Internati
o
nal
Journal of Ele
c
trical
and Computer
Engineering
(IJE
CE)
Vol
.
5
,
No
. 5, Oct
o
ber
2
0
1
5
,
pp
. 98
4~
99
1
I
S
SN
: 208
8-8
7
0
8
9
84
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
A New Approach for SAR Image Denoising
M
u
ra
li Mo
han Ba
bu
Y
*
, Subra
m
a
n
ya
m
M
V
**
,
Giri Pra
s
a
d
M
N
***
*
JNTUA,
Anantapur, AP,
India.
**
EC
E Dept.
& Princip
a
l, SREC, Nand
y
a
l, Kur
nool, AP, India.
**
*
Dept. of
EC
E, JNTUA, An
antapur, AP, India.
Article Info
A
B
STRAC
T
Article histo
r
y:
Received
Ja
n 21, 2015
Rev
i
sed
May 21
, 20
15
Accepte
d
J
u
n 8, 2015
In s
y
nthe
tic
aper
ture rad
a
r (SAR) im
aging, the
tra
n
sm
itted pulses from
space
born antenna in
teracts with ground objects and
returned energ
y
or back
scatt
e
red en
erg
y
will be co
lle
ct
ed
to ge
t ba
ckscat
t
e
red im
age
.
In th
is process,
a speckl
e
noise
will be add
e
d b
ecause of
the co
herent im
ag
ing s
y
stem
and
makes the stud
y
of SAR images ve
r
y
diffi
c
u
lt. For be
tter
SAR im
age
processing,
the s
p
eckle has to
be remove
d in
the
initi
al s
t
ages
of
proces
s
i
ng
and maintain all textur
e features
effi
ciently
.
The BM3D method is generally
cons
idered as
s
t
ate of art m
e
tho
d
in
denoising of SAR i
m
ages. In this paper,
it is
proposed
a technique to
despeckle
the speckle no
ise to
th
e maximum
exten
t
whil
e m
a
i
n
taining
th
e
edg
e
ch
ara
c
ter
i
sti
c
s.
Keyword:
BM3
D
Im
age
SAR
Spec
kle
Wavelet
Copyright ©
201
5 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
:
M
u
ral
i
M
o
ha
n
B
a
bu
Y
,
Depa
rt
em
ent
of El
ect
r
oni
cs
a
n
d
C
o
m
m
uni
cat
i
ons E
n
gi
nee
r
i
n
g
,
Sri Venk
ateswara Co
llege
o
f
Eng
i
n
eeri
n
g, C
h
itto
or,
A
n
d
r
a Pr
ad
esh, Ind
i
a,
5
171
27
Em
a
il: k
i
sn
amo
h
a
n
ece@g
m
a
il.co
m
1.
INTRODUCTION
Si
nce i
t
s
ori
g
i
n
i
n
t
h
e
1
9
50'
s, S
A
R
has
b
een
de
vel
o
ped
i
n
t
o
a m
a
t
u
re t
ech
nol
ogy
and
i
s
n
o
w
reco
g
n
i
zed a
s
a hi
ghl
y
s
u
cce
ssful
i
m
agi
ng t
ool
f
o
r
en
vi
r
o
n
m
ent
a
l
m
oni
t
o
ri
n
g
,
f
o
rest
c
o
ver
m
a
ppi
ng
,
gl
aci
er
m
o
n
ito
rin
g
and
military
ap
p
l
icatio
n
s
th
at req
u
i
re b
r
o
a
d
-
area i
m
ag
in
g
at h
i
gh
reso
lu
tions. Syn
t
h
e
tic apertu
re
rada
r send m
i
crowa
v
e pulse
s towards
the
earth. T
h
ese signals interact
w
ith
g
r
ou
nd
ob
j
ects an
d
th
e b
ack
-
scattered
rad
i
atio
n
of th
e scen
e will b
e
ob
tain
ed
at
radar. Brigh
t
er areas are
produ
ced
b
y
stro
ng
er rad
a
r
responses, due
sum
m
ation
of m
i
crow
a
v
e
re
turns a
n
d
darker areas a
r
e
from
weaker ra
dar res
p
onse
s
. The
back
scat
t
e
red resp
o
n
se gene
r
a
l
l
y
depen
d
s o
n
f
r
e
que
nc
y
or
wa
vel
e
n
g
t
h
o
f
t
h
e ra
dar use
d
,
m
i
crowa
v
e pul
s
e
ori
e
nt
at
i
on o
r
pol
a
r
i
zat
i
on,
i
n
ci
dence an
gl
e of
the m
i
crowa
v
e
pulse
an
d object on
the
earth.
Spec
kle is a c
o
mm
on and special type of
noise in
all co
heren
t
im
ag
in
g
syste
m
s th
at is po
ssib
l
e i
n
SAR im
ag
in
g
syste
m
. Th
e freq
u
e
n
c
y
d
o
m
ain
filters m
a
in
ly started
with revo
lu
tion
from
th
e in
v
e
n
tion
o
f
wavelets.
The
spec
kle has t
o
be m
i
nimized to analy
ze the
SAR im
age correctly. The re
m
ovable of speckle
p
l
ays critical an
d im
p
o
r
tan
t
ro
le in
prepro
cessin
g
of
an
y SAR system
. A v
a
riety of sp
atial d
o
m
ain
filters
and
tran
sform
d
o
m
ain
pro
cedures
are av
ailab
l
e i
n
SAR
do
m
a
in
. Bu
t still th
ere is po
ssi
b
ility t
o
d
e
sp
eck
le
furth
e
r
and
f
u
rt
he
r t
o
achi
e
ve m
a
xim
u
m
reduct
i
o
n o
f
spec
kl
e. S
p
at
i
a
l
dom
ai
n fi
l
t
ers l
i
k
e l
ee [
1
-
4
]
an
d m
a
p fi
l
t
ers [
5
-
7
]
g
a
v
e
b
e
tter d
e
sp
eck
ling
resu
lts
and
are failed
in
p
r
es
erv
i
n
g
th
e edg
e
d
e
tails.
