Internati
o
nal
Journal of Ele
c
trical
and Computer
Engineering
(IJE
CE)
V
o
l.
6, N
o
. 1
,
Febr
u
a
r
y
201
6,
pp
. 13
0
~
13
8
I
S
SN
: 208
8-8
7
0
8
,
D
O
I
:
10.115
91
/ij
ece.v6
i
1.8
483
1
30
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
The Noise Reduction over Wire
less Channel Using Vector
Quantization Compressi
on and Filterin
g
Iman
E
l
aw
ad
y,
Ab
del
m
ou
n
a
i
m
Mo
ul
a
y
L
a
kh
dar
,
K
h
el
i
f
i
M
u
st
aph
a
Departement of
Electronic, Tah
r
i M
ohammed University
,
Be
char
,
Algeri
a
Article Info
A
B
STRAC
T
Article histo
r
y:
Received
J
u
n 29, 2015
Rev
i
sed
Sep 7, 20
15
Accepte
d Oct 3, 2015
The transm
ission of com
p
resse
d data
over wireless channe
l condition
s
represents a big
challenge
. The
idea of providing
robust transmission gets a
lot of atten
tion in field of r
e
sear
ch. In th
is paper
we stud
y
the
ef
fect of
th
e
noise over wirel
e
ss channel
.
W
e
use the m
odel of Gilbert-E
llio
t t
o
represen
t
the wir
e
les
s
cha
nnel.
Th
e par
a
m
e
ters
of
th
e m
odel
are
s
e
le
ct
ed t
o
repres
e
n
t
three
cas
es
of
ch
annel
.
As
dat
a
f
o
r trans
m
is
s
i
on we us
e im
ages
i
n
gra
y
l
e
ve
l
size 512x512.
To minimize ban
d
width usage we compressed th
e image with
vector qu
ant
i
za
ti
on als
o
in
this
c
o
m
p
re
ssion tech
nique we stud
y
the eff
ect of
the
codebook
in
the robustn
ess of transmission so we use diff
eren
t
algorithms
to generate a codebook for the vector
quan
t
ization finally
w
e
stud
y
th
e
res
t
orat
ion eff
i
ci
enc
y
of r
e
c
e
ived
im
age us
ing fi
lt
ering and
indi
ce
s
recov
e
r
y
techn
i
que.
Keyword:
Co
d
e
bo
ok
g
e
ner
a
tio
n
Filterin
g
Gilb
ert-Ellio
t ch
ann
e
l
Indices recovery
Vector qua
ntization
Copyright ©
201
6 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
:
Im
an Elawady,
Depa
rtem
ent of Electronic,
Tah
r
i M
o
h
a
mmed
Un
iv
ersity,
B.P 417
BEC
H
A
R
(08
000
), Alg
e
r
i
a.
Em
a
il: i
m
an
elawad
y
2
@g
m
a
il.
co
m
1.
INTRODUCTION
The t
r
a
n
sm
i
s
si
on
of
dat
a
ove
r n
o
i
s
y
chan
nel
m
a
ke us co
ncer
ne
d ab
o
u
t
t
h
e ro
b
u
st
ness
o
f
transm
ission and t
h
e effectiveness
of re
storation process in case of erroneou
s d
a
ta. Th
ere is lo
t o
f
p
a
ra
m
e
ters
affect in the
quality of receive
d im
age.
