Indonesi
an
Journa
l
of El
ect
ri
cal Engineer
ing
an
d
Comp
ut
er
Scie
nce
Vo
l.
24
,
No.
1
,
Octo
be
r
2021
,
pp.
14
4
~
156
IS
S
N: 25
02
-
4752, DO
I: 10
.11
591/ijeecs
.v
24
.i
1
.
pp
14
4
-
15
6
144
Journ
al h
om
e
page
:
http:
//
ij
eecs.i
aesc
or
e.c
om
A
comp
utation
al e
xperimental
of
noise sup
pressin
g techni
qu
e
stand
on hard d
ecision
threshold
d
issimi
larity
Vo
r
apoj P
ata
navijit
1
,
K
ornkam
ol Th
ak
ul
sukana
nt
2
1
Depa
rt
m
ent
of Electrical a
nd
E
l
ec
tron
ic
Engi
n
eering,
Vin
ce
n
t
M
arr
y
School
of E
ngine
er
ing,
As
sum
pti
on
Univer
sit
y
of
Tha
iland,
Bangk
ok
,
Th
ai
l
and
2
Depa
rtment
of
Mana
gement Inf
orm
at
ion
S
y
s
te
m
s,
Marti
n
de
Tou
rs School
of
Ma
nage
m
ent
and E
conomics,
As
sum
pti
on
Univer
sit
y
of
Th
ai
l
and, Ba
ngkok
,
Thaila
nd
Art
ic
le
In
f
o
ABSTR
A
CT
Art
ic
le
history:
Re
cei
ved
Ju
n
2
,
20
21
Re
vised
A
ug
7
,
20
21
Accepte
d
Aug
11
,
2021
Due
to
the
ext
r
e
m
e
insiste
nce
fo
r
digi
t
al
imag
e
p
roc
essing,
pl
enti
ful
m
oder
n
noise
suppress
ing
te
chni
qu
es
are
embodied
o
f
dissim
il
ari
t
y
proc
ess
and
suppress
ing
proc
ess.
One
of
th
e
ext
r
eme
c
apabili
t
y
d
issim
il
ar
ity
is
har
d
dec
ision
thr
esh
old
(HD
T)
dissimi
la
ri
t
y
,
which
h
a
s
bee
n
rec
en
tly
dec
l
are
d
in
2012,
for
suppr
essing
the
impu
lsive
nois
y
phot
ogra
phs
thus
th
e
computer
expe
riment
al
sta
te
m
ent
attempts
to
inve
stig
at
e
t
he
ca
p
abi
l
ity
of
the
nois
e
suppress
ing
te
ch
n
ique
that
is
stand
on
HD
T
diss
imila
rity
for
th
e
proc
essed
photogra
phs,
which
are
co
rrupt
ed
b
y
fix
ed
-
inte
nsit
y
impuls
e
n
oise
(FIIN
).
Thi
s
pape
r
pro
poses
the
noise
suppress
ing
te
chn
ique
stand
on
HD
T
dissim
il
ari
t
y
fo
r
FIIN
.
There
are
3
primar
y
cont
r
ibut
ions
o
f
thi
s
pape
r.
Th
e
first
cont
r
ibut
ion
is
the
st
at
ist
ical
ave
r
age
of
the
HD
T
dissim
il
arit
y
of
noise
-
fre
e
elem
ent
s,
which
are
comput
ed
from
ple
nti
fu
l
ground
-
trut
h
photogra
phs
b
y
var
y
ing
win
dow
size
for
the
best
HD
T
window
size
.
The
sec
ond
cont
ribution
is
t
he
statistical
av
era
ge
of
th
e
HD
T
dissim
il
arit
y
of
cor
rupt
ed
el
ements,
which
are
computed
from
ple
ntiful
cor
rupt
ed
phot
ogra
phs
b
y
var
y
ing
outl
i
er
d
ensity
for
th
e
be
st
HD
T
window
size
.
Th
e
final
c
ontri
buti
on
is
the
st
at
isti
cal
int
err
elati
on
o
f
the
ca
p
abi
l
ity
o
f
the
noise
s
uppre
ss
ing
te
chn
ique
and
har
d
consiste
n
t
of
HD
T
dissim
il
ari
t
y
are
inv
esti
gated
b
y
var
y
ing the
outli
er
dense
n
ess for the
b
est
HD
T
h
a
rd
consiste
n
ce.
Ke
yw
or
ds:
Digital
im
age p
r
ocessi
ng
Fixed
-
inte
ns
it
y im
pu
lse
no
ise
Hard
decisi
on t
hr
es
hold
dissi
m
il
arit
y
No
ise
sup
pr
ess
ing
tec
hn
i
ques
This
is an
open
acc
ess arti
cl
e
un
der
the
CC
B
Y
-
SA
l
ic
ense
.
Corres
pond
in
g
Aut
h
or
:
Vora
poj
Pata
na
vij
it
Dep
a
rtem
ent o
f
Ele
ct
ri
cal
and
Elec
tro
nic E
ngi
nee
rin
g
Vinc
e
nt Mar
ry Scho
ol of E
ng
ineerin
g
,
As
sum
pt
ion
Un
i
versi
ty
o
f
T
haila
nd
PK
E Bl
dg., 2
nd Fl
r.
,
88 M
oo
8
Ba
ng N
a
-
Tra
d Km
. 2
6, Ba
ngsa
othon
g,
Ba
ngkok
T
haila
nd
Em
a
il
: patanav
ijit
@yahoo.c
om
, p
at
anav
ijit
@g
m
ai
l.co
m
1.
LIT
URATU
R
E REVIE
W
Re
gu
la
rly
,
the
com
pe
te
nce
of
s
ophisti
cat
ed
im
age
pr
oce
ssing
te
c
hn
i
ques
[1
]
-
[
4],
s
uc
h
as
sup
e
r
reso
l
ution
[
5],
re
m
ote
sensing
[
6],
an
d
m
edical
i
m
agi
ng
[
7
]
-
[9]
,
ar
e
def
init
el
y
su
sce
ptible
fro
m
no
ise
there
upon
no
i
se
sup
pr
essi
ng
te
ch
nique
[10
]
-
[
20]
are
bec
om
e
an
irresist
ible
m
o
m
e
nto
us
proces
s.
The
or
et
ic
al
ly
,
the
noise
sup
pressi
ng
te
ch
ni
qu
e
regularly
const
ru
ct
s
t
he
undesira
ble
ef
f
ect
su
ch
as
bl
urrin
g
eff
ect
or
detai
l
losin
g
t
her
e
up
on
the
f
undam
ental
intenti
on
of
no
ise
sup
pressi
ng
te
c
hn
i
que
is
for
c
once
al
ing
no
ise
from
n
oi
sy photo
gr
ap
h wh
e
reas
protec
ti
ng
detai
l. Ord
inaril
y, im
pu
lse
noise [
21
]
-
[
29]
h
as
a great
i
m
pac
t
to
overall
qu
al
it
y
of
the
rec
orde
d
photog
ra
ph
due
to
t
he
fact
that
the
i
m
pu
lse
no
ise
m
od
ifie
s
the
corrupted
pix
el
el
em
ent
with
irre
gula
r
intensit
y.
N
at
ur
al
ly
,
phot
ogra
phs
are
c
orr
up
te
d
by
the
i
m
pu
lse
noise
from
plentif
ul
reas
ons
f
or
i
ns
ta
nt
transm
issi
on
or
receive
r
fail
ur
e
,
overl
oad
of
t
he
ci
rcu
it
sign
al
or
et
c.
Fr
om
m
at
he
m
at
ic
a
lly
analy
ti
cal
per
sp
ect
ive
,
the
im
pu
lse
no
ise
can
be
m
od
el
ed
into
tw
o
ki
nds:
rand
om
-
intensit
y
Evaluation Warning : The document was created with Spire.PDF for Python.
Ind
on
esi
a
n
J
E
le
c Eng &
Co
m
p
Sci
IS
S
N:
25
02
-
4752
A com
pu
t
ational ex
pe
rime
ntal
investi
ga
ti
on
of noise s
uppre
ss
ing t
ech
niqu
e sta
nd on…
(
Vorapoj P
atan
aviji
t
)
145
im
pu
lse
n
oise (R
II
N
),
wh
ic
h
can b
e v
a
ri
ed fr
om
“0
” to “2
55” an
d
fixe
d
-
i
nt
ensity
i
m
pu
lse
n
oise (
F
IIN
),
wh
ic
h
can
be
ei
ther
“0”
or
“2
55.
Fo
r
c
onceal
ing
an
i
m
pu
l
sive
no
ise
,
sp
eci
fic
al
ly
FI
IN
,
the
cl
assic
m
edian
filt
er
(MF),
w
hic
h
a
re
ori
gin
at
e
d
by
Pr
at
t
[21]
in
1975,
is
rea
dily
i
m
ple
m
ented
and
has
a
n
ac
ceptable
c
om
petence
there
upon
MF
has
bec
om
e
the
well
-
kn
own
noise
sup
pressi
ng
te
c
hn
i
que
ne
ver
t
heles
s
no
ise
s
uppr
essing
te
chn
iq
ue
re
gula
rly
con
str
uct
s
the
undesira
ble
ef
fect
su
c
h
as
gr
eat
blurrin
g
e
ff
ect
or
gr
eat
detai
l
losi
ng
because
the
MF
deals
with
bo
t
h
noisy
pix
el
el
em
ents
an
d
aut
he
ntic
pix
el
el
em
e
nts.
