
u
mi
nan
ce)
a
s
th
e
produ
ct of
two cha
r
a
c
teri
stics:
the
first,
call
ed e
n
lightenm
ent and rated 0 <i
(x,
y) <
∞
i
s
the
amo
unt
of light in
cid
ent on th
e
scene
view. The se
con
d
, called
reflecta
nce a
nd rated 0 <r
(x, y) <1, is
the amount o
f
light reflected
from obje
c
ts
in the scen
e, these two chara
c
te
ri
sti
c
make
up the
overall pe
rceived inten
s
ity,
expre
s
sed a
s
a produ
ct.
The illumin
a
tion-refle
c
tan
c
e model
ca
n be used
to develop
a frequ
en
cy domain
pro
c
ed
ure for improving
the
appe
ara
n
ce of a
n
image by si
multaneo
us gray-l
evel
ra
nge
comp
re
ssion
and contra
st
enha
ncement
. An image f(
x,y) can be e
x
presse
d as
the prod
uct o
f
illumination and reflec
tance components [5].
f
x,
y
i
x,
y
∗r
x,
y
(
1
)
Becau
s
e the
Fouri
e
r tra
n
sf
orm of the produ
ct of
two functio
n
s i
s
no
t separable,
we defin
e:
z
x,
y
l
n
f
x,
y
l
n
i
x,
y
l
n
r
x,
y
(
2
)
Then Equ
a
tio
n
(2) b
e
came
by using FT.
z
u,
v
F
u,
v
F
u,
v
(
3
)
Whe
r
e
F
u,
v
and
F
u,
v
are the Fou
r
ie
r transfo
rm
s of
ln
i
x,
y
and
ln
r
x,
y
respect
i
vely.
If we proc
es
s
z
u,
v
by means o
f
a filter function
H
u,
v
, then, we obtain:
S
u,
v
H
u,
v
z
u,
v
H
u,
v
F
u,
v
H
u,
v
F
u,
v
(
4
)
Whe
r
e
S
u,
v
is the Fouri
e
r tra
n
sf
orm of the re
sult. In the sp
atial domain,
s
x,
y
I
FFT
Su,
v
I
FFT
H
u,
v
F
u,
v
I
FFT
H
u,
v
F
u,
v
(5)
Whe
r
e IFFT i
s
the Inverse
Fouri
e
r T
r
an
sform.
So the output
image ca
n b
e
expre
s
sed
by the functio
n
:
g
x,
y
e
,
(
6
)
The filter function of homo
m
orp
h
ic syst
ems can
be shown
belo
w
.
H
u,
v
,
,
D
u,
v
√
u
v
(
7
)
Evaluation Warning : The document was created with Spire.PDF for Python.
TELKOM
NIKA
ISSN:
2302-4
046
A Hyb
r
id the
Non
s
u
b
sam
p
led Co
ntourl
e
t Tran
sform
and Hom
o
m
o
rphic… (Kha
d
douj Taifi)
541
2.2. Nonsub
sampled Co
ntourle
t Tra
n
sforma
tion
Non
s
u
b
Samp
led Co
ntou
rle
t
Tran
sform
(NSCT
)
p
r
opo
sed by
Cun
h
a
, et al [6, 7], is a
invariant version by tran
sla
t
ion of tran
sfo
r
m cont
ou
rlet
s. The tra
n
sfo
r
med into
co
n
t
ourlets
uses
a
Lapla
c
ian
pyramid [8] fo
r t
he m
u
ltiscale
de
comp
os
iti
on,
the No
nS
ubsample
d
Directio
nal Filter
Bancs for direction
a
l deco
m
positio
n. To
ensu
r
e t
he transl
a
tion inva
rian
ce, NS
CT
is implement
ed
usin
g a pyra
midal structu
r
e Non
s
u
b
Sa
mpled
an
d direction
a
l filterban
ks
Non
s
u
b
Sample
d.
