TELKOM
NIKA Indonesia
n
Journal of
Electrical En
gineering
Vol.12, No.4, April 201
4, pp. 2718 ~ 2
7
2
3
DOI: http://dx.doi.org/10.11591/telkomni
ka.v12i4.4305
2718
Re
cei
v
ed Au
gust 14, 20
13
; Revi
sed O
c
t
ober 2
1
, 201
3; Acce
pted
No
vem
ber 1
2
,
2013
An Improved Medical DR Image Enhancement Method
Bi Juntao
Qingd
ao H
o
tel
Mana
geme
n
t Coll
eg
e, Qingd
ao Sha
n
d
ong
2
661
00, Ch
ina
email:
ju
ntao
_b
i@16
3.com
A
b
st
r
a
ct
As a resu
lt of n
o
ise
interfere
n
c
e, impro
per
e
x
posur
e a
nd th
ick hu
man tiss
ue, det
ail
infor
m
ati
on
of
DR i
m
ag
e w
ill
be bur
ied, i
n
clu
d
in
g uncl
ear e
dge a
nd co
nt
ra
st reduction. This pa
per pr
op
oses a
med
i
cal
D
R
imag
e en
ha
nc
ement pr
ocess
i
ng
base
d
o
n
artificial
fis
h
-s
w
a
rm al
gorit
h
m
. Accord
in
g to DR i
m
age c
ontour
mo
de
l an
d DR
image
bo
und
a
r
y, the enh
anc
ement pr
ocess
i
ng s
e
g
m
e
n
t D
R
i
m
ag
e by art
i
ficial fis
h
-sw
a
r
m
clusteri
ng
meth
od, then e
n
h
a
n
c
e DR class
i
fic
a
tion i
m
ag
e throug
h ad
din
g
GAG operator. S
i
mulati
on res
u
lt
s
show
that the
meth
od c
an ef
f
e
ctively
eli
m
i
n
ate no
ise i
n
D
R
i
m
ag
e,
an
d the d
e
tail
infor
m
ati
on
of enh
a
n
ce
d
DR i
m
ag
e is cl
earer tha
n
befo
r
e me
anw
hi
le
w
i
th high effect
iven
ess an
d ro
bustness.
Ke
y
w
ords
:
D
R
imag
e, imag
e enh
anc
e
m
en
t, artificial fish-sw
a
rm
Copy
right
©
2014 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
Digital Radio
g
rap
h
y (DR)
is advan
ce
d m
edical imag
e photog
rap
h
y
system eq
uipment
widely u
s
ed i
n
the current
medical p
r
ofe
ssi
on. Compa
r
ed to
conve
n
t
ional comput
ed ra
diog
rap
h
y
(CR),
DR
ca
n obtain im
a
ge on th
e di
splay more
directly and
qui
ckly, an
d the
amount of X
-ray
irra
diation
pa
tients affecte
d
is
signifi
ca
ntly r
edu
ced
[1-6]. Additio
nally, DR
ha
s hig
h
sen
s
itivity
and
wide
dy
namic range.
DR an
d
CR both
cha
n
g
e
the X
-ray i
m
age i
n
form
ation into
dig
i
tal
image info
rm
ation, their e
x
posu
r
e latitu
de refle
c
t
certain advanta
g
e
s
comp
ari
n
g
to the ordina
ry
intensifying
scre
en
- film
systems:
Usin
g di
gital
tech
nology,
CR a
nd
DR have
a wi
de
dyna
mic
rang
e which i
s
re
sp
on
sible
for a wi
de ex
posure l
a
titud
e
,
therefo
r
e
a
llowing so
me techni
cal erro
r
in the photo
g
r
aphi
c [7
-13].
For exam
pl
e, t
he good i
m
age
can
be
obtained
even at the pa
rts
who
s
e
con
d
itions a
r
e diffi
cult to gra
s
p
[14-16]. CR
and DR
can
be pe
rform
e
d
variou
s ima
ge
pro
c
e
ssi
ng a
c
cordi
ng to the clini
c
al n
e
eds, such
as various im
a
ge filtering, windo
w width
and
positio
n adj
u
s
tment, the
rich
fun
c
tion
ality of roa
m
ing e
n
larg
ement, ima
g
e stitching,
and
distan
ce, a
r
e
a
, den
sity measure
m
ent.