Lat
er
,
the
noise
su
pp
ressin
g
te
chn
i
qu
e
sta
nd
on
MF
a
nd
ad
aptive
wi
ndow
dim
ension
f
or
FI
I
N
was
ori
gi
nated
by
Hw
a
ng
a
nd
Hadda
d
[22]
i
n
1994.
A
fter
ward,
the
noi
se
sup
pr
es
sin
g
te
ch
nique
sta
nd
on
MF
a
nd
detai
l
-
preser
ving
regulariz
at
io
n
for
FII
N
was
or
i
gin
at
ed
by
Chan
et
al.
[
23]
in
20
05.
S
ucceedi
ng,
the
no
ise
s
uppres
sing
te
chn
iq
ue
sta
nd on s
ta
ti
sti
cal
d
et
ect
ion
for
R
IIN
wa
s
origin
at
ed
by
D
ong
e
t al
.
[
24
]
in 2
007. Be
hi
nd,
the
n
oise
su
pp
ressin
g
te
chn
i
qu
e
sta
nd
on
pro
bab
il
it
y
existe
nce
det
ect
ion
f
or
F
IIN
was
or
i
gin
a
te
d
by
A
wad
[25]
in
2018.
Subse
quently
,
the
no
ise
suppressi
ng
te
ch
nique
s
ta
nd
on
in
verse
distance
w
ei
gh
te
d
inter
pola
ti
on
(D
BI
D
WI
)
for
FI
I
N
was
i
nves
ti
gated
by
Pata
nav
i
j
it
[
26
]
i
n
2019. N
e
xt,
th
e
noise
s
uppres
sing
te
ch
nique stand
on
i
nterpolat
io
n
schem
e
fo
r
FI
I
N
at
high
den
sit
y
was
or
i
gin
at
ed
by
Kishore
ba
bu
et
al.
[27]
in
2019
.
Ther
ea
fter
,
the
no
ise
detect
io
n
te
ch
niques
st
and
on
sta
ti
stical
analy
sis
schem
e
for
RI
IN
was
in
vestigat
e
d
by
Pata
nav
i
j
it
and
Thakulsu
ka
na
nt
[28]
in
2019.
Lat
er,
the
noise
suppressi
ng
te
ch
nique
s
ta
nd
on
m
ulti
-
filt
ers
was
in
vestigat
e
d
by
Abd
urraz
zaq
et
al
.
[29] in 20
19. A
s a
re
su
lt
s,
ple
ntifu
l
no
ise
s
uppressi
ng
t
ec
hn
i
qu
e
s [21
]
-
[
32]
ha
ve
been
m
od
ifie
d
f
ro
m
cl
assic
MF
f
or
only
co
nceali
ng
im
pu
lsi
ve
noise
pi
xels
an
d
un
-
to
uc
hing
noise
-
fr
ee
pix
el
s
with
bette
r
c
om
petence
there
up
on
t
he
hi
ndra
nc
e
of
posit
ion
i
ng
i
den
ti
ficat
ion
of
co
rru
pted
pi
xel
el
e
m
ents
is
to
i
den
ti
fy
the
c
orrupted p
ixel
el
e
m
ent
as
no
isy
and
i
den
ti
fy
th
e
no
ise
-
f
ree
pi
xel
el
e
m
ent
as
no
ise
-
fr
ee.
I
n
orde
r
t
o
se
pa
rate
the
corrupted
area
from
no
ise
-
fr
e
e
pi
xel
el
em
ent,
the
at
tri
bu
te
of
noise
-
fr
ee
pi
xels,
wh
ic
h
a
re
in
s
m
oo
th
area
(or
alm
os
t
al
l
pix
el
intensit
y
in
t
his
area
are
sli
gh
tl
y
eq
ual)
or
in
c
orner
area
(or
al
l
pix
el
intensit
y
in
this
area
are
separ
at
ed
int
o
two
le
vels)
ne
ver
t
heless
the
alm
os
t
area
of
pix
el
s
is
sm
oo
th
area
bu
t
only
few
a
rea
of
pix
el
s
is
corner
area
.
Iron
ic
al
ly
,
the
noisy
pix
e
ls
are
gr
eat
heter
oge
neity
,
wh
ic
h
c
an
be
diffuse
d
f
r
om
0
to
255
due
to
i
m
pu
lsi
ve
no
i
se
there
upon
noisy
area
has
great
heter
ogen
ei
ty
of
pixe
l
in
te
ns
it
y
than
the
heter
ogeneit
y
of
nois
e
-
f
ree
area.
As
a
resu
lt
s,
the
hard
de
ci
sion
t
hr
es
hold
(
HDT
)
dissim
il
ar
ity
idea
was ori
gin
at
ed
by
A
wa
d
[
33]
in 201
2
f
or posi
ti
on
in
g
ide
ntifi
cat
ion
of
co
rru
pted pixel el
e
m
ents and,
la
te
r,
was
beco
m
e
on
e
of
the
great
c
om
petence
diss
i
m
i
la
rity
fo
r
noisy
/no
ise
-
f
ree
posit
ion
i
ng
identific
at
io
n
t
hat
is
inco
rpor
at
e
d
in
sop
histi
cat
ed
noise
s
uppressi
ng
te
c
hn
i
que.
As
co
ns
e
quence
,
the
H
DT
dissim
il
ari
ty
was
inv
est
igate
d durin
g 0
-
100%
densit
y by [
34
]
-
[
35]
in 2
020.
2.
THE
FUN
DAMENT
AL TH
EORY OF
H
DT DISSI
MI
LARIT
Y
The
par
ti
ti
on
com
pr
ehe
ns
i
vely
pr
e
faces
the
noise
s
uppressi
ng
te
chn
i
qu
e
sta
nd
on
both
t
he
po
sit
io
ning
i
de
ntific
at
ion
te
c
hniq
ue
us
in
g
H
DT
dissim
il
arity,
w
hich
is
as
s
iduousl
y
el
ucidated
i
n
Sect
io
n
2.1,
and noise
resto
rin
g
te
ch
nique
us
in
g
cl
assic
al
MF fil
te
r,
w
hich
is assi
duousl
y el
ucidated
in
Sect
ion
2.2.
2
.
1.
P
os
iti
on
i
ng
ide
nt
ific
at
i
on
of
im
pulse
no
ise
s
tand
on HDT
dissi
mi
larit
y
Du
e
to
the
al
ge
br
a
m
od
el
of
no
isy
im
ages,
wh
ic
h
are
co
rrup
te
d
by
fix
va
lue
i
m
pu
lsi
ve
ou
tl
ie
r,
the
corrupted
p
i
xel
ele
m
ent (
,
x
i
j
) c
an
b
e al
gebraica
ll
y reveale
d
as:
0
1
if
,
ot
he
r
wis
e
i
no
i
sy
i
i
ori
gi
na
l
x
x
C
x
i
j
x
wh
ere
n
o
isy
x
and
original
x
is no
isy
a
nd
or
i
gi
nal p
i
xels
(1)
w
he
re
i
C
is t
he
th
i
noisy
area a
nd
0
i
is t
he n
um
ber
of
no
isy
a
rea.
The
H
DT
dissi
m
il
arity
,
d
i
j
[33]
c
an
be
al
gebrai
cal
ly
rev
eal
ed
as
f
ollo
wing
te
chn
ic
al
e
xpres
sion
s
wh
e
re
t
he
phot
ogra
ph
siz
e
is
def
i
ned
as
be
nm
,
the
wind
ow
si
ze
is
de
fine
d
a
s
kl
,
an
d
th
e
batch
siz
e
is
def
i
ned a
s
nm
.
,
,
,
kl
s
k
t
l
d
i
j
y
s
t
x
i
j
w
her
e
0
.
5
1
kk
and
0
.
5
1
ll
(2)
,
,
,
kl
s
k
t
l
d
i
j
x
i
j
y
s
t
k
l
(3)
Evaluation Warning : The document was created with Spire.PDF for Python.
IS
S
N
:
2502
-
4752
Ind
on
esi
a
n
J
E
le
c Eng &
Co
m
p
Sci,
Vo
l.
24
, N
o.
1
,
Oct
ober
20
21
:
144
-
15
6
146
By
var
yi
ng
the
intensit
y
of
pi
xel
el
e
m
ent
fr
om
0”
to
“25
5”
(
,
0
,
2
5
5
x
i
j
),
the
sim
p
lif
ic
at
ion
of
the up
per te
ch
nical
expres
sio
ns
ca
n be al
gebraica
ll
y reveale
d
as
:
,
2
2
5
5
2
kl
s
k
t
l
y
y
s
t
k
l
b
a
(4)
The H
DT dissi
m
il
arity
,
d
i
j
at
,
x
i
j
is
def
i
ned as
:
,
2
5
5
2
,
d
i
j
x
i
j
(5)
The u
ncer
ta
in
o
f
the a
ver
a
ge
of d
issi
m
il
ariti
es
,
d
i
j
can
be
al
ge
br
ai
cal
ly
r
e
vea
le
d
,
0
,
1
2
7
.