Figure
1 ill
u
s
trate
s
th
e p
r
inci
ple
of NSCT. Th
ese
combi
ne t
w
o
su
cce
ssive
stage
s
of
decompo
sitio
n
invariant by
translatio
n
:
(1)
NonSu
b
samp
led Pyramid
(
NSP) whi
c
h p
r
ovide
s
multi-scale
(2)
NonSu
b
samp
led Dire
ctio
nal
Filte
r
Banks (NS
D
FB)
allo
wi
ng
d
e
comp
osition
according diff
erent o
r
ientati
ons) [9].
The re
sult i
s
a flexible image d
e
com
positio
n, mul
t
iscal
e
, tran
slation invaria
n
t and
Multidire
c
tion
expan
sion
that ha
s
better di
re
cti
onal frequ
en
cy lo
calization a
nd a
fast
impleme
n
tation. NSCT
co
nsi
s
ts of two
fi
lter b
anks, i.e. the NonSu
b
sam
p
led Pyramid Filter B
ank
(NSPFB) a
n
d
the NonSu
b
s
ampl
ed Di
re
ctional Filte
r
Bank (NS
D
F
B
) as
sho
w
n
in
Figure 1
(
a),
whi
c
h split the 2-D frequ
en
cy plane in th
e sub
ban
ds ill
ustrate
d
in Fi
gure 1
(
b
)
.
Figure 1. Non
s
ub
sam
p
led
conto
u
rl
et tra
n
sfor
m. (a)
N
S
FB struct
u
r
e
that impleme
n
ts the NS
CT
;
(b) Ide
a
lized frequ
en
cy part
i
tioning
Figure 2 pre
s
ent
s a com
parative stu
d
y
in
digital mammog
r
ap
hy image enh
a
n
cem
ent
based on th
ree differe
nt algorith
m
s: h
o
mmomo
rp
hi
que filter, un
sha
r
p m
a
skin
g and p
r
o
p
o
s
ed
method
s (u
sing a hybrid
method Co
mbining
No
n
s
ubSa
m
ple
d
Contou
rlet Tran
sfo
r
m a
nd
hommom
o
rp
hique filte
r).
Finally the
enha
nced im
age i
s
obt
ain
ed with
cl
arit
y and fre
e
from
noise.
Figure 2. a) ROIs of the ori
g
inal imag
e,
b) the NS
CT, c)the h
o
mom
r
phi
c filtering,
d) the Un
sh
arp masking,e
)
the
prop
osed enha
ncement
Evaluation Warning : The document was created with Spire.PDF for Python.
ISSN: 23
02-4
046
TELKOM
NI
KA
Vol. 16, No. 3, Dece
mb
er 201
5 : 539 – 545
542
To verify the prop
osed
method, exp
e
ri
me
nts a
r
e
perfo
rmed
with data b
a
s
e MIAS
mammogram
.
In the experi
m
ent, four contra
st enh
a
n
ce
m
ent met
hod
s are
performed b
)
th
e NSCT
c)the h
o
mo
mrphi
c filteri
ng and, d) the Un
sharp masking
and e) the
propo
se
d robu
st
enha
ncement
using m
odified homo
m
orphic filter in
conto
u
rlet (propo
sed e
nha
ncem
ent). Th
e
prop
osed
me
thod e
s
timate
s n
o
ise
cha
r
acteri
stic
in b
a
ckgroun
d re
gion and eli
m
inates
noi
se
in
brea
st
a
r
e
a
i
n
co
rpo
r
atio
n with contrast enha
ncem
ent
of micro
c
al
ci
fication. Expe
rimental
re
sul
t
s
sho
w
that, the prop
osed e
nhan
cem
ent signifi
cant
ly redu
ce
s noi
se
in high noi
se
mammog
r
a
m
s.
3.
Ev
aluation of Con
t
ra
st E
nhanc
ement Technique
s for Mammog
r
aphic
To gaug
e the quality of the contou
rlet based e
nhan
ce
d ima
ge,three q
u
a
n
titative
measures
su
ch a
s
Di
strib
u
tion Sepa
ra
tion Meas
ure
(DSM), Ta
rget to Back-grou
nd contrast
Enhancem
e
nt measure based on
Standard deviati
on (TBES),
Target
to Background
contrast
Enhancem
e
nt measure based on Ent
r
opy
(TBEE)
S. Singh and K. Bovis proposed three
different qua
n
t
itative measure
s
for evalu
a
tion of the enhan
ce
d ima
ge quality [10
,
11].