Som
e
researche
r
s thi
n
k
that the
problem
can be
solved by optim
i
ze
in
th
e sou
r
ce cod
i
ng
as sh
own
in
th
e articles [1
],
[2
] an
d
[3
], o
t
h
e
rs t
h
ink
th
at th
e
so
lu
tion
ex
ists in
th
e
opt
i
m
i
zati
on i
n
t
h
e cha
nnel
codi
ng as s
h
o
w
n i
n
t
h
e art
i
c
l
e
s [4]
and [
5
]
,
ot
her g
u
esse
s t
h
i
nk t
h
at
we
sho
u
l
d
m
a
ke com
b
i
n
a
t
i
on
bet
w
een
t
h
e s
o
u
r
ce a
n
d
cha
n
nel
co
di
ng
o
r
w
h
at
w
e
cal
l
j
o
i
n
t
s
o
urce
cha
n
nel
c
odi
ng
(JSCC) as sh
own in
t
h
e artic
les [6
], [7
] an
d
[8
], i
n
th
is prop
o
s
ition
t
h
ey th
ink
we
h
a
v
e
to
o
p
tim
ize th
e sou
r
ce
and t
h
e cha
n
n
e
l
codi
n
g
acc
or
di
n
g
t
o
t
h
e
chan
nel
co
n
d
i
t
i
ons
whi
c
h i
s
un
pre
d
i
c
t
a
bl
e
and
de
pen
d
i
n
g o
n
p
r
ob
ab
ility also
it co
nsu
m
es lo
t
o
f
ti
m
e
an
d resou
r
ces of
p
r
o
cessi
ng
u
n
it
b
ecau
s
e th
ey are iterativ
e
alg
o
rith
m
s
.
I
n
t
h
e o
p
t
i
m
i
z
at
i
on o
f
t
h
e so
urce c
odi
ng t
h
ey
depe
nd
on
addi
ng s
o
m
e
trans
f
orm
s
t
o
enha
nce t
h
e
q
u
a
lity o
f
recon
s
tru
c
ted
im
ag
e b
u
t
t
h
is is not a g
u
a
ran
t
ee t
h
at th
e tran
sm
i
ssio
n
is
ro
bu
st
o
r
t
h
at go
ing
t
o
m
a
k
e
the restoration
process m
o
re efficien
t with
less p
r
o
cessi
n
g
;
h
o
wev
e
r in
th
e
channel coding they create som
e
al
go
ri
t
h
m
t
h
at
add s
o
m
e
dat
a
as sho
w
n i
n
t
h
e art
i
c
l
e
s [9]
and [
1
0]
(co
rrec
t
or bi
t
s
or a
dd
som
e
redun
da
n
c
y
)
i
n
t
h
e
ori
g
i
n
al
i
n
f
o
rm
at
i
on o
r
c
h
angi
ng
t
h
e
dat
a
o
r
ga
ni
zat
i
o
n
(interleavi
n
g) i
n
way that
we
ca
n estim
ate or
guess
th
e o
r
ig
i
n
al in
form
at
io
n
as sh
own
in
th
e article [1
1
]
o
r
get it
fro
m
th
e redun
d
a
n
t
d
a
ta wh
ich
is in
gen
e
ral
m
i
nim
i
ze t
h
e c
o
m
p
ressi
o
n
rat
i
o an
d m
a
xi
m
i
ze i
n
ban
d
w
i
d
t
h
usa
g
e.
The vect
or
quantization provi
des hi
gh
efficiency
as
a source coding technique with hi
gh
co
m
p
ression
ratio
and
reason
ab
le qu
ality;
h
o
wev
e
r we
can
’
t
k
e
ep
th
is
en
h
a
n
cem
en
t after tran
sm
issi
o
n
.
As
sh
own
in
th
e article [1
2
]
th
at
th
e in
d
i
ces are
tran
sm
it
te
d ov
er a n
o
i
s
y
chan
nel
,
w
h
i
c
h i
s
o
bvi
ou
sl
y
m
o
st
oft
e
n
the case, trans
m
ission errors
usually
occur (the receive
d indices are no
t all the sa
me of the trans
m
itted
Evaluation Warning : The document was created with Spire.PDF for Python.
I
S
SN
:
2
088
-87
08
IJEC
E V
o
l
.
6, No
. 1, Feb
r
uar
y
20
1
6
:
13
0 – 13
8
13
1
in
d
i
ces). Sin
ce th
e v
ector
q
u
a
n
tizatio
n
is a blo
c
k
co
d
i
n
g
tech
n
i
q
u
e
t
h
e erro
r also
will b
e
all th
e p
i
x
e
ls in
the
bl
oc
k a
nd t
o
m
a
ke re
st
orat
i
o
n i
t
bec
o
m
e
s a
bi
g c
h
al
l
e
n
g
e
beca
use we
d
o
n
’
t
ha
ve a
n
y
i
n
f
o
rm
at
i
on ab
out
t
h
e
nat
u
re of pi
xel
s
t
h
at
cont
ai
ns
i
n
t
h
at
bl
ock.