5
d
i
j
(6)
and
,
6
3
.
7
5
d
i
j
(7)
The
a
ver
a
ge
of
d
issi
m
i
la
riti
es
c
D
of the
pro
ce
ss
ed photo
grap
h ca
n be a
lg
eb
raical
ly
r
eveale
d
,
11
,
1
1
n
k
m
l
cc
j
k
i
l
D
d
i
j
n
k
m
l
(
8)
As
a
res
ult
of
the
com
pu
te
r
com
pu
ta
ti
on
resu
lt
s
[
33
]
,
t
he
al
te
ring
of
regulariz
ed
offset
can
be
al
gebraica
ll
y revealed,
c
D
T
h
D
(9)
and
2
c
T
h
D
D
(10)
The
c
om
pr
ehe
ns
ive
pr
ocessi
ng
of
posit
io
ning
i
den
ti
fic
at
ion
of
im
pu
lse
noise
sta
nd
on
HD
T
dissim
il
arit
y can
be
al
ge
brai
cal
ly
r
eveale
d
a
s
Figure
1.
As
a
res
ult
of
po
sit
io
ning
identific
at
ion
processin
g
f
or
al
l
pix
el
el
e
m
ents
in
t
he
co
rru
pte
d
photog
raph,
th
e
cor
r
upte
d
pi
xel
el
e
m
ents
c
an
be
se
par
at
e
d
to
be
noise
-
fr
ee
pi
xel
el
em
ents
or
nois
y
pix
el
el
e
m
ents as,
if
,
,
othe
r
w
is
e
no
i
sy
origi
na
l
x
d
i
j
T
h
x
i
j
x
(11)
2.2
.
Restor
ati
on
of impul
se
no
ise
stand
on medi
an
filte
r
As
a
resu
lt
of
posit
ion
in
g
identific
at
ion
processi
ng
f
or
al
l
pix
el
el
em
ents
in
the
cor
r
upte
d
photog
raph,
th
e
cor
r
upte
d
pi
xel
el
e
m
ents
c
an
be
se
par
at
e
d
to
be
noise
-
fr
ee
pi
xel
el
em
ents
or
nois
y
pix
el
el
e
m
ents ther
ef
or
e
only
grou
p of n
oisy pi
xel
el
e
m
ents are
s
uppresse
d
as,
,
,
,
,
,
,
0
,
0
M
e
d
i
j
m
e
d
i
a
n
w
i
s
j
t
x
i
s
j
t
k
s
t
k
s
t
(12)
Evaluation Warning : The document was created with Spire.PDF for Python.
Ind
on
esi
a
n
J
E
le
c Eng &
Co
m
p
Sci
IS
S
N:
25
02
-
4752
A com
pu
t
ational ex
pe
rime
ntal
investi
ga
ti
on
of noise s
uppre
ss
ing t
ech
niqu
e sta
nd on…
(
Vorapoj P
atan
aviji
t
)
147
Figure
1.
The
c
o
m
pr
ehe
ns
ive
processi
ng of
posit
ion
i
ng ide
nt
ific
at
ion
of im
pu
lse
noise sta
nd on H
DT
dissim
il
arit
y
Evaluation Warning : The document was created with Spire.PDF for Python.
IS
S
N
:
2502
-
4752
Ind
on
esi
a
n
J
E
le
c Eng &
Co
m
p
Sci,
Vo
l.
24
, N
o.
1
,
Oct
ober
20
21
:
144
-
15
6
148
3.
THE
LE
SS
ON
OF
T
HE
COMP
UTER
CA
L
CU
L
ATIO
N
OF
NOI
S
E
SU
PP
RES
SING
TE
CHN
I
QUE
STA
ND O
N HDT
DISSIM
ILARI
T
Y
T
h
e
p
a
r
t
i
t
i
o
n
c
om
p
r
e
h
e
n
s
i
v
e
l
y
p
r
e
f
a
c
e
s
t
h
e
c
om
p
u
t
e
r
c
a
l
c
ul
a
t
i
o
n
o
f
n
o
i
s
e
s
u
p
p
r
e
s
s
i
n
g
t
e
c
h
n
i
q
u
e
s
t
a
n
d
o
n
H
D
T
d
i
s
s
i
m
i
l
a
r
i
t
y
i
n
t
w
o
c
om
p
u
t
e
r
c
a
l
c
u
l
a
t
i
o
n
l
e
s
s
o
n
,
w
h
i
c
h
c
a
n
b
e
a
l
g
e
b
r
a
i
c
a
l
l
y
r
e
v
e
a
l
e
d
a
s
F
i
g
u
r
e
2
and
Figure
3
f
or
f
orci
ng
the
bib
li
ophile
the
ob
vi
ou
s
per
ce
ptio
n
on
t
he
c
om
puta
ti
on
process
of
noise
sup
pressi
ng
te
chn
iq
ue
sta
nd
on
HD
T
di
ssi
m
il
arity
in
ever
y
c
om
pu
ta
ti
on
ste
p.
Th
e
first
cal
cula
ti
on
le
sson
ca
n
be
al
gebraica
ll
y
rev
eal
ed
as
Fig
ur
e
2
w
he
re
t
he
proce
ssin
g
pi
xel
el
em
ent
,
ij
y
is
a
no
ise
-
f
ree
pi
xel
el
em
ents
that
is i
den
ti
fie
d
as
no
ise
-
f
ree
by im
pu
lse
n
oise i
den
ti
ficat
io
n process.
1
,
1
133
256
y
i
j
1,
133
256
y
i
j
1
,
1
255
256
y
i
j
,
128
256
y
i
j
,1
0
256
y
i
j
1
,
1
120
256
y
i
j
1,
120
256
y
i
j
,1
128
256
y
i
j
1
,
1
0
256
y
i
j
1
,
is a
n
oise
-
f
r
e
e
y
i
j
1
,
2
133
256
y
i
j
,2
128
256
y
i
j
1
,
2
120
256
y
i
j
1
,
3
255
256
y
i
j
,3
128
256
y
i
j
1
,
3
120
256
y
i
j
1
,
2
113
256
y
i
j
,2
255
256
y
i
j
1
,
2
120
256
y
i
j
1
,
3
255
256
y
i
j
,3
0
256
y
i
j
1
,
3
125
256
y
i
j
2
,
1
125
256
y
i
j
2,
125
256
y
i
j
2
,
1
255
256
y
i
j
2
,
2
0
256
y
i
j
2
,
3
125
256
y
i
j
2
,
2
0
256
y
i
j
2
,
3
119
256
y
i
j
3
,
1
125
256
y
i
j
3,
125
256
y
i
j
3
,
1
125
256
y
i
j
3
,
2
125
256
y
i
j
3
,
3
125
256
y
i
j
3
,
2
123
256
y
i
j
3
,
3
119
256
y
i
j
2
,
1
116
256
y
i
j
2,
0
256
y
i
j
2
,
1
0
256
y
i
j
1
,
2
116
256
y
i
j
2
,
3
116
256
y
i
j
2
,
2
115
256
y
i
j
2
,
3
112
256
y
i
j
3
,
1
123
256
y
i
j
3,
123
256
y
i
j
3
,
1
123
256
y
i
j
1
,
2
123
256
y
i
j
3
,
3
123
256
y
i
j
3
,
2
120
256
y
i
j
3
,
3
116
256
y
i
j
1
,
1
0.5195
y
i
j
1,
0.5195
y
i
j
1
,
1
0.0000
y
i
j
,
0.50
00
y
i
j
,1
0.00
00
y
i
j
1
,
1
0.4688
y
i
j
1,
0.46
88
y
i
j
,1
0.9961
y
i
j
1
,
1
0.9961
y
i
j
,
d
i
j
1
,
2
0.5195
y
i
j
,2
0.50
00
y
i
j
1
,
2
0.4688
y
i
j
1
,
3
0.9961
y
i
j
,3
0.5000
y
i
j
1
,
3
0.4688
y
i
j
1
,
2
0.4414
y
i
j
,2
0.99
61
y
i
j
1
,
2
0.4688
y
i
j
1
,
3
0.9961
y
i
j
,3
0.0000
y
i
j
1
,
3
0.4883
y
i
j
2
,
1
0.4883
y
i
j
2,
0.48
83
y
i
j
2
,
1
0.0000
y
i
j
2
,
2
0.0000
y
i
j
2
,
3
0.4883
y
i
j
2
,
2
0.0000
y
i
j
2
,
3
0.4648
y
i
j
3
,
1
0.4883
y
i
j
3,
0.4883
y
i
j
3
,
1
0.0000
y
i
j
3
,
2
0.4883
y
i
j
3
,
3
0.4883
y
i
j
3
,
2
0.4805
y
i
j
3
,
3
0.4648
y
i
j
2
,
1
0.4531
y
i
j
2,
0.00
00
y
i
j
2
,
1
0.4688
y
i
j
1
,
2
0.4531
y
i
j
2
,
3
0.4531
y
i
j
2
,
2
0.4492
y
i
j
2
,
3
0.4375
y
i
j
3
,
1
0.4805
y
i
j
3,
0.4805
y
i
j
3
,
1
0.9961
y
i
j
1
,
2
0.4805
y
i
j
3
,
3
0.4805
y
i
j
3
,
2
0.4688
y
i
j
3
,
3
0.4531
y
i
j
1
,
is a
n
oise
-
f
r
e
e
y
i
j
1
,
1
0.0195
y
i
j
1,
0.019
y
i
j
1
,
1
0.4961
y
i
j
,
0.00
00
y
i
j
,1
0.50
00
y
i
j
1
,
1
0.0313
y
i
j
1,
0.03
13
y
i
j
,1
0.0000
y
i
j
1
,
1
0.5000
y
i
j
1
,
2
0.0195
y
i
j
,2
0.00
00
y
i
j
1
,
2
0.0313
y
i
j
1
,
3
0.4961
y
i
j
,3
0.0000
y
i
j
1
,
3
0.0313
y
i
j
1
,
2
0.0586
y
i
j
,2
0.49
61
y
i
j
1
,
2
0.0313
y
i
j
1
,
3
0.4961
y
i
j
,3
0.5000
y
i
j
1
,
3
0.0117
y
i
j
2
,
1
0.0117
y
i
j
2,
0.01
17
y
i
j
2
,
1
0.4961
y
i
j
2
,
2
0.5000
y
i
j
2
,
3
0.0117
y
i
j
2
,
2
0.5000
y
i
j
2
,
3
0.0352
y
i
j
3
,
1
0.0117
y
i
j
3,
0.0117
y
i
j
3
,
1
0.0117
y
i
j
3
,
2
0.0117
y
i
j
3
,
3
0.0117
y
i
j
3
,
2
0.0195
y
i
j
3
,
3
0.0352
y
i
j
2
,
1
0.0469
y
i
j
2,
0.50
00
y
i
j
2
,
1
0.5000
y
i
j
1
,
2
0.0469
y
i
j
2
,
3
0.0469
y
i
j
2
,
2
0.0508
y
i
j
2
,
3
0.0625
y
i
j
3
,
1
0.0195
y
i
j
3,
0.0195
y
i
j
3
,
1
0.0195
y
i
j
1
,
2
0.0195
y
i
j
3
,
3
0.0195
y
i
j
3
,
2
0.0313
y
i
j
3
,
3
0.0469
y
i
j
,,
,
,
1
2
a
nd
1
2
kl
s
k
t
l
y
i
j
y
k
l
d
i
j
k
k
l
l
kl
33
33
,,
,
0.14
33
33
st
y
i
j
y
k
l
d
i
j
,
0.1
922
,
is a
no
ise
l
e
ss pixe
l
N
oise
_D
e
te
c
ti
on
,
0
d
i
j
T
h
y
i
j
ij
IF
T
HE
N
an
d
Figure
2
.