3.1. Distribu
tion Separati
on Measu
re
(DSM
)
The DSM rep
r
esents h
o
w
sep
a
rate
d are the
distrib
u
tions of ea
ch
mammog
r
am
and is
defined by Eq
uation (8
).
DSM
μ
μ
μ
μ
(
8
)
Whe
r
e,
μ
,
μ
are the mean of the micro
c
al
cificati
on regio
n
of the enhan
ced a
nd ori
g
i
nal
image respe
c
tively.
μ
,
μ
are the mean of the
surrou
ndin
g
tissue of the e
nhan
ce
d and
origin
al
image respe
c
tively.
3.2. Targ
et-to-Backg
rou
nd Cont
r
a
s
t
Enhanceme
n
t M
easu
re
Bas
e
d o
n
Standard
Dev
i
ation
(TB
C
S)
A key obje
c
t
i
ve of a co
n
t
rast e
nhan
cement is t
o
maximize th
e differen
c
e
betwe
en
backg
rou
nd
and target
mean g
r
ay l
e
vel and
en
su
re that the
homog
eneit
y
of the ma
ss i
s
increa
sed
aid
i
ng the
visu
al
ization
of its
boun
dari
e
s a
nd lo
catio
n
.
Usi
ng th
e
rati
o of th
e
stan
dard
deviation of the grayscale
s
within the t
a
rget
b
e
fore
and after the
enha
ncement
, we can
qua
ntify
this improve
m
ent usin
g the TBCS given in (9).
TBCS
μ
μ
μ
μ
σ
σ
(
9
)
Whe
r
e,
,
are
the stan
da
rd deviatio
n
of the gr
a
y
scal
e
s
com
p
risi
ng the t
a
rget a
nd
backg
rou
nd
before
and af
ter the enh
an
ceme
nt. Assuming that th
e target ha
s
a smalle
r me
an
before a
nd af
ter enha
ncem
ent comp
ared
to the background.
3.3. Target-to-Backg
rou
nd Con
t
ras
t
Enha
nceme
n
t Mea
s
ure
Bas
e
d on En
trop
y
(TBCE
)
This mea
s
u
r
e is an
exten
s
ion
of the
T
B
C met
r
ic.TB
C
E i
s
b
a
sed
on the
ent
rop
y
of the
regio
n
s
rathe
r
than in the standard devia
tions an
d is d
e
fined by Equ
a
tion (10
)
.
TBCE
μ
μ
μ
μ
£
£
(
1
0
)
Whe
r
e,
£
and £
,are the entro
py of the target in the origi
nal and e
nha
nce
d
image
s.
Evaluation Warning : The document was created with Spire.PDF for Python.
TELKOM
NIKA
ISSN:
2302-4
046
A Hyb
r
id the
Non
s
u
b
sam
p
led Co
ntourl
e
t Tran
sform
and Hom
o
m
o
rphic… (Kha
d
douj Taifi)
543
4. Experimenta
l
Results
Figure 3, Figure 4 an
d Fi
gure 5
sho
w
the
plots of DSM, TBCS
and TBCE
metrics
respe
c
tively use
d
to m
e
asu
r
e th
e e
nhan
cem
ent
ability of the
NSCT,
u
n
sh
arp
ma
sking,
hommom
o
rp
hic filter and
prop
osed met
hod (p
ro
po
se
dM)
Figure 3.
Plots of DSM met
r
ics
Figure 4.
Plots of TBCS metrics
Figure 5.
Plots of TBCE metrics
From th
ese
evaluation
of cont
ra
st en
han
ceme
nt tech
niqu
e, we
con
c
lu
de th
at hybri
d
method (NSCT and hom
o
m
orp
h
ic filter) gives good result
s for enh
anci
ng mam
m
ograms.