All the
advan
tages above provide
te
ch
n
i
cal supp
ort
f
o
r
the details
of the diagn
osti
c imagi
ng ob
servatio
n, bef
ore a
nd after comp
ari
s
on
and qu
antitative
analysi
s
[17].
Ho
wever, a
s
a result of noise interfe
r
e
n
c
e, imprope
r exposure a
n
d
thick hum
an
tissue,
detail inform
ation of DR i
m
age will b
e
buried, in
clu
d
ing un
cle
a
r
edge a
nd co
ntrast redu
cti
on,
whi
c
h m
a
y le
ads to mi
sdi
a
gno
se. T
here
f
ore it
ca
n
im
prove
diag
no
stic
efficien
cy
of phy
sici
an
s,
and hel
p do
ctors ma
ke
rig
h
t diagno
si
s to enha
nce DR imag
e. Thi
s
pap
er p
r
op
ose
s
a me
dical
DR i
m
age
en
han
ceme
nt p
r
ocessin
g
ba
sed
on a
r
tifi
ci
al fish
-swa
rm
algorith
m
. Accordi
ng to
DR
image conto
u
r mod
e
l an
d DR im
age
bound
ary, the enh
an
ce
ment pro
c
e
s
sing
segm
en
t DR
image throug
h artificial fish-swa
rm
clustering me
th
o
d
[18-2
0
]. Simulation resu
lts sho
w
that
the
method
can
e
ffectively eliminate noi
se i
n
DR i
m
age,
a
nd the d
e
tail i
n
formatio
n of
enha
nced
DR
image is
clea
rer tha
n
before mean
while
with high effe
ctivene
ss a
n
d
robu
stne
ss.
2.
Medical DR Image Enhan
cement Pro
c
essing
2.1. Establis
h Contour M
odel
The esta
blishment of co
ntour mo
del
for
medical
image is a
b
le to determine the
Periph
eral
co
ntour of the i
m
age. Co
nc
rete step
s are
descri
bed a
s
follows:
In the mod
e
l, conto
u
r pixel
s
fish i
s
the
p
i
xel set
j
Y
which co
nsi
s
ts
of the gro
up of
pixels
with larg
er g
r
ey value, from which ca
n obtain the
pe
rforma
nce co
efficient of art
i
ficial fish. Using
the followin
g
way to cal
c
ul
ate the minim
u
m value of pixels set:
Evaluation Warning : The document was created with Spire.PDF for Python.
TELKOM
NIKA
ISSN:
2302-4
046
An Im
proved
Medical DR Im
age Enhan
cem
ent Meth
od (Bi Ju
ntao
)
2719
Usi
ng form
ul
a (1), the di
stance betwee
n
pixels sets
1
Y
and
2
Y
can be calcul
ated:
12
12
()
eY
Y
Y
Y
(1)
Usi
ng form
ul
a (2), the differen
c
e b
e
twe
en pixels
sets
1
Y
and
2
Y
can be
cal
c
ulate
d
:
''
2
12
()
(
)
PY
Y
Y
Y
(2)
Usi
ng form
ul
a (3), Midp
oin
t
pixels of m
edical ima
ge contour d
a
ta can be calcula
t
ed:
12
1,
1
(,
,
,
)
/
n
nj
k
jk
DY
Y
Y
Y
Y
n
(3)
Thro
ugh th
e
above, we a
r
e able
to get
the pixe
l
s
set of medi
cal
DR imag
e
contour in
orde
r to provi
de data to DR image
seg
m
entation.
2.2. DR Imag
e Segmenta
tion
Assu
me that B is medical
DR ima
ge to be
se
gmente
d
, E is the collecte
d
environment
pixel of the
i
m
age,
C
pre
s
ent
s the
pix
e
l set of
im
a
ge
conto
u
r,
D i
s
key d
e
tail characte
ri
stic
coeffici
ent of
the imag
e,
C
jk
Q
present the
probability of
j
(g
rey
v
a
lue
of the ima
g
e
) a
nd
k
(gradie
n
t coe
fficient) bel
o
ngs to
DR
(edge pixel
s
),
N is the
nu
mber
of pixels calculated
by
formula (4):
1,
1
/
N
C
j
kj
k
j
k
jk
j
QD
D
1,
2
/
N
D
jk
jk
jk
jk
j
QD
D
(4)
From the formula above,
we can get th
e maximum value of grey value of DR image,
expre
s
sed a
s
formula (5
):
ma
x
lg
lg
CC
D
D
jk
jk
jk
j
k
IQ
Q
Q
Q
(5)
A
ssu
me t
hat
(,
)
gy
z
pre
s
ent
s the
pixels of DR image,
M
and
j
present the
grey
level and g
r
e
y
value of the image sepa
rately, the to
tal numbe
r of pi
xels in the en
tire frame i
s
N
,
then the den
sity of image pixel can be
ca
lculate
d
by formula (6):
/(
,
1
,
2
,
3
,
,
1
)
jk
j
k
Qg
N
j
k
M
(6)
Formul
a (6
) should me
et the followin
g
co
ndition:
1,
1
2
N
jk
jk
Q
Then u
s
ing fo
rmula (7) to g
e
t the averag
e
value of the grey value of the image:
2
1,
1
1
,
1
()
NN
Uj
k
j
k
jk
j
k
jQ
k
Q
(7)
Evaluation Warning : The document was created with Spire.PDF for Python.