The
fi
rst calc
ulati
on lesso
n of n
ois
e sup
pr
essi
ng techn
i
qu
e
stan
d o
n HD
T
d
issi
m
il
arity
(wher
e
the
pro
cessi
ng p
i
xel e
lem
ent
,
ij
y
is a n
oi
se
-
f
ree
pix
el
el
e
m
ents)
Lat
er,
the
cal
culat
ion
le
ss
on
can
be
al
ge
br
a
ic
al
ly
rev
eal
ed
as
Fig
ur
e
2
w
her
e
t
he
proce
ssing
pi
xel
el
e
m
ent
,
ij
y
is
a
no
isy
pix
el
el
e
m
ents
that
is
identifie
d
as
noise
-
fr
ee
by
im
pu
lse
noise
ide
ntific
at
ion
pro
cess
and m
us
t be s
uppresse
d by im
pu
lse
noise
res
torati
on
proces
s.
Evaluation Warning : The document was created with Spire.PDF for Python.
Ind
on
esi
a
n
J
E
le
c Eng &
Co
m
p
Sci
IS
S
N:
25
02
-
4752
A com
pu
t
ational ex
pe
rime
ntal
investi
ga
ti
on
of noise s
uppre
ss
ing t
ech
niqu
e sta
nd on…
(
Vorapoj P
atan
aviji
t
)
149
1
,
1
125
256
y
i
j
1,
255
256
y
i
j
1
,
1
0
256
y
i
j
,
0
256
y
i
j
,1
133
256
y
i
j
1
,
1
255
256
y
i
j
1,
0
256
y
i
j
,1
255
256
y
i
j
1
,
1
255
256
y
i
j
1
,
is
a
n
im
p
u
l
si
v
e
n
o
is
e
y
i
j
1
,
2
125
256
y
i
j
,2
255
256
y
i
j
1
,
2
128
256
y
i
j
1
,
3
255
256
y
i
j
,3
255
256
y
i
j
1
,
3
128
256
y
i
j
1
,
2
125
256
y
i
j
,2
133
256
y
i
j
1
,
2
255
256
y
i
j
1
,
3
123
256
y
i
j
,3
113
256
y
i
j
1
,
3
111
256
y
i
j
2
,
1
125
256
y
i
j
2,
255
256
y
i
j
2
,
1
0
256
y
i
j
2
,
2
125
256
y
i
j
2
,
3
125
256
y
i
j
2
,
2
0
256
y
i
j
2
,
3
123
256
y
i
j
3
,
1
125
256
y
i
j
3,
255
256
y
i
j
3
,
1
0
256
y
i
j
3
,
2
125
256
y
i
j
3
,
3
125
256
y
i
j
3
,
2
0
256
y
i
j
3
,
3
123
256
y
i
j
2
,
1
120
256
y
i
j
2,
120
256
y
i
j
2
,
1
120
256
y
i
j
1
,
2
120
256
y
i
j
2
,
3
120
256
y
i
j
2
,
2
0
256
y
i
j
2
,
3
0
256
y
i
j
3
,
1
116
256
y
i
j
3,
116
256
y
i
j
3
,
1
255
256
y
i
j
1
,
2
116
256
y
i
j
3
,
3
116
256
y
i
j
3
,
2
116
256
y
i
j
3
,
3
255
256
y
i
j
1
,
1
0.4883
y
i
j
1,
0.9961
y
i
j
1
,
1
0.0000
y
i
j
,
0.00
00
y
i
j
,1
0.51
95
y
i
j
1
,
1
0.9961
y
i
j
1,
0.00
00
y
i
j
,1
0.9961
y
i
j
1
,
1
0.9961
y
i
j
,
d
i
j
1
,
2
0.9961
y
i
j
,2
0.99
61
y
i
j
1
,
2
0.5000
y
i
j
1
,
3
0.9961
y
i
j
,3
0.9961
y
i
j
1
,
3
0.5000
y
i
j
1
,
2
0.4883
y
i
j
,2
0.51
95
y
i
j
1
,
2
0.9961
y
i
j
1
,
3
0.4805
y
i
j
,3
0.4414
y
i
j
1
,
3
0.4336
y
i
j
2
,
1
0.4883
y
i
j
2,
0.99
61
y
i
j
2
,
1
0.0000
y
i
j
2
,
2
0.4883
y
i
j
2
,
3
0.4883
y
i
j
2
,
2
0.0000
y
i
j
2
,
3
0.4805
y
i
j
3
,
1
0.4883
y
i
j
3,
0.9961
y
i
j
3
,
1
0.0000
y
i
j
3
,
2
0.4883
y
i
j
3
,
3
0.4883
y
i
j
3
,
2
0.0000
y
i
j
3
,
3
0.4805
y
i
j
2
,
1
0.4688
y
i
j
2,
0.46
88
y
i
j
2
,
1
0.4688
y
i
j
1
,
2
0.4688
y
i
j
2
,
3
0.4688
y
i
j
2
,
2
0.0000
y
i
j
2
,
3
0.0000
y
i
j
3
,
1
0.4531
y
i
j
3,
0.4531
y
i
j
3
,
1
0.9961
y
i
j
1
,
2
0.4531
y
i
j
3
,
3
0.4531
y
i
j
3
,
2
0.4531
y
i
j
3
,
3
0.9961
y
i
j
1
,
is
a
n
im
p
u
l
si
v
e
n
o
is
e
y
i
j
1
,
1
0.4883
y
i
j
1,
0.9961
y
i
j
1
,
1
0.0000
y
i
j
,
0.00
00
y
i
j
,1
0.51
95
y
i
j
1
,
1
0.9961
y
i
j
1,
0.00
00
y
i
j
,1
0.9961
y
i
j
1
,
1
0.9961
y
i
j
1
,
2
0.9961
y
i
j
,2
0.99
61
y
i
j
1
,
2
0.5000
y
i
j
1
,
3
0.9961
y
i
j
,3
0.9961
y
i
j
1
,
3
0.5000
y
i
j
1
,
2
0.4883
y
i
j
,2
0.51
95
y
i
j
1
,
2
0.9961
y
i
j
1
,
3
0.4805
y
i
j
,3
0.4414
y
i
j
1
,
3
0.4336
y
i
j
2
,
1
0.4883
y
i
j
2,
0.99
61
y
i
j
2
,
1
0.0000
y
i
j
2
,
2
0.4883
y
i
j
2
,
3
0.4883
y
i
j
2
,
2
0.0000
y
i
j
2
,
3
0.4805
y
i
j
3
,
1
0.4883
y
i
j
3,
0.9961
y
i
j
3
,
1
0.0000
y
i
j
3
,
2
0.4883
y
i
j
3
,
3
0.4883
y
i
j
3
,
2
0.0000
y
i
j
3
,
3
0.4805
y
i
j
2
,
1
0.4688
y
i
j
2,
0.46
88
y
i
j
2
,
1
0.4688
y
i
j
1
,
2
0.4688
y
i
j
2
,
3
0.4688
y
i
j
2
,
2
0.0000
y
i
j
2
,
3
0.0000
y
i
j
3
,
1
0.4531
y
i
j
3,
0.4531
y
i
j
3
,
1
0.9961
y
i
j
1
,
2
0.4531
y
i
j
3
,
3
0.4531
y
i
j
3
,
2
0.4531
y
i
j
3
,
3
0.9961
y
i
j
,,
,
,
1
2
a
nd
1
2
kl
s
k
t
l
y
i
j
y
k
l
d
i
j
k
k
l
l
kl
33
33
,,
,
0.53
79
33
st
y
i
j
y
k
l
d
i
j
,
0.1922
,
i
s a
noi
sy
pi
xe
l
Nois
e
_De
t
e
c
t
i
on
,
1
d
i
j
T
h
y
i
j
ij
IF
T
H
E
N
an
d
Figure
3
.