4.1. Confu
s
ion Matrix
To mea
s
u
r
e
accuracy a
medical test
is don
e. Let'
s
say
we test some pe
opl
e for the
pre
s
en
ce
of dise
ase. Some of these peopl
e hav
e
the dise
ase, and ou
r te
st demon
strate that
they are
po
sit
i
ve. They a
r
e
call
ed true
p
o
sitives
(TP
)
. Some
have t
he di
sea
s
e,
b
u
t the te
st sa
ys
they do not. They are
call
ed false n
ega
tives (FN). Some do n
o
t have the dise
a
s
e, and th
e test
s
a
id th
e
y
do
n
o
t
- tr
ue
ne
ga
tive
(
T
N
)
.
F
i
n
a
lly
, there m
i
ght be
pe
opl
e in
goo
d
he
alth who
hav
e a
positive resul
t
- false po
sitives (FP). Th
us, t
he num
b
e
r of true p
o
s
itives, false
negative
s
, tru
e
negative
s
an
d false po
sitives ad
d up to 100 % of the whol
e (Ta
b
le
1).
Table 1. Co
nfusio
n matrix
Actual
Predicted
Positive
Negative
Positive
T
P
(
T
r
ue Positive)
F
P
(
F
alse Positive)
Negative FN(
F
alse
Negati
v
e)
TN(
T
rue
Negative)
TP: correct cl
assificatio
n
of abnormal
FP: incorre
c
t cla
ssifi
cation
of abnormal
TN: c
o
rrec
t c
l
ass
i
fic
a
tion of normal
FN:
incorrect cla
ssifi
cation of
normal
Evaluation Warning : The document was created with Spire.PDF for Python.
ISSN: 23
02-4
046
TELKOM
NI
KA
Vol. 16, No. 3, Dece
mb
er 201
5 : 539 – 545
544
4.2. Data
bas
e
of Tes
t
ing
Set and Trai
ning Set Images
A compa
r
ative study is done between
wavele
t and
steera
b
le p
y
ramid tran
sf
orm to
cla
ssify microcal
cification
s into norm
a
l or
abn
ormal (Benig
n
or Malign
a
n
t) ca
se
s u
s
ing
multiorintatio
n and multiresol
ution re
p
r
es
entation
s
.
140 mamm
ogra
m
s o
b
tai
ned from MI
AS
databa
se
wa
s used in this
study (Ta
b
le
2).
Table 2. Nu
m
ber of traini
ng
and testing
set
Catego
ry
database
No. of training se
t
No. of testing set
Normal
36
18
Anormal
58
28
4.3. Featur
e Extrac
tion
Feature extra
c
tion i
n
volves simplifying
the
am
ount
of re
sou
r
ce
s re
quire
d to
de
scrib
e
a
large
set of data accu
rat
e
ly. In the propo
sed me
th
od of the GIST descripto
r built by Torralba
[12], it is close enoug
h to the Gabo
r filter ban
k.
The new versio
ns use the stee
rable pyramid
s
.
The "ra
w
" de
scripto
r
is co
nstru
c
ted a
s
f
o
llows:
It turns the image into a
ban
k of Gabo
r filters with
N
σ
sc
ales
N
θ
orientation
s
scale, we
obtain N
= N
θ
x N
σ
image
s. Each im
ag
e wa
s divide
d into-M x M sub
-
ima
g
e
s
. We calculate
the
energy of ea
ch
sub im
ag
e, so
we get
a vect
or
of N x M x M.and we use the conventio
nal
para
m
eters:
1) N
σ
= 4
2) N
θ
= 8
3)
M = 4
Then the vect
or si
ze is 5
1
2
.
4.4. Classific
a
tion Phas
e
In this phase, knn cla
s
sifier is used to cl
a
ssify the imag
es. The enh
a
n
ce
d image i
s
given
as the
inp
u
t to cla
s
sify mammog
r
am
s into
norm
a
l and
ab
no
rmal. If it’s
contai
ns tu
m
o
r
(micro
cal
c
ification).