W
a
v
e
let
d
o
m
ain
filters
[8-11
]
h
a
v
e
p
r
od
uced b
e
tter resp
on
se th
an
sp
atial filters. Th
at
is th
e reaso
n
;
the research
ers h
a
v
e
con
cen
t
r
ated
on
tran
sform
d
o
m
ain
f
ilters [12-1
4
]
.
A co
m
b
in
atio
n o
f
tran
sfo
r
m
do
m
a
in
an
d
sp
atial d
o
m
ain
i
m
ag
e
d
e
no
ising
alg
o
rith
m
s
is p
r
esen
ted
in
BM3
D
algo
rit
h
m
.
It co
n
s
ists o
f
h
a
rd
thresh
o
l
d
i
ng
and
wien
er filter in
wav
e
let d
o
main
. Howev
e
r, th
e
sm
o
o
t
h
i
n
g
of h
o
m
o
g
e
n
e
ou
s areas
an
d
th
e p
r
eserv
i
ng
o
f
ed
g
e
s are still n
o
t
well b
a
lanced
in th
ese meth
od
s.
Evaluation Warning : The document was created with Spire.PDF for Python.
I
S
SN
:
2
088
-87
08
I
J
ECE Vo
l. 5
,
N
o
. 5
,
O
c
tob
e
r
20
15
:
984
–
9
91
98
5
But still the BM3D m
e
thod is ge
ne
rally considered as
state of a
r
t m
e
thod i
n
de
noisi
ng
of im
ages a
nd
des
p
eckl
i
n
g o
f
SAR
im
ages [1
5]
. The c
o
m
p
ressi
ve sen
s
i
ng
base
d 3
D
(
C
S3D
)
des
p
ec
kl
i
ng
fram
e
wo
rk i
s
com
p
ri
sed
of
t
h
ree
m
a
jor st
e
p
s;
sel
ect
i
o
n
o
f
s
ubset
s
o
f
pi
xel
s
f
r
om
SA
R
im
ages, rec
onst
r
uct
i
o
n
o
f
SA
R
i
m
age from
each subset
of
pi
xels
us
ing CS t
h
eory, a
n
d stat
istical co
m
b
in
ing of m
u
ltiple reco
nstructed images
b
y
em
p
l
o
y
in
g
selectiv
e 3D
filterin
g
.
In
t
h
is
p
a
p
e
r, we
propo
sed
a tech
n
i
qu
e to desp
eck
l
e th
e SAR im
a
g
es t
o
t
h
e m
a
xim
u
m
ext
e
nt
w
h
i
l
e
m
a
i
n
t
a
i
n
i
ng t
h
e ed
ge c
h
a
r
act
eri
s
t
i
c
s.
W
e
c
o
m
p
ared
o
u
r
p
r
o
p
o
sed
m
e
t
hod
wi
t
h
state-o
f
-art m
e
t
h
od
s i
n
term
s of
q
u
a
lity p
a
rameters lik
e ENL, SSI, PSNR
and
ESI.
2.
DE-SPE
CKLI
NG METHO
D
S
2.
1.
B
M
3
D
Met
h
o
d
The B
M
3
-
D
a
l
go
ri
t
h
m
cont
ai
ns t
w
o m
a
jor
st
eps. T
h
e
fi
r
s
t
one c
o
nt
ai
n
s
t
r
ans
f
o
r
m
dom
ai
n hard
th
resh
o
l
d
i
ng
to b
u
ild
a relatively clean
i
m
ag
e fo
r estim
atin
g
statistics, an
d
th
e second
on
e is h
a
v
i
n
g
i
m
ag
e
d
e
no
ising
u
s
ing
wien
er filteri
n
g
in
t
h
e sam
e
tran
sfo
r
m
d
o
main
. In
b
o
t
h
t
h
e cases sim
i
la
rity b
e
tween
a
g
r
ou
p
o
f
b
l
o
c
k
s
will b
e
ev
al
u
a
ted
li
k
e
in
n
o
n
l
o
cal
ap
pro
ach. Th
e
size o
f
th
e co
llected
group
wi
ll p
r
odu
ce d
i
fferen
t
resp
o
n
ses.
A c
o
m
b
i
n
at
i
on
of
t
r
ans
f
o
r
m
dom
ai
n an
d s
p
a
tial do
m
a
in
i
m
ag
e
d
e
n
o
i
sing
al
go
rith
m
s
is p
r
esen
ted
in
BM3
D
al
g
o
rith
m
[1
6
]
. The selectio
n
of t
h
ese thresh
o
l
d
i
n
g
lev
e
ls v
a
ry
th
e qu
ality o
f
o
u
t
p
u
t
im
ag
es an
d
t
h
e
sel
ect
i
on
of
su
b
ban
d
i
n
t
r
a
n
s
f
o
r
m
dom
ai
n pl
ay
s a m
a
jor r
o
l
e
i
n
de
n
o
i
s
i
n
g
t
h
e i
m
age.