ISSN: 23
02-4
046
TELKOM
NI
KA
Vol. 12, No. 4, April 2014: 2718 – 2
723
2720
Equation (8) i
s
discrete gre
y
matrix:
2
(,
)
()
/
tu
l
l
U
j
TQ
Y
(8)
Whe
n
the m
a
trix value rea
c
he
s the
max
i
mum value, i
t
is po
ssible t
o
achieve th
e
desi
r
e
d
segm
entation
.
3.
Medical Image Enhanc
e
m
ent Tec
h
n
o
log
y
The no
nline
a
r
enh
an
ceme
nt algorithm f
o
r DR ima
ge
base on
Cont
ourlet transf
o
rm is a
algorith
m
u
s
i
ng a
nonli
nea
r fun
c
tion fo
r
pro
c
e
ssi
ng th
e lo
w fre
que
n
c
y coefficie
n
ts a
nd
sub
-
b
a
nd
coeffici
ents, i
n
ord
e
r to
co
rrect tran
sform coe
ffi
cient
s, and th
en in
verting the a
d
justed
co
effici
ent
to realize the enha
ncement
of image con
t
rast. T
he no
nlinea
r enh
an
ceme
nt ope
ra
tor prom
oter i
s
Anti-symmet
r
y, can enla
r
ge the tra
n
sf
orm
coefficie
n
ts within
a
certai
n ampli
t
ude ra
nge a
nd
enha
nce e
d
g
e
detail
s
in th
e ba
ckg
r
ou
nd
, acco
rdin
g to
the
co
rrelati
on b
e
twe
en
a
m
plitude
of th
e
decompo
se
d coeffici
ent an
d the image
detail. T
he n
on-lin
ea
r enh
ancement op
erato
r
de
scri
bed
in the article i
s
sh
own as fo
llow:
))
exp(
1
(
)
tanh(
)
(
)
(
u
u
c
u
b
w
sign
w
w
E
(9)
w
presents
Co
ntourlet coeffi
cient
s of each sub
-
ima
ge,
]
1
,
1
[
w
.
)
(
w
E
is Co
ntou
rlet
coeffici
ents af
ter enha
ncem
ent.
ma
x
(
|
|
)
w
w
w
5.
0
(
m
a
x(
)
m
i
n
(
)
)
/
(
m
a
x
(
)
)
uw
w
w
t
w
(10)
b and c
determine the amp
litude rag
e
, which i
s
cr
u
c
ial
for cont
ra
st enforcem
ent effects.
The gai
n fun
c
tion p
r
op
erl
y
cho
s
en
ca
n en
sure the
detail ch
ara
c
teri
stics an
d
contrast
can
be
enforced
whil
e the noise coefficient
s wil
l
not be
incre
a
se
d. Figure 1 sho
w
s the output cu
rves of
the enforcem
ent operator i
n
this pap
er.
Figure 1. The
Output of Adaptive Enforc
ement Fun
c
ti
on wh
en c=6;
b=0.9; t=0.8
The M
R
imag
e co
ntra
st en
force
m
ent al
g
o
ri
thm b
a
sed
on Contou
rle
t
transfo
rmati
on can
be gen
erali
z
e
d
as:
(1)
Cho
o
se p
r
ope
r de
co
m
positio
n layer, Lapla
c
e filter and
dire
ctio
nal filter to Contourl
e
t
decompo
se t
he image;
(2) Pro
c
e
s
s the sub
-
b
and i
m
age with n
o
n
linea
r Co
nto
u
rlet co
efficie
n
ts acco
rdin
g
to th
e
Equation (9)
Evaluation Warning : The document was created with Spire.PDF for Python.