Expe
riment
s we
re
don
e wi
th several
value
s
of k and D0
value of cutoff
need
ed for th
e homom
orp
h
ic filter.
The pe
rform
ance of the prop
osed a
p
p
roa
c
h
e
s fo
r the cla
ssifi
cation of No
rmal and
Anormal p
a
ttern
s is me
asured by cl
assi
fica
tion a
c
curacy, sen
s
itivity and spe
c
ificity.
It has been
comp
uted ba
sed o
n
the numbe
r of
correctly dete
c
ted no
rmal/
abno
rmal
image
s i
n
o
r
der to eval
ua
te the
efficie
n
cy a
n
d
ro
bu
stne
ss of th
e
algo
rithm. T
he M
e
tric i
s
as
follows
:
%
(
1
1
)
MIAS databa
se i
s
empl
oye
d
for expe
rim
ents.
94 traini
ng imag
es
an
d 46 Te
sting I
m
age
s
are u
s
ed.
Figure 6. Cla
ssifi
cation a
c
curan
c
y rate the NSCT and
propo
se
d me
thod with different value of
cutoff need
ed
for the homo
m
orp
h
ic filter
Evaluation Warning : The document was created with Spire.PDF for Python.
TELKOM
NIKA
ISSN:
2302-4
046
A Hyb
r
id the
Non
s
u
b
sam
p
led Co
ntourl
e
t Tran
sform
and Hom
o
m
o
rphic… (Kha
d
douj Taifi)
545
Figure 6 illustrates the
classifi
cation accura
ncy rate
obtained from the method used.
From th
e fig
u
re a
r
e
co
ncluding th
at the propo
se
d
method i
s
better tha
n
ather m
e
tho
d
in
cla
ssif
i
cat
i
on rat
e
.
From
the
Fig
u
re
6 th
e
be
st results f
r
o
m
all th
e tra
n
sforms (NS
C
T, p
r
op
ose
d
M
with
D_0
=
0.2, D_
0=0.5 and D_0=0.
9) are alway
s
obtai
ned
fo
r
k
=
1. but for th
e tran
sforms are
individually produ
cing thei
r best re
sult
s for differe
nt values of k.
The
be
st re
sults a
r
e
obtai
ned fo
r
k
=
5; wh
ere a
s
for p
r
op
osed
M with
D_0
=
0.2 the
corre
s
p
ondin
g
value
of
k =
5
as
for
propo
sed
Meth
od (pro
po
sed
M
) with
D_0
=
0.9 K=5
a
s
for
prop
osedM with
D_0
=
0.5
k
= 1 sho
w
s the best a
c
cura
ncy rate 95%.
5. Conclu
sion
Contrast
enh
ancement
ha
s an
impo
rta
n
t role
i
n
Im
age p
r
o
c
e
ssi
ng a
s
it extract the
useful info
rm
ation from th
e disto
r
ted i
m
age.
Image
enhan
ce
me
nt is used fo
r improving the
visual
quality
of an im
age.I
n
this pa
per,
we
pro
p
o
s
e
a
hybrid
the
Contourl
e
t an
d
Homm
omo
r
p
h
ic
Filtering fo
r Enhan
cin
g
Ma
mmogram
s. The propo
se
d method giv
e
s an im
pre
s
sive enh
an
ce
ment
of the mi
cro
c
alcification
s a
nd a
n
autom
atic
cla
ssifi
ca
tion for cl
assi
fying the
digit
a
l mam
m
og
ram
has be
en propo
sed. Preli
m
inary expe
riments
are carri
ed out on
MIAS database. From the
experim
ental
results, it is
o
b
se
rved that
the pro
p
o
s
ed
mammog
r
a
m
cla
ssifi
cati
on sy
stem ba
sed
on Gist featu
r
e with D_
0=0
.
5 set gives the bette
r pe
rf
orma
nce of 95% of classification rate.
Referen
ces
[1]
G Kom, A T
i
e
deu, M
Kom,
C Ng
uem
gne,
J Gonsu.