Th
e SAR-BM
3
D
al
g
o
rith
m
d
e
sp
eck
les SAR i
m
ag
es b
y
co
m
b
in
in
g
t
h
e co
n
c
ep
ts
o
f
no
nlo
cal filtering
an
d wav
e
let hard
shrink
age, wh
ich
h
a
s a b
e
tter cap
acity to
preserv
e
relev
a
n
t
d
e
tails wh
ile sm
o
o
t
h
i
ng
h
o
m
o
g
e
n
e
ou
s areas. Ho
wev
e
r,
t
h
e
sm
o
o
t
h
i
ng
o
f
ho
m
o
g
e
n
e
o
u
s
areas an
d th
e
p
r
eserv
i
ng
o
f
edg
e
s
are
still n
o
t
wel
l
bal
a
nce
d
i
n
t
h
ese
m
e
tho
d
s. T
h
e B
M
3D m
e
t
hod i
s
general
l
y
consi
d
ere
d
as s
t
at
e of art
m
e
t
h
o
d
i
n
den
o
i
s
i
n
g
of
i
m
ages an
d
des
p
eckl
i
n
g
o
f
S
A
R
im
ages as Fi
gu
re
1.
Fi
gu
re
1.
B
M
3
D
m
e
t
hod
2.
2. CS
3
D
Me
tho
d
The com
p
ressi
ve sensi
ng (C
S) t
h
eo
ry
pr
o
v
e
d t
h
at
any sparse signal or im
ag
e can be reconstructe
d
fro
m
sa
m
p
les fewer th
an
nu
m
b
er of elemen
ts in
a
signal
or im
age [17]. The
s
e
subsets are
ta
ken as
measu
r
em
en
t v
ectors in CS fram
e
work to ob
tain
m
u
ltip
le SAR im
ag
es b
y
so
lv
i
n
g
co
nv
ex
o
p
tim
i
zatio
n
pr
o
b
l
e
m
.
The
pi
xel
-
wi
se a
v
e
r
agi
n
g
o
f
m
u
l
t
i
pl
e com
p
ressi
v
e
rec
onst
r
uct
e
d
im
ages w
o
ul
d
l
ead t
o
bet
t
e
r
r
e
sul
t
s
com
p
ared t
o
con
v
e
n
t
i
onal
des
p
eckl
i
n
g t
e
chni
que
s.
In t
h
i
s
w
o
rk
, em
pl
oy
sel
ect
i
v
e
3
di
m
e
nsi
o
n
a
l
(3
D)
filterin
g
of m
u
ltip
le recon
s
tructed
im
ag
es to
furth
e
r im
p
r
o
v
e d
e
sp
eck
lin
g resu
lts.
Thi
s
despe
c
kl
i
n
g
f
r
am
ewor
k i
s
com
p
ri
sed o
f
t
h
ree
m
a
jor s
t
eps;
sel
ect
i
on of su
bset
s of
p
i
xel
s
fr
om
SAR im
ages,
reconstruction
of im
age from each s
ubset
of
pi
xels
using
CS the
o
ry, and statistical combining
o
f
m
u
ltip
le reco
n
s
t
r
u
c
ted
im
a
g
es
b
y
em
p
l
o
y
in
g selectiv
e
3D filtering
, as
sh
own
in Fi
g
u
re 2
.
Evaluation Warning : The document was created with Spire.PDF for Python.
I
J
ECE
I
S
SN
:
208
8-8
7
0
8
A New Appr
oach for
SAR
Image
Denoising
(Mu
r
a
li Moha
n
Bab
u
Y)
98
6
Fi
gu
re
2.
C
S
3
D
m
e
t
hod
2.
3.
Pr
op
ose
d
Me
th
od
The radar im
age will be cropped according to user'
s
applications size.
If the im
age
is in geotiff
form
at
, generall
y
i
t
i
s
noi
se fr
ee one;
ot
herw
i
s
e it
i
s
cont
ami
n
at
ed wi
t
h
noi
se. Th
e necessary sized
image will
b
e
g
i
v
e
n
to
n
o
n
d
ecim
a
ted
wav
e
let tran
sfo
r
m
an
d
ex
tract th
e all p
o
ssib
l
e
co
efficien
ts from
it. After separatin
g
the coefficients, the required coefficien
t
s
wi
l
l
be col
l
e
ct
ed and t
h
ey
wi
l
l
be gi
ven t
o
i
n
verse no
n deci
m
a
ted
wav
e
let tran
sform
.
Th
e o
u
t
pu
t i
m
a
g
e o
f
prev
iou
s
step
will b
e
d
i
v
i
d
e
d
in
to
d
i
fferen
t b
l
o
c
k
s
wi
th
a fix
e
d
size. Each
b
l
o
c
k
will b
e
co
m
p
a
r
ed
with
o
t
h
e
r b
l
o
c
k
s
an
d eu
clid
ean
d
i
stan
ce will b
e
calcu
lated
.
If th
e d
i
stan
ce is b
e
lo
w
a
th
resh
o
l
d v
a
lue, th
en
th
e
b
l
o
c
k
will b
e
g
i
v
e
n
to d
i
sc
rete wav
e
let tran
sfo
r
m
an
d
th
e co
efficien
ts will b
e
ext
r
act
ed. M
a
ni
pul
at
i
on of co
effi
ci
ent
s
wi
ll
be do
ne and
an
i
nverse di
scret
e
wavel
e
t
t
r
ansform
wi
ll
be appl
i
e
d
to reconstruct the im
age as shown in Figure
3.
Fi
gu
re
3.
Pr
o
p
o
se
d m
e
t
hod
The l
a
st
obt
ai
ned i
m
age wi
ll
be gi
ve
n t
o
di
scret
e
wavel
e
t
t
r
ansform
once agai
n a
n
d
cal
cul
a
t
e
all
p
o
ssib
l
e co
effi
cien
ts. A wien
er filter wil
l
b
e
ap
p
lied
o
n
sp
ecific set o
f
co
efficien
t v
a
l
u
es an
d
an
inv
e
rse
d
i
screte wav
e
let tran
sfo
r
m
will
b
e
app
lied
to
reco
n
s
tru
c
t th
e imag
e.