The
fi
rst calc
ulati
on lesso
n of n
ois
e sup
pr
essi
ng techn
i
qu
e
stan
d o
n HD
T
d
issi
m
il
arity
(wher
e
the
pro
cessi
ng p
i
xel e
lem
ent
,
ij
y
is a n
o
i
sy pixel el
em
e
nts)
4.
RESU
LT
S
AND DI
SCUS
S
ION
In
this
pa
rtit
ion
of
c
om
pu
te
r
si
m
ulati
on
,
the
si
m
ulati
on
pr
ogram
is
the
M
ATL
AB,
w
hich
is
instal
le
d
and
e
xec
uted
on
plentif
ul
w
orkstat
ions
at
this
capa
bili
ty
par
ti
cula
rizat
ion
:
m
ai
n
proc
essor
i7
-
6700
HQ
a
nd
m
ai
n
execu
ti
ng
m
e
m
or
y
16
GB
RAM.
All
work
sta
ti
ons
execu
te
plentif
ul
gro
und
-
t
ru
t
h
phot
ogra
ph
s
(which
are
com
pounde
d
of
Ai
rp
la
ne
,
Ba
boon,
Girl
,
H
ou
se
,
Le
na,
Mob
il
e,
Pe
pper,
Pe
ntag
on
and
Re
s
olu
ti
on
)
with
bounti
fu
l
noise
f
re
quency.
4.1
.
The
com
put
er
simul
ati
on
c
orrela
tio
n
o
f
H
DT
dissi
mi
larity an
d
w
indow dime
nsion
Fr
om
the
resu
l
ts
of
com
pu
te
r
si
m
ulati
on
on
plentiful
ground
-
tr
uth
photogra
phs,
the
fir
st
sta
ti
st
ic
al
m
o
m
ent
and
the
seco
nd
st
at
ist
ic
al
m
o
m
ent
of
the
norm
alized
HDT
dissim
il
arity
of
al
l
gr
ou
nd
-
tr
uth
photo
grap
hs
(
wh
ic
h
are
the
no
ise
-
f
ree
phot
ogra
ph
s
)
are
0.0
394±0
.02
21,
0.
05
06±
0.027
5
an
d
0.0
585±
0.031
6
at
w
in
dow dim
ensio
n 3
x3, 5x5 a
nd 7x
7,
res
pe
ct
ively
.
Nex
t,
the
fir
st
sta
ti
sti
cal
m
o
ment
of
norm
al
i
zed
H
DT
dissi
m
il
arity
,
d
i
j
of
al
l
photog
raphs,
wh
ic
h
are
fl
uctuate
d
from
0%
to
90%
noise
f
re
quency
of
F
IIN
a
t
3x3,
5x
5
a
nd
7x7
co
uld
be
la
id
out
as
Fi
gure
4
,
Figure
5
a
nd
Figure
6,
res
pe
ct
ively
.
Fr
om
these
resu
lt
s
of
com
pu
te
r
si
m
ulati
on
,
the
norm
al
iz
ed
HD
T
dissim
il
arit
y
w
it
h
window d
im
ension
7x7
pro
vid
es
t
he
hi
ghest
norm
al
iz
e
d
ab
so
l
ute
dif
f
eren
t.
C
onseq
ue
ncely
,
the
no
ise
s
uppressi
ng
te
ch
nique
sta
nd
on
H
DT
dissim
il
ari
ty
pr
ocides
t
he
hig
he
st
peak
sign
al
to
noise
rati
o
(
PS
NR
)
.
Evaluation Warning : The document was created with Spire.PDF for Python.
IS
S
N
:
2502
-
4752
Ind
on
esi
a
n
J
E
le
c Eng &
Co
m
p
Sci,
Vo
l.
24
, N
o.
1
,
Oct
ober
20
21
:
144
-
15
6
150
Figure
4
.
The
c
om
pu
te
r
sim
ulati
on
c
orrelat
io
n of h
dt
dissim
il
arit
y at
d
i
m
ension
3x
3
a
nd
noise
den
s
ity
Figure
5
.
The
c
om
pu
te
r
sim
ulati
on
c
orrelat
io
n
of HDT
dissi
m
il
arity at di
m
ensio
n
5x5 an
d
noise
den
sit
y
Figure
6
.
The
c
om
pu
te
r
sim
ulati
on
c
orrelat
io
n
of HDT
dissi
m
il
arity at di
m
ensio
n
7x7 an
d
noise
den
sit
y
Evaluation Warning : The document was created with Spire.PDF for Python.
Ind
on
esi
a
n
J
E
le
c Eng &
Co
m
p
Sci
IS
S
N:
25
02
-
4752
A com
pu
t
ational ex
pe
rime
ntal
investi
ga
ti
on
of noise s
uppre
ss
ing t
ech
niqu
e sta
nd on…
(
Vorapoj P
atan
aviji
t
)
151
4.2
.
The
com
put
er
sim
ulat
i
on
c
orrela
tion
between
n
ois
e
suppres
sing
capabil
ity
and
ha
rd
co
nsiste
n
t
of
HDT
dissi
mi
l
arit
y
F
r
o
m
t
h
e
r
e
s
u
l
t
s
o
f
c
om
p
u
t
e
r
s
i
m
ul
a
t
i
o
n
o
n
f
o
u
r
g
r
o
u
n
d
-
t
r
u
t
h
p
h
o
t
o
g
r
a
p
h
s
,
w
h
i
c
h
a
r
e
c
om
p
o
u
n
d
e
d
o
f
G
i
r
l
,
L
e
n
a
,
A
i
r
p
l
a
n
e
a
n
d
P
e
p
pe
r
,
w
i
t
h
t
h
e
F
IIN
,
t
h
e
c
om
p
u
t
e
r
s
i
m
u
l
a
t
i
o
n
c
or
r
e
l
a
t
i
o
n
b
e
t
w
e
e
n
n
o
i
s
e
s
u
p
p
r
e
s
s
i
n
g
c
a
p
a
b
i
l
i
t
y
a
n
d
h
a
r
d
c
o
n
s
i
s
t
e
n
t
o
f
H
D
T
d
i
s
s
i
m
i
l
a
r
i
t
y
c
o
u
l
d
b
e
l
a
i
d
o
u
t
a
s
T
a
b
l
e
1
,
T
a
b
l
e
2
,
T
a
b
l
e
3
a
n
d
T
a
b
l
e
4
,
r
e
s
p
e
c
t
i
v
e
l
y
(
w
h
e
r
e
t
h
e
b
o
l
d
f
o
r
m
a
t
i
s
r
e
p
r
e
s
e
n
t
e
d
t
h
e
h
i
g
h
e
s
t
P
S
N
R
)
.
Fro
m
th
e
resu
lt
s
of
com
pu
te
r
sim
ulati
on
consum
m
ations
of
Girl
phot
ogra
ph
ic
,
t
he
HD
T
ha
rd
co
nsi
ste
nt
for
the
m
os
t
capab
il
ity
m
us
t
be
set
durin
g
0.025
-
0.3
75
r
oughtl
y.
By
us
i
ng
al
ge
br
ai
c
i
nvest
igati
on,
t
he
first
sta
ti
sti
cal
m
om
ent
and
the
sec
ond
sta
ti
sti
ca
l
m
o
m
ent
o
f
the
no
rm
al
iz
ed
HD
T
dissim
ilarity
,
wh
ic
h
are
cal
culat
ed
fr
om
these
com
pu
te
r
si
m
ulati
on
consum
m
ations
in
Table
1,
a
re
0.241
7±0.1
392.
F
ro
m
the
resu
lt
s
of
c
om
pu
te
r
si
m
ulatio
n
co
nsum
m
a
ti
on
s
of
Lena
phot
ogra
ph
ic
,
the
H
DT
hard
co
ns
ist
ent
fo
r
the
m
os
t
c
apab
il
it
y
m
us
t
be
set
du
ri
ng
0.0
25
-
0.350
r
ough
tl
y.