Déte
ction
autom
ati
que
des
o
pacit
és d
ans
le
s
mammogra
p
h
i
es par
la
méth
ode
de
min
i
m
i
satio
n
d
e
l
a
s
o
mme d
e
l
’in
er
tie.
ELSEVIER.
200
5: 34
7-
356.
[2] http://
w
w
w
.
co
n
t
releca
ncer.ma
/
docume
n
ts/rapport-d
a
ctivites
-200
6/cons
ulté
le
12/04/
201
2
[3]
Lure F
Y
M, Jo
nes PW
, Gaborski RS.
Multi
-resoluti
on
uns
harp
mask
in
g techni
qu
e for ma
mmogr
a
m
imag
e enh
anc
ement
. In SPIE Proceed
in
gs. 199
6; 27(1
0
): 830-8
39.
[4]
S Hela
n, Je
an-C
h
ristop
he
Burie, T
h
ierry
Bo
u
w
ma
ns,
Stépha
ne B
a
zeil
le. Obj
e
ct detectio
n
i
n
und
er
w
a
ter im
ages. 20
06.
[5]
W
enli L
i
u, Pen
g
He, Hui
Li, Hon
gbo Y
u
.
Impr
ove
m
ent o
n
the Alg
o
rith
m of Ho
mo
mo
rphic F
ilter
ing
.
Internatio
na
l
Confer
ence
o
n
El
ectrical
and
Co
m
pute
r
Eng
i
ne
eri
n
g
Adva
nces
i
n
Bi
omed
ica
l
Engi
neer
in
g. 2012; 11.
[6]
MN Do, M Vetterli.
T
he
contour
let tra
n
sform: a
n
e
fficient
dir
e
ctiona
l multir
eso
l
utio
n i
m
a
g
e
repres
entati
o
n
.
IEEE
T
r
ans.Image Proc. 2
005
; 14(12).
[7]
Yue Lu, Min
h
N.
A new contour
let trans
form w
i
th
sha
r
p freque
ncy l
o
cali
z
a
tio
n
. Proc. of IEEE
Internatio
na
l C
onfere
n
ce o
n
Image Proc
essi
ng. 200
6.
[8]
PJ Burt, EH A
dels
on. T
he L
apl
acia
n p
y
r
a
mid as
a c
o
mp
act ima
ge c
o
d
e
.
IEEE Trans.
Comm
. 19
83
;
31: 532-
54
0.
[9]
RH Bamberger, MJ
T
Smith.
A filter b
ank
for the d
i
rectio
nal
dec
ompos
i
t
ion of
imag
es
: theory a
n
d
desi
g
n
. IEEE
Trans. Sign
al Pr
oc. 1992; 4
0
(4)
:
882-89
3.
[10]
K Bovis, S S
i
ngh.
E
nha
nce
m
e
n
t T
e
ch
niq
ue Ev
alu
a
tio
n
usin
g Qu
anti
t
ative Me
asur
es on
Di
gita
l
Mammogram
s.
Proceed
in
gs o
f
5th Internatio
nal W
o
rksh
op
on Di
gital Mam
m
ogra
p
h
y
. 20
0
1
; 547-5
53.
[11]
Vikrant Bh
ateja
,
S
w
a
p
n
a
Devi.
An Improv
ed
Non-
Lin
ear T
r
a
n
sformatio
n
F
unction for En
ha
nce
m
e
n
t o
f
Mammogra
p
h
i
c Breast Masses
. 201
1 3
r
d Internati
o
n
a
l Co
nfere
n
ce
on Electro
n
i
cs Compute
r
T
e
chnolog
y (IC
E
CT
2011). 2011.
[12]
Aude Ol
iva, An
tonio T
o
rralb
a.
Mode
lin
g the s
hap
e of the sc
ene: a
ho
listic r
epres
entati
on
of the spati
a
l
enve
l
op
e.
Internatio
nal J
ourn
a
l of Co
mp
uter
Vision.
20
01;
42(3): 14
5-1
7
5
.
Evaluation Warning : The document was created with Spire.PDF for Python.