The quality param
e
ters
like equi
valent num
ber
looks (ENL), speck
le
suppression index (S
SI),
correl
a
t
i
on coeffi
ci
ent
(C
C
)
, edge savi
ng
or
preservi
n
g
i
n
d
e
x (ESI) an
d p
eak si
gnal
t
o
noi
se rat
i
o
(PSNR
)
wi
ll
be m
e
asured f
o
r al
l
t
h
e out
put
im
ages of di
ffe
rent
l
a
t
e
st
despeckl
i
ng m
e
t
hod
s al
ong
wi
t
h
pr
op
osed al
go
ri
t
h
m
.
1. Pe
ak
Si
g
n
a
l
t
o
N
o
i
s
e R
a
t
i
o
(
P
SN
R)
:
PSNR is th
e
ratio
b
e
tween
the
m
a
x
i
m
u
m
si
g
n
a
l
p
o
wer and
t
h
e
cor
r
u
p
t
i
n
g n
o
i
s
e po
wer
.
It
i
s
the fact
o
r
t
h
at
j
u
d
g
es w
h
et
her
a
m
e
t
hod i
s
p
r
ovi
di
n
g
g
o
o
d
den
o
i
s
i
n
g sche
m
e
or
n
o
t
. Hi
g
h
e
r t
h
e v
a
lu
e m
ean
s hig
h
e
r im
ag
e q
u
ality.
PSNR
=
10l
og
1
0
* (2
n
-
1
)
/
M
S
E
(
1
)
Evaluation Warning : The document was created with Spire.PDF for Python.
I
S
SN
:
2
088
-87
08
I
J
ECE Vo
l. 5
,
N
o
. 5
,
O
c
tob
e
r
20
15
:
984
–
9
91
98
7
2. E
q
ui
val
e
nt
N
u
m
b
er
of
Lo
oks (
E
N
L
)
:
Ot
her t
h
an
PS
N
R
val
u
e,
ENL
val
u
e
plays critical role in
cohe
re
nt
syste
m
s like SAR
processing. T
h
e E
N
L
va
lues s
p
eak ab
out
t
h
e e
ffi
ci
e
n
cy
i
n
sm
oot
h
i
ng
spec
kl
e
no
i
s
e o
f
im
age ove
r
h
o
m
ogeneo
u
s a
r
e
a
s.
ENL =
(m
ean/
st
anda
rd
de
vi
at
i
o
n
)
2
(
2
)
3
.
C
o
efficien
t
o
f
Co
rrel
a
tio
n (CC
)
:
Correla
tion coefficient gi
ves
how
fa
r the two im
ages correlated t
o
each
othe
r, that m
eans
how
far t
h
e des
p
eckle
d i
m
age is near
by the
ori
g
inal
im
age. It indi
cates the stre
ngth
o
f
l
i
n
ear rel
a
t
i
o
ns
hi
p
bet
w
ee
n t
h
e
ori
g
i
n
al
(
x
)
an
d
den
o
i
s
e
d
m
a
ges (^
x)
.
I
f
t
h
e
val
u
e i
s
near
t
o
1,
t
h
e
n
t
h
ere
ex
ists stro
ng
er
p
o
s
itiv
e correl
a
tio
n
b
e
tween
t
h
e
x
an
d ^x imag
e.
C
x,^x
= E[(x-µ
x
) (
^
x-µ
^x
)]
/
σ
x
.
σ
^x
(
3
)
whe
r
e µ
x
and µ
^x
are
m
ean values of
ori
g
inal and de
s
p
ec
kled SAR im
ag
es respectively
and
σ
x
and
σ
^x
are
st
anda
rd
de
vi
at
i
ons
o
f
ori
g
i
n
al
an
d s
p
ec
kl
e re
m
oved i
m
ages res
p
ect
i
v
el
y
.
4. Speckle
Suppression
Index
(SSI)
:
The rat
i
o o
f
st
a
nda
rd
devi
at
i
o
n t
o
m
ean is
use
d
to m
easure the s
p
eckle
streng
th
in
an
i
m
ag
e. Let x
an
d
^
x
are orig
i
n
al and
d
e
sp
eck
l
ed
SAR i
m
a
g
es. Th
e SSI plays a critica
l
ro
le in
rada
r im
age processing
steps. It is
defi
ned as
SSI
= [(
var(
^x
))
1/2
/
m
ean(^x
)
]
* [m
ean(
x
) /
(
va
r(
x)
)
1/2
]
(
4
)
It shou
ld
b
e
less th
an
1
.
Lower th
e v
a
l
u
e m
e
a
n
s
higher the
s
p
eckle
re
duction.
Ideal val
u
e i
s
zero.
5 E
d
ge
Save
I
ndex
:
Edg
e
sav
e
ind
e
x
(ESI) exp
l
ain
s
t
h
e cap
a
b
ility o
f
the i
m
ag
e how far th
e edg
e
pro
p
e
rties
have
bee
n
m
a
intained.
ESI
h
=
|
̂
i,
j
1
̂
i,
j
|
/
|
x
i
,
j
1
i,
j
|
(5
)
ESI
v
=
|
̂
i,
j
1
̂
i,
j
|
/
|
x
i
,
j
1
i,
j
|
(6
)
whe
r
e
^
x
is reco
v
e
red
im
ag
e;
x
is orig
i
n
al SAR im
ag
e;
m
and n a
r
e the
num
b
er of
rows
and col
u
m
n
s of t
h
e
SAR im
age.
3.