TELKOM
NIKA
ISSN:
2302-4
046
An Im
proved
Medical DR Im
age Enhan
cem
ent Meth
od (Bi Ju
ntao
)
2721
(3)
Re
-con
st
ruct
MR ima
ge from
Con
t
ourle
t
c
o
effic
i
ent
s
to attain contras
t
enforced
image.
4. Experiment
and
An
aly
s
is
The size of
the cho
s
en
cell
DR imag
e (a
s
sho
w
n
in Figu
re
2(a)) i
s
512
×5
12. Th
e
overall of the
origin
al imag
e is d
a
rk an
d
the di
stri
buti
on ra
nge
of the gray
is n
a
rro
w. The
ed
ges
of the tissue
are not very
obliviou
s
. All
of thes
e will
affect the do
ctors’ analy
s
i
s
and di
agn
o
s
is
and
will not
benefit furth
e
r
process.
In
orde
r to ve
rify the algo
rith
m, the propo
sed
algo
rithm
is
comp
ared
wit
h
some
no
rm
al DR imag
e
enforcem
ent
me
thod
s. (
e
),
(f),
(g
)
and
(
h
) i
n
Fi
gur
e
3
is
corre
s
p
ond
ed
to the hi
stog
ram
s
of (a),
(b), (c)
and
(d
) in Fig
u
re 2.
The
simulati
on run
s
in th
e
environ
ment
Matlab 7.0 on
the compute
r
with
Pentium
(
R) E5200 2.
5GHz an
d 1G
memory.
(a) Original image
(b) Enh
a
n
c
ed
by
histog
ram
(c)Enha
nced by
wav
e
let
(d) Ima
ge en
han
ced
Figure 2. The
Original a
nd
Enhan
ced Im
age
s
(e) Original image
(f) Image Enh
anced by hist
ogra
m
(g) Ima
ge en
han
ced by wavelet
(h) Ima
ge en
han
ced
with prop
osed alg
o
rithm
Figure 3. Orig
inal Image an
d Enhan
ced I
m
age
In Figure
2, the overall co
ntrast
of (b) a
nd (c) in Fi
gu
re 2 is
enforced in which some detail
s
can
be d
e
tecte
d
.
Ho
wever the
image
is pron
e to b
r
ig
ht an
d loo
k
s
rou
g
h
.
The
parti
cle
s
feeli
ng
of (b)
is obvio
us which ill
ust
r
ate
s
the
histo
g
rams
bala
n
ce
pro
c
e
s
sing
amplifies the
image
noi
ses.
There are fa
ked
sha
d
o
w
in the tissu
e
edge
s par
t
s
of the image in (c). Ho
wever the im
age
Evaluation Warning : The document was created with Spire.PDF for Python.
ISSN: 23
02-4
046
TELKOM
NI
KA
Vol. 12, No. 4, April 2014: 2718 – 2
723
2722
enforced by the algo
rithm in this pape
r doe
sn
’t have
obvious noi
ses and
sha
d
o
ws. The layer
feeling of the image is e
n
fo
rce
d
and the
detail ch
ara
c
t
e
risti
cs a
r
e hi
ghlighte
d
. The image is m
o
re
unambi
guo
us. The noi
se
s
are
effectivel
y restri
ct
ed
when e
n
force
s
the imag
e d
e
tails, a
s
sho
w
n
in Figu
re
2(d). In the
gray image
hi
stogram
s
of
Figu
re 3
a
nd Fig
u
re
2,
the g
r
ay in
(e
)
con
c
e
n
trate
s
belo
w
0.25, the gray in (f)
con
c
e
n
trate
s
betwe
en 0.5
and 1. The g
r
ay distributio
n of
the image
s processe
d by prop
os
ed me
thod, which is (h
) is uniform and it’s be
tween 0 an
d 0.3
and
between
0.45
and
0.
8. The
di
spla
y dynamic ra
ge i
s
exten
d
ed. Th
e visio
n
effect
of th
e
image is imp
r
oved totally and som
e
dark details ca
n be seen
clea
rl
y, which are
benefi
c
ial for the
corre
c
t analy
s
is a
nd dia
g
n
o
se.
In Table
1, the sum of t
he info
rmatio
n
entropy of
the hi
stogra
m
s e
n
forced
image
(PSNR) is th
e least, whil
e
the sum in the image
pro
c
ee
d by pro
p
o
se
d algo
rith
m is larg
est t
hat
can attain m
o
re dia
gno
se
related info
rmation.