By
us
ing
al
ge
br
ai
c
in
vestiga
ti
on
,
the
fir
st
sta
ti
st
ic
al
m
o
m
ent
and
the
second
sta
ti
sti
cal
m
o
m
ent
of
the
norm
al
iz
ed
HD
T d
issi
m
il
arity,
wh
ic
h
are ca
lc
ulate
d
f
ro
m
t
hese
c
om
pu
te
r
si
m
ulati
on
co
nsum
m
a
ti
on
s
in Tab
le
2,
a
r
e
0.2
278±0.132
0.
Table
1.
T
he
c
om
pu
te
r
sim
ulati
on
c
orrelat
io
n of n
oise s
uppressi
ng ca
pa
bili
ty
an
d har
d
c
onsist
ent
of
HD
T
dissim
il
a
rity
(G
irl)
Hard Co
n
sis
ten
t
PSNR (dB
)
10
20
30
40
50
60
70
80
90
0
.02
5
3
4
.69
3
1
.90
2
7
.69
2
3
.37
2
0
.17
1
8
.45
1
6
.73
1
6
.11
1
2
.64
0
.05
0
3
0
.71
3
1
.27
2
7
.64
2
3
.37
2
0
.17
1
8
.45
1
6
.73
1
6
.11
1
2
.64
0
.07
5
2
8
.12
2
9
.43
2
7
.45
2
3
.39
2
0
.18
1
8
.45
1
6
.73
1
6
.11
1
2
.64
0
.10
0
2
6
.82
2
7
.86
2
7
.01
2
3
.36
2
0
.17
1
8
.45
1
6
.73
1
6
.11
1
2
.64
0
.12
5
2
6
.89
2
7
.01
2
6
.56
2
3
.35
2
0
.19
1
8
.45
1
6
.73
1
6
.11
1
2
.64
0
.15
0
2
7
.43
2
6
.96
2
6
.10
2
3
.70
2
0
.23
1
8
.48
1
6
.74
1
6
.11
1
2
.64
0
.17
5
2
8
.28
2
7
.24
2
6
.22
2
4
.05
2
0
.41
1
8
.56
1
6
.75
1
6
.11
1
2
.64
0
.20
0
2
9
.13
2
7
.96
2
6
.42
2
4
.73
2
0
.87
1
8
.79
1
6
.77
1
6
.11
1
2
.64
0
.22
5
2
9
.88
2
8
.64
2
7
.18
2
5
.29
2
1
.77
1
9
.27
1
6
.83
1
6
.13
1
2
.63
0
.25
0
3
0
.28
2
8
.94
2
7
.62
2
6
.10
2
3
.08
1
9
.87
1
6
.94
1
6
.14
1
2
.63
0
.27
5
3
0
.12
2
8
.91
2
7
.74
2
6
.55
2
4
.61
2
1
.17
1
7
.42
1
6
.24
1
2
.60
0
.30
0
2
9
.55
2
8
.20
2
7
.46
2
6
.77
2
5
.42
2
2
.40
1
8
.30
1
6
.42
1
2
.50
0
.32
5
2
9
.24
2
7
.46
2
6
.68
2
6
.37
2
5
.67
2
3
.58
1
9
.43
1
6
.61
1
2
.23
0
.35
0
2
9
.33
2
7
.01
2
6
.04
2
5
.55
2
5
.21
2
3
.97
2
0
.71
1
6
.77
1
1
.45
0
.37
5
2
9
.53
2
6
.85
2
5
.49
2
4
.79
2
4
.37
2
3
.66
2
1
.40
1
6
.56
1
0
.17
0
.40
0
2
9
.63
2
6
.89
2
5
.16
2
4
.17
2
3
.66
2
2
.81
2
1
.33
1
6
.10
9
.19
0
.42
5
2
9
.59
2
6
.85
2
4
.95
2
3
.75
2
3
.07
2
2
.14
2
0
.74
1
5
.68
8
.81
0
.45
0
2
9
.57
2
6
.77
2
4
.80
2
3
.50
2
2
.63
2
1
.30
19
.57
1
5
.24
8
.80
0
.47
5
2
9
.51
2
6
.73
2
4
.76
2
3
.25
2
2
.19
2
0
.54
1
8
.11
1
3
.92
8
.46
0
.50
0
2
9
.51
2
6
.67
2
4
.67
2
3
.07
2
1
.74
1
9
.48
1
6
.44
1
2
.09
7
.56
Table
2
.
T
he
c
om
pu
te
r
sim
ulati
on
c
orrelat
io
n of n
oise s
uppressi
ng ca
pa
bili
ty
an
d har
d
c
onsist
ent
of HD
T
dissim
il
arit
y
(Len
a)
Hard Co
n
sis
ten
t
PSNR (dB
)
10
20
30
40
50
60
70
80
90
0
.02
5
3
4
.73
3
2
.15
2
7
.91
2
3
.79
2
0
.57
1
8
.17
1
7
.12
1
6
.38
1
4
.17
0
.05
0
3
1
.40
3
1
.98
2
7
.91
2
3
.79
2
0
.57
1
8
.17
1
7
.12
1
6
.38
1
4
.17
0
.07
5
2
7
.99
3
1
.07
2
7
.90
2
3
.79
2
0
.57
1
8
.17
1
7
.12
1
6
.38
1
4
.17
0
.10
0
2
6
.4
8
2
8
.79
2
7
.74
2
3
.79
2
0
.57
1
8
.17
1
7
.12
1
6
.38
1
4
.17
0
.12
5
2
6
.15
2
6
.87
2
7
.30
2
3
.82
2
0
.58
1
8
.17
1
7
.12
1
6
.38
1
4
.17
0
.15
0
2
6
.91
2
5
.87
2
6
.37
2
3
.99
2
0
.58
1
8
.17
1
7
.12
1
6
.38
1
4
.17
0
.17
5
2
8
.29
2
6
.30
2
5
.59
2
4
.13
2
0
.66
1
8
.18
1
7
.12
1
6
.38
1
4
.17
0
.20
0
2
9
.79
2
7
.27
2
5
.7
1
2
4
.58
2
0
.94
1
8
.19
1
7
.12
1
6
.38
1
4
.17
0
.22
5
3
1
.24
2
8
.66
2
6
.43
2
4
.93
2
1
.65
1
8
.25
1
7
.08
1
6
.35
1
4
.16
0
.25
0
3
2
.46
2
9
.79
2
7
.53
2
5
.47
2
2
.79
1
8
.56
1
7
.07
1
6
.31
1
4
.13
0
.27
5
3
2
.31
3
0
.13
2
8
.24
2
6
.36
2
3
.67
1
9
.47
1
7
.01
1
6
.11
1
4
.00
0
.30
0
3
1
.13
2
9
.24
2
8
.19
2
6
.79
2
4
.6
4
2
0
.73
1
7
.08
1
5
.46
1
3
.70
0
.32
5
2
9
.37
2
7
.50
2
6
.98
2
6
.21
2
4
.86
2
1
.83
1
7
.25
1
4
.35
1
2
.60
0
.35
0
2
8
.06
2
5
.62
2
5
.08
2
4
.49
2
3
.77
2
2
.18
1
7
.47
1
3
.13
1
0
.92
0
.37
5
2
6
.52
2
3
.93
2
2
.98
2
2
.59
2
2
.02
2
1
.21
1
7
.75
1
1
.89
9
.24
0
.40
0
2
5
.20
2
2
.54
2
1
.19
2
0
.73
1
9
.88
1
9
.29
1
7
.08
1
1
.85
7
.96
0
.42
5
2
4
.13
2
1
.34
1
9
.68
1
8
.91
1
7
.91
1
7
.04
1
5
.60
1
2
.25
7
.78
0
.45
0
2
3
.16
2
0
.10
1
8
.42
1
7
.22
1
6
.12
1
5
.06
1
3
.84
1
1
.83
8
.53
0
.47
5
2
2
.14
1
9
.07
1
7
.26
1
5
.82
1
4
.59
1
3
.32
1
2
.26
1
0
.75
8
.79
0
.50
0
2
1
.25
1
8
.05
1
6
.19
1
4
.66
1
3
.32
1
1
.93
1
0
.93
9
.66
8
.28
Evaluation Warning : The document was created with Spire.PDF for Python.
IS
S
N
:
2502
-
4752
Ind
on
esi
a
n
J
E
le
c Eng &
Co
m
p
Sci,
Vo
l.
24
, N
o.
1
,
Oct
ober
20
21
:
144
-
15
6
152
Fr
om
the
res
ults
of
c
om
pu
te
r
si
m
ulati
on
co
ns
umm
at
ion
s
of
Air
plane
photogra
phic
,
the
HD
T
ha
r
d
consi
ste
nt
f
or
t
he
m
os
t
capa
bili
ty
m
us
t
be
se
t
durin
g
0.0
25
-
0.350
r
oughtl
y.
By
us
i
ng
al
ge
br
ai
c
i
nv
e
sti
ga
ti
on
,
the
first
sta
ti
sti
cal
m
o
m
ent
an
d
the
seco
nd
st
at
ist
ic
al
m
o
m
e
nt
of
the
norm
al
iz
ed
H
DT
dissi
m
il
arity,
whic
h
are
cal
culat
ed
f
rom
these
com
pu
te
r
sim
ulati
on
consum
m
ation
s
in
Ta
ble
3,
are
0.1
778±0
.1308.