E
X
PER
I
ME
NTAL
RES
U
LTS
All th
e filterin
g
and
tran
sfo
r
m d
o
m
a
i
n
tech
n
i
qu
es will
b
e
ap
p
lied
and
tested
m
a
in
ly to
t
h
e i
m
ag
es of
RISAT-1
senso
r
. Tab
l
e
1
g
i
v
e
s
d
i
fferen
t
qu
ality facto
r
valu
es to
test imag
e. Gen
e
ral
l
y ENL v
a
lu
e
p
l
ays a
m
a
jor r
o
l
e
i
n
deci
di
n
g
t
h
e d
e
spec
kl
e al
gori
t
hm
t
o
be
used in microwa
v
e im
age proce
ssing. The hi
gher the
val
u
e
of E
N
L
,
t
h
e bet
t
e
r t
h
e
p
e
rf
orm
a
nce of
des
p
eckl
i
n
g t
e
chni
que
. Am
ong al
l
t
h
e t
ech
ni
q
u
es, t
h
e p
r
o
pos
e
d
m
e
t
hod gi
v
e
s b
e
t
t
e
r
ENL
val
u
e.
Al
o
ng
wi
t
h
ENL val
u
e
,
SSI
i
s
an im
port
a
nt
param
e
t
e
r in t
h
e fi
el
d of
SAR
im
agery
.
General
l
y
,
spec
kl
e su
p
p
re
ssi
on i
nde
x i
s
l
e
ss t
h
an
1
.
L
o
wer t
h
e val
u
e
means highe
r
the s
p
eckle
redu
ctio
n.
Am
o
n
g
all th
e
t
echni
q
u
es
, t
h
e
p
r
o
p
o
sed
m
e
tho
d
gi
ve
s l
east
SSI
val
u
e.
Th
at
m
eans m
a
xim
u
m
speckl
e
h
a
s bee
n
rem
oved.
Correlation
c
o
efficient gives how
far the
two im
ages corre
lated to each
other, t
h
at m
e
a
n
s
how fa
r
t
h
e des
p
eckl
e
d
im
age i
s
near by
t
h
e spec
kl
e
d
i
m
age. From
t
a
bl
e 2, we ca
n say
t
h
at
ou
r
m
e
t
hod
gi
ves
bet
t
e
r
CC v
a
lu
e.
Ed
ge savi
ng i
nde
x i
s
t
h
e c
r
i
t
i
cal
param
e
t
e
r i
n
any
de
n
o
i
s
i
ng m
e
t
hod.
W
i
t
h
o
u
t
bet
t
e
r
edge s
a
vi
n
g
param
e
t
e
r, any
m
e
t
hod
h
o
l
d
s
not
hi
n
g
.
It
can
be o
b
se
rve
d
t
h
at
t
h
e
pr
op
os
ed t
ech
ni
q
u
e
g
i
ves bet
t
e
r E
S
I
val
u
es
am
ong al
l
m
e
tho
d
s.
Si
m
i
l
a
rl
y
,
t
h
e PS
NR
v
a
l
u
e t
h
at
i
s
o
b
t
ai
ned by
t
h
e
pr
o
pose
d
m
e
t
hod i
s
hi
g
h
.
It
can be
n
o
t
ed
th
at,
ou
r d
e
sp
eck
ling
t
ech
n
i
q
u
e
p
e
rform
s
b
e
tter th
an
o
t
h
e
r m
e
th
od
s in
qu
ality facto
r
s term
s of ENL,
SSI,
C
C
,
ES
I a
nd
PS
NR
. T
h
e
des
p
eckl
i
n
g
r
e
sul
t
s
o
f
t
h
e
p
r
op
ose
d
t
ech
ni
q
u
e a
nd
ot
he
r m
e
t
h
o
d
s t
o
i
m
ag
e are
gi
ve
n i
n
Fi
g
u
r
e
4 an
d Tabl
e
2. T
h
e PS
NR
val
u
es f
o
r di
f
f
e
rent
des
p
ec
kl
i
ng t
ech
ni
ques
on
di
ffe
re
nt
R
I
SAT
-
1
im
ages ha
ve
b
een
gi
ve
n i
n
Fi
gu
re
5.
The area t
h
at we conside
r
e
d
for
des
p
eckling is
MAYKOP, RUSSIA. T
h
e cent
r
a
l
latitude a
nd
l
o
n
g
i
t
ude
val
u
es are
4
4
.
6
09
a
n
d
4
0
.
0
94
re
sp
ect
i
v
el
y
.
The
c
h
aract
eri
s
t
i
c
s
o
f
t
h
e
sce
n
e
hav
e
bee
n
gi
ve
n
b
e
l
o
w
i
n
t
a
bl
e
1.
It
i
s
a R
I
S
A
T-
1 c
o
a
r
se
resol
u
t
i
o
n
s
canS
A
R
m
ode
C
R
S
dat
a
Evaluation Warning : The document was created with Spire.PDF for Python.
I
J
ECE
I
S
SN
:
208
8-8
7
0
8
A New Appr
oach for
SAR
Image
Denoising
(Mu
r
a
li Moha
n
Bab
u
Y)
98
8
Tab
l
e
1
.
C
h
aracteristics o
f
Data
Radar
Carr
ier Fr
eq
uency
5.
35 GHz (
C
-
b
and
)
I
n
cidence angle
41.
337 Deg
Polarizations HH
Datu
m
W
G
S
84
Sensor
height at the equator
541 k
m
Revisit ti
m
e
(O
rbit repeat c
y
cl
e)
12 days
Resolution 36
m
e
ters
Mode ASCEND
ING
Sensor
L
ook
Right
Mean Local
Ti
m
e
6 AM
Tabl
e
2.
Q
u
ant
i
t
a
t
i
v
e com
p
ari
s
on
o
f
des
p
ec
k
l
i
ng t
ech
ni
ques
f
o
r R
I
S
A
T-
1
dat
a
ENL SSI
CC
ESI
PSNR
L
E
E
17.
741
84
0.
2870
47
0.
5119
41
0.
2498
04
10.
760
49
W
A
VE
LE
T
0.
4743
52
0.
5844
96
0.
5936
18
0.
1702
58
11.