Owi
n
g to enforce
the noise wh
en enfo
r
ce the
image
whe
n
usin
g hi
stogram bala
n
ce, its inform
atio
n
entropy i
s
lo
wered. F
r
om
the co
mpa
r
ison
of some n
o
rmal co
ntra
st enfor
cem
ent method
s, the
prop
osed m
e
thod is
an e
ffective MR image
contrast met
hod with so
me
ap
plicatio
n
value
no
matter from
obje
c
tive eva
l
uation i
ndex
es
or
subj
ective vision effects.
Table 1. Co
m
pari
s
on of Informatio
n Entr
opy of Image before a
nd af
ter Enhan
ce
ment
Figure1 (a)
Figure 1
(
b)
Figure 1
(
c)
Figure 1
(
d)
Information ent
ro
p
y
2.4855
2.4502
3.2474
3.9380
PSNR
（）
dB
—
3.8633
27.8419
36.9311
5. Conclu
sion
This
pap
er propo
se
s a
me
dical
DR im
a
ge e
nhan
ce
ment p
r
o
c
e
s
sing b
a
sed o
n
artificial
fish-swarm
al
gorithm. Accordin
g to DR image
cont
our m
odel
a
nd DR ima
g
e
bou
nda
ry, the
enha
ncement
pro
c
e
ssi
ng
segm
ent DR image by a
r
tifi
cial fish-swarm
clust
e
rin
g
method, th
en
enha
nce DR
cla
ssifi
cation
image th
rou
g
h
addi
ng GA
G ope
rato
r. Simulation re
sults
sh
ow th
at
the meth
od
can effe
ctively eliminate
noi
se i
n
DR
ima
ge, an
d the
d
e
tail informati
on of
enh
an
ced
DR ima
ge is
clea
re
r than b
e
fore me
an
while with high
effectivene
ss
and ro
bu
stne
ss.
Referen
ces
[1]
T
S
Huang. Mo
deli
ng a
n
a
l
ysis
and visu
al
izati
on of no
nrig
id
obj
ect motion.
In Proc. of 10th Intl. Conf. on
Pattern Reco
g
n
itio
n
. Atlantic. 201
1; 1: 361-3
64.
[2]
JK Aggar
w
a
l, Q Cai, W
Liao, B Sabata.
Articulate
d and e
l
astic non-ri
gi
d moti
on: a revi
e
w
.
In Proc. of
IEEE Comput
e
r
Societ
y
Work
shop
on M
o
tio
n
of No
n-
Ri
gi
d
and Artic
u
l
a
te
d Objects. Aus
t
in. 201
2: 16-
22.
[3]
JK Agg
a
ra
w
a
l,
Q Cai, W
Li
a
o
, B Sa
bata.
Nonri
g
i
d
moti
o
n
a
nal
ys
is: arti
culate
d a
n
d
el
astic moti
on.
Co
mp
uter Visi
on an
d Ima
ge
Und
e
rstand
in
g
. 2010; 7
0
(2): 1
42-1
46.
[4]
K Rohr, HS
Stiehl, R Spr
eng
el, W
Beil
,
T
M
Buzug, J W
eese, MH
Kuhn. Po
int-
base
d
el
astic
registrati
on
of medic
a
l im
age
data
usin
g a
p
p
ro
ximati
ng th
i
n
-pl
a
te spl
i
n
e
s.
Vi
su
al
i
z
atio
n
in
Bi
ome
d
i
cal
Co
mp
uting
. 2
0
11; 54(3): 2
97-
306.
[5]
D T
e
rzopou
los
,
D Meta
xas.
D
y
namic
3D
mode
ls
w
i
t
h
l
o
cal
an
d g
l
o
b
a
l d
e
formati
on
s: deforma
bl
e
super
qua
drics.
IEEE Trans. o
n
PAMI
. 2011; 13(7): 70
3-7
1
4
.
[6]
E Bard
inet,
LD
Co
hen,
N A
y
a
c
he. T
r
acking
medic
a
l
3D
da
ta
w
i
th
a
par
a
m
etric d
e
forma
ble
mod
e
l.
In
Proc. of ECCV.
1996; 5
7
(15):
317-
328.
[7]
L Z
hou, C Ka
mbham
ettu. Hierarchic
a
l stru
cture and n
o
n
r
igid moti
on re
cover
y
from 2
D
monoc
ul
a
r
vie
w
s.