From
the
resu
lt
s
of
c
om
pu
te
r
s
i
m
ulati
on
co
nsum
m
a
ti
on
s
of
Pe
pp
e
r
phot
ogra
ph
ic
,
t
he
HD
T
ha
rd
co
ns
ist
ent
f
or
th
e
m
os
t
c
apab
il
it
y
m
us
t
be
set
du
rin
g
0.025
-
0.3
50
r
oughtl
y.
By
us
i
ng
al
ge
br
a
ic
inv
est
igati
on,
t
he
first
sta
ti
sti
ca
l
m
o
m
ent
and
the
seco
nd
sta
ti
sti
cal
m
o
m
ent
of
the
norm
al
i
zed
H
DT
dissi
m
il
arity,
wh
ic
h
are
cal
culat
ed
from
these c
om
pu
te
r
sim
ulatio
n
c
onsu
m
m
at
ion
s in
Ta
ble
4,
a
re
0.194
4±0.1
429.
Table
3
.
T
he
c
om
pu
te
r
sim
ulati
on
c
orrelat
io
n of n
oise s
uppressi
ng ca
pa
bili
ty
an
d har
d
c
onsist
ent
of HD
T
dissim
il
arit
y
(A
ir
plane)
Hard Co
n
sis
ten
t
PSNR (dB
)
10
20
30
40
50
60
70
80
90
0
.02
5
3
4
.15
3
1
.38
2
7
.13
2
3
.01
1
9
.62
1
7
.66
1
6
.25
1
5
.64
1
3
.89
0
.05
0
3
0
.96
3
1
.20
2
7
.12
2
3
.01
1
9
.62
1
7
.66
1
6
.25
1
5
.64
1
3
.89
0
.07
5
2
7
.72
2
9
.95
2
7
.07
2
3
.01
1
9
.62
1
7
.66
1
6
.25
1
5
.64
1
3
.89
0
.10
0
2
6
.10
2
7
.77
2
7
.04
2
3
.03
1
9
.62
1
7
.66
1
6
.25
1
5
.64
1
3
.89
0
.12
5
2
5
.89
2
6
.39
2
6
.80
2
3
.07
1
9
.62
1
7
.66
1
6
.25
1
5
.64
1
3
.89
0
.15
0
2
6
.34
2
5
.55
2
5
.73
2
3
.41
1
9
.64
1
7
.66
1
6
.25
1
5
.64
1
3
.89
0
.17
5
2
6
.82
2
5
.59
2
4
.97
2
3
.87
1
9
.80
1
7
.67
1
6
.25
1
5
.64
1
3
.89
0
.20
0
2
7
.23
2
5
.98
2
5
.03
2
4
.15
2
0
.24
1
7
.73
1
6
.24
1
5
.64
1
3
.89
0
.22
5
2
7
.20
2
6
.17
2
5
.39
2
4
.10
2
1
.17
1
7
.96
1
6
.25
1
5
.63
1
3
.89
0
.25
0
2
7
.04
2
5
.93
2
5
.16
2
4
.11
2
2
.00
1
8
.47
1
6
.24
1
5
.60
1
3
.84
0
.27
5
2
7
.10
2
5
.37
2
4
.68
2
3
.82
2
2
.73
1
9
.45
1
6
.36
1
5
.45
1
3
.77
0
.30
0
2
7
.10
2
4
.87
2
4
.04
2
3
.32
2
2
.88
2
0
.43
1
6
.70
1
5
.03
1
3
.60
0
.32
5
2
6
.88
2
4
.40
2
3
.29
2
2
.66
2
2
.50
2
1
.14
1
7
.46
1
4
.49
1
2
.98
0
.35
0
2
6
.48
2
4
.01
2
2
.60
2
1
.93
2
1
.76
2
1
.03
1
8
.00
1
3
.87
1
1
.63
0
.37
5
2
6
.03
2
3
.40
2
1
.93
2
1
.17
2
0
.81
2
0
.27
1
7
.79
1
3
.06
1
0
.14
0
.40
0
2
5
.50
2
2
.91
2
1
.30
2
0
.46
1
9
.83
1
9
.29
1
7
.53
1
3
.02
8
.85
0
.42
5
2
5
.16
2
2
.47
2
0
.72
1
9
.77
1
8
.93
1
8
.16
1
6
.79
1
3
.03
8
.44
0
.45
0
2
4
.84
2
2
.10
2
0
.22
1
9
.17
18
.15
1
7
.07
1
5
.51
1
2
.81
8
.86
0
.47
5
2
4
.58
2
1
.70
1
9
.77
1
8
.59
1
7
.37
1
6
.08
1
4
.14
1
1
.94
8
.90
0
.50
0
2
4
.32
2
1
.36
1
9
.41
1
8
.02
1
6
.65
1
4
.97
1
2
.79
1
0
.84
8
.33
Table
4
.
T
he
c
om
pu
te
r
sim
ulati
on
c
orrelat
io
n of n
oise s
uppressi
ng ca
pa
bili
ty
an
d har
d
c
onsist
ent
of HD
T
di
ssim
il
arit
y
(P
epp
e
r)
Hard Co
n
sis
ten
t
PSNR (dB
)
10
20
30
40
50
60
70
80
90
0
.02
5
3
5
.40
3
1
.64
2
6
.77
2
3
.50
2
0
.22
1
8
.11
1
6
.59
1
6
.04
1
3
.93
0
.05
0
3
1
.40
3
1
.45
2
6
.76
2
3
.50
2
0
.22
1
8
.11
1
6
.59
1
6
.04
1
3
.93
0
.07
5
2
7
.93
3
0
.26
2
6
.73
2
3
.50
2
0
.22
1
8
.11
1
6
.59
1
6
.04
13.
93
0
.10
0
2
6
.12
2
7
.93
2
6
.58
2
3
.50
2
0
.22
1
8
.11
1
6
.59
1
6
.04
1
3
.93
0
.12
5
2
5
.96
2
5
.82
2
6
.14
2
3
.50
2
0
.22
1
8
.11
1
6
.59
1
6
.04
1
3
.93
0
.15
0
2
6
.56
2
5
.20
2
5
.39
2
3
.52
2
0
.22
1
8
.11
1
6
.59
1
6
.03
1
3
.93
0
.17
5
2
7
.48
2
5
.58
2
5
.15
2
3
.79
2
0
.27
1
8
.11
1
6
.59
1
6
.02
1
3
.93
0
.20
0
28
.40
2
6
.54
2
4
.91
2
4
.01
2
0
.45
1
8
.14
1
6
.59
1
6
.01
1
3
.93
0
.22
5
2
9
.04
2
7
.29
2
5
.34
2
4
.25
2
1
.01
1
8
.24
1
6
.59
1
6
.00
1
3
.93
0
.25
0
2
9
.28
2
7
.64
2
6
.12
2
4
.45
2
1
.92
1
8
.61
1
6
.56
1
5
.94
1
3
.91
0
.27
5
2
9
.28
2
7
.50
2
6
.44
2
4
.85
2
2
.95
1
9
.37
1
6
.52
1
5
.81
1
3
.80
0
.30
0
2
9
.11
2
6
.93
26
.23
2
4
.96
2
3
.64
2
0
.27
1
6
.41
1
5
.29
1
3
.38
0
.32
5
2
8
.42
2
6
.03
2
5
.47
2
4
.47
2
3
.77
2
1
.21
1
6
.71
1
4
.49
1
2
.63
0
.35
0
2
7
.64
2
5
.08
2
4
.34
2
3
.53
2
2
.90
2
1
.25
1
6
.91
1
3
.24
1
1
.08
0
.37
5
2
6
.73
2
4
.10
2
3
.07
2
2
.29
2
1
.53
2
0
.45
1
6
.84
1
2
.29
9
.28
0
.40
0
2
5
.78
2
2
.99
2
1
.77
2
0
.86
19.
95
1
8
.98
1
6
.74
1
2
.14
8
.02
0
.42
5
2
4
.69
2
1
.84
2
0
.51
1
9
.40
1
8
.24
1
7
.16
1
5
.47
1
2
.49
7
.87
0
.45
0
2
3
.60
2
0
.80
1
9
.17
1
7
.78
1
6
.59
1
5
.40
1
3
.95
1
2
.07
8
.57
0
.47
5
2
2
.49
1
9
.66
1
7
.89
1
6
.38
1
5
.08
1
3
.79
1
2
.49
1
1
.01
8
.81
0
.50
0
2
1
.60
1
8
.50
1
6
.75
1
5
.15
1
3
.81
1
2
.55
1
1
.21
9
.91
8
.33
4.
3
.
The
comprehensive ca
pa
bil
ity
of
t
he no
ise
sup
pressive technique st
an
d
on
HD
T
dissi
mi
larity
In
co
ns
i
der
at
i
on
of
the
PS
NR
capab
il
it
y
of
the
noise
su
ppres
sin
g
te
chn
iq
ue
sta
nd
on
HDT
dissim
il
arit
y,
t
he
ex
per
im
ental
si
m
ulati
on
of
the
noise
s
u
ppressi
ve
te
ch
ni
qu
e
sta
nd
on
H
DT
dissim
il
arity
with
the
opti
m
iz
ed
HD
T
ha
r
d
c
onsist
ent,
w
hich
are
in
vestigat
e
d
f
r
om
the
pr
e
vious
sim
ulatio
n,
a
re
in
vestigat
ed
on
four
groun
d
-
tr
uth
ph
otogra
phs
(G
i
rl,
Le
na,
Air
plane
a
nd
Pepper
)
by
c
orrupti
ng
F
IIN
a
t
plentif
ul
fr
e
quency
Evaluation Warning : The document was created with Spire.PDF for Python.