404
77
CURVE
L
E
T
0.
5650
33
0.
6770
36
0.
7333
1
0.
3342
37
17.
946
99
PCA-
L
P
G
0.
5718
52
0.
4337
72
0.
7735
26
0.
2748
61
14.
249
52
BM
3D
11.
342
83
0.
5234
7
0.
9063
64
0.
6244
99
16.
995
52
CS3D
12.
090
2
0.
0777
5
0.
9055
26
0.
6207
49
16.
953
02
PROPOSE
D
25.
970
77
0.
0547
31
0.
9616
31
0.
6886
64
27.
224
32
(a)
(b
)
(c)
(d
)
(e)
(
f)
(g)
(h)
(i)
Fi
gu
re
4.
R
I
S
A
T-1
dat
a
set
:
R
u
ral
a
r
ea
near
m
a
y
kop ci
t
y
i
n
R
u
ssi
a.
(5
1
2
*
5
1
2
)
(a)
O
r
i
g
i
n
a
l
, (
b
)
N
o
i
s
y
,
(c
) Lee,
(d
)
Wavelet,
(e
) Cu
r
v
elet, (
f)P
CA-LP
G
,
(
g
)
BM
3D,
(
h
)
CS
3D
an
d
(i)
Pr
o
pos
ed
.
Th
e
o
t
h
e
r study area is con
s
i
d
ered
in th
e
wo
rk
is lo
cated
o
n
th
e east coast o
f
In
d
i
a at
a latitu
d
e
of
17
°4
2'
No
rt
h
and l
o
n
g
i
t
u
de
of
83
°
23'
East
and t
h
e t
i
m
e
zone
i
s
GM
T
+ 5:
3
0
. It
i
s
one
of
fam
ous a
n
d
m
a
jor
p
o
rt
s i
n
In
di
a
a
nd t
h
e b
i
ggest
po
rt
o
f
An
d
h
ra
Pra
d
es
h
state. It is Vi
sakhapat
nam
P
o
rt area
. T
h
e table
3
shows t
h
e c
h
a
r
acteristics of
the Terr
aS
AR
-
X
dat
a
.
The
d
e
spec
kl
i
ng
res
u
l
t
s
of
t
h
e
pr
o
pos
ed t
e
c
hni
q
u
e an
d
ot
he
r m
e
t
hods
t
o
i
m
age are gi
ven
i
n
Fi
g
u
re
5
an
d Ta
bl
e
4.
Evaluation Warning : The document was created with Spire.PDF for Python.
I
S
SN
:
2
088
-87
08
I
J
ECE Vo
l. 5
,
N
o
. 5
,
O
c
tob
e
r
20
15
:
984
–
9
91
98
9
Tabl
e 3.
C
h
ara
c
t
e
ri
st
i
c
s
of Te
rraS
A
R
-
X Dat
a
Radar
Carr
ier Fr
eq
uency
9.
65 GHz (
X
-
b
and)
I
n
cidence angle r
a
nge for
:
Str
i
p m
a
p/ScanSAR m
odes
SpotL
i
ght m
odes
20°
-
45°
full per
f
o
r
m
a
nce
20°
-
55°
full per
f
or
m
a
nce
(15°-60° accessibl
e)
Polarizations HH,VH
,HV
,
VV
Pulse Repetition frequency
2.2KHz-6.5KHz
No
m
i
nal or
bit heig
ht at the equator
514 k
m
Revisit ti
m
e
(O
rbit repeat c
y
cl
e)
11 days
I
n
clination 97.
44°
Ascending node
E
quator
i
al Cr
ossing tim
e
18:00 +/-
0.
25h(
lo
cal ti
m
e
)
Tabl
e 4. Q
u
ant
i
t
a
t
i
v
e
com
p
ari
s
on
o
f
des
p
ec
k
l
i
ng
t
ech
ni
ques
f
o
r Ter
r
aS
AR
-
X
dat
a
ENL SSI
CC
ESI
PSNR
L
E
E
10.
581
6
0.
2525
0.
5851
0.
2619
12.
603
8
W
A
VE
LE
T
0.
3666
0.
6509
0.
6518
0.
1656
13.
207
4
CURVE
L
E
T
0.
2209
0.
8803
0.
8836
0.
4415
18.
366
6
PCA-
L
P
G
0.
2259
0.
8456
0.
9914
0.
6940
25.
774
9
BM
3D
11.
480
7
0.
4265
0.
9096
0.
6113
18.
429
5
CS3D
12.
054
3
0.
0757
0.
9118
0.
6172
18.
527
3
PROPOSE
D
12.
816
0
0.
0583
0.
9691
0.
6877
27.
994
1
(a)
(b
)
(c)
(d
)
(e)
(
f)
(g
)
(h
)
(i)
Fi
gu
re 5.
Ter
r
a
S
AR
-
X
dat
a
set
:
Visakhapatnam P
o
r
t
area
in In
dia. (5
1
2
*
5
1
2
)
(a)
O
r
igina
l
,
(
b
) N
o
isy
,
(c
)
Lee, (
d
)
Wavel
e
t, (e)
Cu
rvelet
, (
f)PC
A
-
L
P
G
,
(
g
) BM
3D
, (
h
)
CS3
D
a
n
d (i
)
Pro
p
o
se
d
The p
r
o
p
o
se
d
m
e
t
hod co
ns
i
s
t
s
of B
M
3D
m
e
t
hod al
on
g wi
t
h
u
n
sa
b
s
am
pl
ed deci
m
a
t
i
on. The
decim
a
t
i
on re
duce
s
the qua
lity of the i
m
age that is to
be denoised. It is becau
se of down sampling at
trasnitting e
n
d
and
upsam
p
ling at receivi
ng
end i
n
wa
ve
let
transform
a
t
i
on. The c
o
nversions
of sam
p
li
ng a
r
e
Evaluation Warning : The document was created with Spire.PDF for Python.