CVPR'00.
2000; 2
4
(2): 752-
759.
[8]
C T
o
masi,
T
Kana
de. Sha
p
e
and moti
on
from im
age st
reams un
der
orthogr
ap
h
y
: A
factorization
method.
Int. Journa
l of Co
mp
uter Visio
n
. 19
92; 9(2): 13
7-1
54.
[9]
QIU Bao-zh
i,
CHEN
Xu. Bo
rder Ima
ge S
e
gmentati
o
n
T
e
chno
log
y
f
o
r
Serio
u
s C
o
ll
isi
on V
ehic
l
e
i
n
Accident.
Computer Si
mu
latio
n
. 2013; 3
0
(1): 215-
218.
[10]
C Harris, MJ
Stephe
ns. A c
o
mbi
ned c
o
rn
e
r
and
ed
ge
de
tector.
In Alvey Vision Conference
. 19
98
;
54(6): 14
7-1
5
2
.
[11]
C Schmid, R
Mohr, C Ba
uc
khag
e. Ev
al
ua
tion of i
n
terest
poi
nt detector
s
.
Internation
a
l
Journ
a
l o
f
Co
mp
uter Visi
on
. 200
0; 37(2)
: 151–1
72.
[12]
CAO Jian, LI Hai- sh
eng, C
A
I Qiang. Res
earch o
n
F
eat
ure E
x
traction
of Image T
a
rget.
Compute
r
Simulation
. 20
13; 30(1): 4
09-
413.
Evaluation Warning : The document was created with Spire.PDF for Python.
TELKOM
NIKA
ISSN:
2302-4
046
An Im
proved
Medical DR Im
age Enhan
cem
ent Meth
od (Bi Ju
ntao
)
2723
[13]
BD Lucas, T
Kana
de. An Iterative Image R
e
gistrati
o
n
T
e
chniq
ue
w
i
th a
n
Appl
icatio
n to Stereo Visi
on.
Internatio
na
l Joint Co
nfere
n
c
e
on Artificial Intelligence
.1
98
1; 65(32): 6
74-
679.
[14]
Can
n
y
J. A Co
mputatio
nal
A
p
proac
h to Edge
Detection.
IEEE Trans on Pattern Analys
is and Machine
Intelli
genc
e
. 19
86; 8(6): 67
9-6
98.
[15]
Bregl
er, J Mal
i
k
.
T
r
acking
pe
opl
e w
i
th tw
ists and
exp
o
n
e
n
t
ial
ma
ps
. In P
r
oc. of 19
98 IE
EE Comp
uter
Societ
y Co
nf.on CVPR. Santa
Barbara. 19
98
: 8-15.
[16]
Z
heng B
a
o
dao
, W
ang Hu
aiji
e. Desi
gn of
Remote
Ima
g
e
Alarmin
g
S
y
s
t
em Based
on
ARM/GPRS.
Co
mp
uter Mea
s
ure
m
e
n
t & Control
. 201
3; 21(
1): 149-1
51, 15
9.
[17]
KP Horn, BG Schunck. Deter
m
inin
g optic
al flo
w
.
Artificial In
tellig
enc
e
. 198
1; 17(3): 18
5-2
03.
[18]
BD Luc
as, T
Kana
de.
An it
erative i
m
a
ge r
e
gistratio
n
tech
niq
ue w
i
th an
app
licati
on to s
t
ereo visi
on
. In
Proc. of DARPA Image Und
e
r
s
tandi
ng W
o
rk
shop. S
an F
r
a
n
sisco: Morg
an
Kaufmann. 1
9
81: 121-
13
0.
[19]
C Cross, AK Jain. Markov r
a
ndom fie
l
d te
xt
ure mod
e
ls.
IEEE Trans on P
a
ttern Ana
l
ysis
and Mac
h
i
n
e
Intelli
genc
e. 19
83; 5(1)
: 25-3
9
.
[20]
Sann
idh
i
KS, MZ
Kurian. Cr
eat
io
n of Dig
ital Elev
ated M
ode
l usi
ng lu
n
a
r imag
es of Cha
ndra
y
a
n
–
1
.
Internatio
na
l Journ
a
l of Electr
ical a
nd Co
mp
uter Engi
ne
erin
g
. 2011; 1(
2): 182-1
86.
Evaluation Warning : The document was created with Spire.PDF for Python.