Ind
on
esi
a
n
J
E
le
c Eng &
Co
m
p
Sci
IS
S
N:
25
02
-
4752
A com
pu
t
ational ex
pe
rime
ntal
investi
ga
ti
on
of noise s
uppre
ss
ing t
ech
niqu
e sta
nd on…
(
Vorapoj P
atan
aviji
t
)
153
durin
g
0%
to
90%
an
d
are
c
om
par
ed
to
M
F
,
s
m
oo
thin
g
filt
er
(
SF
)
a
nd
a
da
ptive
m
edian
filt
er
(
AMF
),
wh
ic
h
cou
l
d
be
la
id
out
as
Table
5.
Fr
om
the
resu
l
ts
of
com
pu
te
r
si
m
ulati
on
co
nsum
m
a
ti
on
s
of
four
ph
otog
ra
ph
s
i
n
Table
5,
the
no
ise
s
uppr
e
ss
ing
te
ch
nique
sta
nd
on
H
DT
dissim
il
ari
ty
is
su
perb
than
al
l
oth
e
r
no
ise
su
pp
ressin
g
te
chn
i
qu
e
s
beca
us
e
the
im
pu
lsi
ve
no
ise
posit
ion
in
g
ide
ntific
at
ion
sta
nd
on
H
DT
dissim
i
la
rity
is
ver
y
hi
gh
acc
uracy
thu
s
the
noise
su
ppressi
ng
te
chn
i
qu
e
sta
nd
on
HD
T
dis
si
m
il
arity
on
ly
con
ceal
s
im
pu
lsi
ve
no
ise
p
i
xels and
un
-
to
uc
hes n
oise
-
fr
ee
pix
el
s
.
By
cause
of
th
e
publica
ti
on
c
ircum
sp
ect
ion
of
pa
ges,
the
pa
rtit
ion
of
a
c
om
pu
ta
ti
on
al
ex
per
im
ental
inv
est
igati
on
pro
vid
es
few
s
uppresse
d
ph
otogra
phs,
w
hich
are
processe
d
by
MF,
SF,
A
MF
and
HD
T
,
wh
ic
h
cou
l
d
be
la
id
ou
t
as
Fi
gure
7.
F
ro
m
the
resu
lt
s
in
Fig
ur
e
7,
the
qu
a
nlit
y
of
the
HD
T
su
pp
resse
d
im
ages
ar
e
sli
gh
tl
y bett
er t
han A
MF
s
uppresse
d
im
ages an
d d
ram
at
ic
a
l
b
et
te
r
t
han M
F and SF s
uppr
e
ssed
im
ages.
Table
5
.
T
he
c
om
pu
te
r
sim
ulati
on
c
orrelat
io
n of n
oise s
uppressi
ng ca
pa
bili
ty
an
d har
d
c
onsist
ent
of HD
T
dissim
il
arit
y
(
pep
pe
r
)
PSNR (dB
)
Proces
sed
Pho
to
g
raph
s
No
ise
Frequ
en
cy
No
isy
Ph
o
to
grap
h
s
No
ise Su
p
p
ressin
g
T
echn
iq
u
e
SMF
(3x
3
)
SF
(3x
3
)
AMF
HDT
F
ilter
Lena
(25
6
x
2
5
6
)
D=0
.10
1
5
.65
6
4
3
0
.70
7
6
1
9
.38
1
2
3
5
.30
3
2
3
4
.72
9
4
D=0
.20
1
2
.63
8
9
2
7
.62
5
7
1
6
.32
0
8
3
2
.15
5
8
3
2
.15
0
7
D=0
.30
1
0
.89
7
1
2
3
.68
1
1
1
4
.58
2
9
2
7
.91
4
1
2
8
.23
8
3
D=0
.40
9
.64
8
1
1
9
.00
8
0
1
3
.24
7
9
2
3
.79
0
3
2
6
.79
2
2
D=0
.50
8
.65
5
3
1
5
.47
5
8
12
.21
4
6
2
0
.57
2
5
2
4
.85
5
9
D=0
.60
7
.78
1
3
1
2
.32
8
0
1
1
.29
3
9
1
8
.17
4
7
2
2
.18
2
3
D=0
.70
7
.16
9
7
1
0
.28
6
1
1
0
.65
0
9
1
7
.11
5
3
1
7
.74
7
9
D=0
.80
6
.58
4
6
8
.33
3
1
1
0
.00
5
7
1
6
.45
5
4
1
6
.37
5
3
D=0
.90
6
.06
0
4
6
.82
4
1
9
.43
5
6
1
6
.53
5
2
1
4
.17
2
0
Pep
p
er
(25
6
x
2
5
6
)
D=0
.10
1
5
.37
9
8
3
0
.61
1
6
1
9
.0
6
7
7
3
6
.03
9
1
3
5
.40
4
1
D=0
.20
1
2
.35
9
3
2
6
.58
8
8
1
5
.98
0
4
3
1
.64
8
5
3
1
.64
4
0
D=0
.30
1
0
.62
4
2
2
2
.06
6
3
1
4
.17
4
8
2
6
.76
5
0
2
6
.76
5
0
D=0
.40
9
.39
9
8
1
8
.43
2
1
1
2
.90
7
6
2
3
.49
9
5
2
4
.95
6
7
D=0
.50
8
.38
4
3
1
4
.85
0
6
1
1
.81
1
7
2
0
.22
0
3
2
3
.77
2
0
D=0
.60
7
.61
8
9
1
2
.01
2
8
1
0
.95
6
3
1
8
.11
1
6
21
.24
8
9
D=0
.70
6
.92
4
6
9
.77
0
4
1
0
.20
3
9
1
6
.59
2
3
1
6
.91
0
6
D=0
.80
6
.37
1
0
8
.01
6
6
9
.58
5
3
1
6
.08
9
6
1
6
.03
5
2
D=0
.90
5
.85
8
2
6
.57
6
7
9
.02
1
4
1
6
.29
3
2
1
3
.93
4
3
Airplan
e
(25
6
x
2
5
6
)
D=0
.10
1
4
.83
2
0
2
9
.65
3
2
1
8
.44
2
6
3
4
.63
1
1
3
4
.15
1
4
D=0
.20
1
1
.80
4
5
2
6
.43
5
6
1
5
.31
8
1
3
1
.38
4
4
31
.37
8
2
D=0
.30
1
0
.05
1
0
2
1
.88
6
2
1
3
.45
2
6
2
7
.13
4
7
2
7
.13
4
6
D=0
.40
8
.87
3
5
1
7
.64
1
2
1
2
.13
9
7
2
3
.01
4
7
2
4
.15
4
7
D=0
.50
7
.86
0
0
1
4
.26
9
7
1
1
.00
9
1
1
9
.62
0
1
2
2
.87
6
5
D=0
.60
7
.09
2
0
1
1
.52
9
0
1
0
.12
0
2
1
7
.65
8
6
2
1
.13
9
8
D=0
.70
6
.41
2
8
9
.30
4
2
9
.32
3
8
1
6
.25
1
4
1
8
.00
0
6
D=0
.80
5.
8647
7
.58
3
5
8
.68
9
3
1
5
.74
2
8
1
5
.64
2
3
D=0
.90
5
.33
3
5
6
.02
7
8
8
.03
8
1
1
6
.08
3
4
1
3
.88
9
6
Girl
(25
6
x
2
5
6
)
D=0
.10
1
3
.68
9
0
3
1
.55
8
3
1
7
.25
3
0
3
6
.91
9
7
3
4
.68
6
3
D=0
.20
1
0
.65
6
7
2
5
.51
5
3
1
3
.95
9
3
3
2
.04
3
7
3
1
.89
6
4
D=0
.30
8
.86
7
7
2
0
.77
3
8
1
1
.95
9
9
2
7
.69
3
0
2
7
.73
6
6
D=0
.40
7
.57
9
8
1
6
.51
4
6
1
0
.45
4
3
2
3
.37
3
6
2
6
.77
1
7
D=0
.50
6
.57
1
2
1
3
.03
1
9
9
.23
6
7
2
0
.17
1
2
2
5
.66
7
2
D=0
.60
5
.86
0
9
1
0
.49
8
1
8
.35
9
0
1
8
.45
1
8
2
3
.97
4
1
D=0
.70
5
.13
1
1
8
.04
6
3
7
.42
7
1
1
6
.73
3
4
2
1
.39
7
1
D=0
.80
4
.56
7
4
6
.25
2
0
6
.68
8
1
1
6
.27
9
5
1
6
.76
6
2
D=0
.90
4
.05
7
3
4
.74
6
5
5
.99
8
6
1
6
.74
6
3
1
2
.63
7
1
Evaluation Warning : The document was created with Spire.PDF for Python.