IJECE
ISS
N
:
2088-8708
A New Appr
oach for
SAR
Image
Denoising
(Mu
r
a
li Moha
n
Bab
u
Y)
99
0
avoi
ded i
n
ou
r
pr
o
pose
d
m
e
t
h
od
. The PS
NR
val
u
es f
o
r di
f
f
e
rent
des
p
ec
kl
i
ng t
ech
ni
q
u
es
on
di
ffe
re
nt
R
I
SAT
-
1 im
ages (C
i
r
cul
a
r Fi
ne
res
o
l
u
t
i
o
n st
ri
pm
ap m
ode-R
i
g
h
t
Hori
z
ont
al
(
I
m
a
ge-1)
,
C
i
rcul
ar Fi
ne r
e
so
l
u
t
i
on
st
ri
pm
ap m
o
d
e
-R
i
ght
Ve
r
tical (Im
age-2), coa
r
se re
sol
u
tion sca
n
S
A
R-H
o
riz
ontal Ho
ri
zontal (Image-3),
coarse
res
o
l
u
t
i
on
sca
n
SAR
-
Ho
ri
zo
nt
al
Ve
r
t
i
cal
(Im
age-4)
)
have
bee
n
gi
v
e
n i
n
Fi
g
u
r
e
6.
Fi
gu
re
6.
PS
N
R
val
u
es
f
o
r
di
ffe
rent
i
m
ages
4.
CON
C
L
U
SIONS
We tested with d
i
fferen
t ex
istin
g
al
g
o
ri
t
h
m
s
and
p
r
op
ose
d
a
l
go
ri
t
h
m
on
di
f
f
ere
n
t
i
m
ages of
di
f
f
ere
n
t
m
odes of R
I
S
A
T-
1i
m
a
ges and Te
rra
SAR
-
X im
ages. Th
ey
have bee
n
t
e
st
ed wi
t
h
d
i
ffere
nt
n
o
i
s
e l
e
vel
s
(va
r
i
a
nces of 0
.
1, 0.
25
a
n
d
0.
5)
a
nd di
ffe
re
n
t
st
andar
d
si
zes
(2
56
*
2
5
6
an
d 51
2
*
5
1
2
)
.
W
e
con
s
i
d
ere
d
di
ff
erent
m
odes of
R
I
S
A
T-
1 l
i
k
e
coa
r
se res
o
l
u
t
i
o
n s
canS
A
R
m
ode
(C
R
S
),
m
e
di
um
resol
u
t
i
on
s
canS
A
R
m
ode
(M
R
S
),
fine res
o
lutio
n
stripm
ap
m
ode (FRS
) an
d d
i
ffere
nt p
o
larized im
ages (H
H, H
V
, V
H
,
VV
, RH an
d RV) o
f
RISAT-1
fo
r testin
g
.
We
also
tested
o
n
d
i
fferen
t
TerraSAR-X im
ag
es an
d
m
easu
r
ed
d
i
fferen
t
qu
ality
param
e
ters.
Ev
en
tho
ugh
t
h
e lee filter
rem
o
v
e
s
m
u
ch
sp
eck
le, it canno
t preserv
e
t
h
e edg
e
d
e
tails an
d
it
d
o
e
s no
t
m
a
i
n
t
a
i
n
t
h
e cor
r
el
at
i
on wi
t
h
i
nput
i
m
age. It
i
s
a
m
a
jor dr
awbac
k
i
n
st
at
i
s
t
i
cal
fi
lt
ers. The t
r
ans
f
o
r
m
dom
ai
n
t
echni
q
u
es c
o
nve
rt
t
h
e SAR
im
age dat
a
int
o
di
ffe
rent
f
r
eq
ue
ncy
ban
d
i
n
whi
c
h t
h
e
si
gnal
an
d n
o
i
se are
separate
d. T
h
e
elimination of noise is
sim
p
le at that
m
o
m
e
nt. Because
of
this particula
r
r
eason the tra
n
sform
b
a
sed
tech
n
i
qu
es will p
r
o
d
u
ce sign
ifican
t
p
eak
sign
al
to
n
o
i
se ratios.
W
h
ereas the latest b
l
o
c
k
b
a
sed
techniques like BM3D and CS3D ar
e p
r
eservi
ng t
h
e e
dge
det
a
i
l
s
and su
pp
ress t
h
e speckl
e
bet
t
e
r, b
u
t
pr
o
pose
d
m
e
t
hod
p
r
ese
r
ves
e
dge
s a
n
d
su
p
p
r
e
ss t
h
e
spec
kl
e
m
u
ch bet
t
e
r t
h
an B
M
3
D
a
n
d
C
S
3D
t
ech
ni
q
u
e
s.
It
i
s
evi
d
e
n
t
l
y
obs
er
ved t
h
at
t
h
e p
r
o
p
o
se
d
des
p
eckl
i
n
g t
e
chni
que
pe
rf
or
m
s
bet
t
e
r t
h
an ot
he
r l
a
t
e
st
m
e
thods
in te
rm
s of quality factors like
E
N
L, SS
I, CC, E
S
I a
n
d PS
NR.
REFERE
NC
ES
[1]
Lee J.S.
and
E. Pottier
,
Pol
a
rim
e
t
r
ic R
a
dar
Im
ag
in
g from
Basics
to
Applications, C
R
C Press, 2009.
[2]
Lee J
.
S
.,
L. J
u
rk
evich
,
P
.
Dewa
e
l
e, P
.
W
a
m
b
acq,
and A. Oos
t
erl
i
n
ck, S
p
eck
le f
ilt
ering of s
y
nthe
ti
c aper
ture r
a
dar
images: A r
e
view," Remote Sen
s
ing Re
views, V
o
l. 8
,
No
. 4
,
313
-340, 1994
.
[3]
Frost V.S., J.A.
Stiles, K
.
S. Shanm
ugan, and J
.
C. Holt
zm
an,
A m
odel for ra
dar im
ages and
its appl
ic
ation
t
o
adapt
i
ve d
i
git
a
l
filte
ring of m
u
lt
i
p
lic
ativ
e noise"
,
IEEE
Transa
ctio
ns on P
a
ttern
Anal
y
s
is and Ma
ch
ine Int
e
ll
igen
ce
,
Vol. 4
,
No. 2, 15
7- 166, Mar. 198
2.
[4]
Lopes A, Nezr
y
E, Tou
z
i R, and
Laur
H, “
M
axim
um
a posteriori speckl
e
filt
er
ing
and first order texture models in
SAR images”,
in
Proc. IEEE In
t.
Geosci.
R
e
mote
Sens. S
y
mp, vol. 3, pp. 2409-241
2, 1990
.
[5]
Gagnon L and J
ouan A, “Speckle filtering o
f
SAR images
-A comparative stud
y between
complex-wavelet-b
a
sed
and stand
a
rd f
ilters”, in Proc. SP
IE, pp
. 80-9
1
, 19
97.
[6]
Argenti.F
and A
l
parone L, “Speckle
removal fro
m SAR images in the undeci
mated wavelet domain”, I
EEE Tr
an
s.
Geosci. R
e
mote
Sens. vol. 40,
no
. 11
, pp
. 2363-2
374, Nov. 2002.
Evaluation Warning : The document was created with Spire.PDF for Python.
I
S
SN
:
2
088
-87
08
I
J
ECE Vo
l. 5
,
N
o
. 5
,
O
c
tob
e
r
20
15
:
984
–
9
91
99
1
[7]
Zhang L, Dong
W, Zhang D, an
d Shi G,
“Two-stage
image d
e
no
ising b
y
prin
cip
a
l componen
t
an
aly
s
is with
local
pixel grouping”,
Pattern
Recog
.
,
vol. 43
, no
. 4
,
pp
. 1531-1549
, Ap
r. 2010
.
[8]
Coupé P, Helli
er P, Kervrann C, and
Barillo
t C,
“Bay
esi
a
n non l
o
cal m
eans-based speckle fi
lter
i
ng”, in Proc. 5
t
h
IEEE Int. S
y
mp. Biomed. Im
agin
g, pp
. 1291-129
4, May
2008.
[9]
Dabov K, Foi A, Katkovnik V, and Egia
zarian K
,
“Image denoising b
y
spar
se 3D transform-doma
i
n collaborative
filtering”,
I
EEE
Trans. Im
age
Pr
ocess.,
vol. 16, n
o
. 8
,
pp
. 2080-2
095, Aug. 2007.
[10]
Parrilli S., M. P
oderico
, C.V. A
ngelino
,
and L
.
Verdo
liva, A nonlocal SAR im
age denoising
alg
o
rithm
based on
llmmse wavelet shrinkage", IE
EE Transactions on Geoscience an
d Remote
Sensing, Vol. 50,
No. 2, 606-616, Feb.
2012.
[11]
Goodman J.W,
“So
m
e fundamental proper
tie
s o
f
speckle”, Journal of the Optical
Society
of A
m
erica, vo
l. 66,
no.
11, pp
. 1145-11
50, 1976
.
[12]
Lee J
.
S, Grunes
M.R, and M
a
n
go S.A,
“Speckle reduction
in
multipolariz
atio
n, multifrequenc
y S
A
R im
ager
y”,
IEEE Tr
ansactio
ns on
Geoscience and
Remote Sensing,
vol. 29
, n
o
. 4
,
pp
. 535-54
4, 1991
.
[13]
J.
S.
Le
e,
M.
R.
Grune
s,
D.
L.
Schule
r
,
E. Pottier, and L. Ferro-F
amil, “Scatte
rin
g
-model-based speckle filtering
of
Polarimetric SAR data”, IEEE
Transactions on G
e
oscien
ce
and
Remote Sensing,
vol. 44
, no
. 1
,
pp
. 176-
187, 2006
.
[14]
Chang S.G, Yu
B, and Vette
rli
M, “
S
patial
l
y
adapt
i
ve wav
e
le
t thre
shold
i
ng with
contex
t modeling for
image
denoising”, I
E
EE Tr
ans. Imag
e
Proce
ssing, vol.
9, pp
. 1522–153
1, Sept. 2000
.
[15]
Sveinsson J.R and Benediktss
on J.A, “
S
peckle reduct
i
on and en
hancem
en
t of SAR im
ages using m
u
lti wavelet
s
and ad
aptiv
e th
r
e
sholding”, in
Proc. SPIE Conf
. I
m
age and Si
gn
al Processing for
Remote Se
nsing
V, S. B. Serpico
,
Ed:
EUROPTO Series, vo
l.
3871
, pp
. 239–250
., 1
999.
[16]
Y. Murali Mohan Babu, M.V
.
Subraman
y
a
m &
M.
N.
Giriprasad “PCA base
d image deno
ising",
Signal & Imag
e
Processing: An I
n
ternational Jour
nal (SIP
IJ) Vol.
3, No. 2, 236-24
4, April 2012.
[17]
Iqbal M., J. Che
n
W
.
Yang, P. W
a
ng,
and B. Sun, "SAR im
age despeckling b
y
selec
tive 3D fil
t
e
ring of m
u
ltipl
e
compressive reconstructed
imag
es,"
Progress In
Electromagnetics Research
, Vol. 134, No. 12
, 20
9-226, 2013
.
Evaluation Warning : The document was created with Spire.PDF for Python.