Our
pr
op
osi
t
i
on
based
on w
o
r
k
i
n
g o
n
t
h
e nat
u
re of i
n
di
c
e
s by
u
s
ing
d
i
fferen
t
co
d
e
b
ook
s and
try to
co
llect
in
g
so
m
e
d
a
ta
th
at will b
e
se
n
d
an
d
use in
resto
r
ation
p
r
o
c
ess to
st
udy
t
h
e
i
n
fl
u
e
nce
of c
h
a
ngi
ng
t
h
e c
o
deb
o
o
k
creat
i
o
n i
n
t
h
e rece
pt
i
o
n
and
rest
orat
i
o
n
pr
ocess
an
d i
n
t
h
e
q
u
a
lity of receiv
e
d im
ag
e.
Th
is p
a
p
e
r is o
r
g
a
n
i
zed
as
fo
llo
ws in
th
e
first p
a
rt we try to
in
trodu
ce th
e con
c
ep
t of th
e
v
ector
q
u
a
n
tizatio
n
an
d
t
h
e
g
e
n
e
ratio
n
of ind
i
ces t
h
at will b
e
tr
an
sm
it
ted
o
v
e
r t
h
e ch
ann
e
l, after th
at th
e creatio
n
of
co
d
e
bo
ok
u
s
i
n
g
two
d
i
fferen
t
ap
p
r
o
a
ch
es, than
th
e p
a
ram
e
t
e
rs u
s
ed
fo
r the
m
o
d
e
l o
f
Gil
b
ert-Ellio
t as
wireless
ch
ann
e
l. In
th
i
r
d
p
a
rt we
will in
trod
u
c
e a tech
n
i
q
u
e
o
f
rest
o
r
ation
(i
n
d
i
ces recov
e
ry
). Fi
n
a
lly we p
r
esen
t th
e
si
m
u
latio
n
results, d
i
scussion
an
d con
c
lusion.
2.
VECTO
R
QU
ANTIZ
ATIO
N
Recently im
ag
e com
p
ression, especia
lly at low
bit rate, ha
s assum
e
d
a major
role in
applications
suc
h
as
st
o
r
a
g
e
on
l
o
w m
e
m
o
ry
de
vi
ces
, na
rr
o
w
-
b
a
n
d
cha
n
n
e
l
t
r
a
n
s
m
i
t
t
i
ng,
wi
rel
e
ss t
r
a
n
sm
i
t
t
i
ng a
nd
st
ream
i
ng dat
a
on t
h
e i
n
t
e
r
n
et
[13]
. S
h
an
n
on
fi
rst
su
gge
s
t
ed t
h
at
enco
d
i
ng a seq
u
e
n
c
e
of sam
p
l
e
s from
a
sou
r
ce ca
n
p
r
ovi
de
bet
t
e
r r
e
sul
t
t
h
a
n
e
n
c
odi
ng
i
n
di
vi
d
u
a
l
sam
p
l
e
s i
n
t
e
rm
s of co
m
p
ressi
on e
ffi
ci
ency
[14].Im
a
ge
Dat
a
com
p
ression using vector quanti
zation (VQ)
has
recei
ve
d a l
o
t of attent
ion.
Since VQ has a
high
coding effici
ency a
n
d
sim
p
le decode
r arc
h
itecture
,
it is very s
u
ita
ble for low-
b
it rate ap
p
licatio
n
s
.
Th
e
g
e
n
e
ral
VQ algorith
m
h
a
s th
ree m
a
in
step
s
[15
]
:
Fi
rst
di
vi
de t
h
e
im
age i
n
t
o
bl
o
c
ks
(us
u
al
l
y
t
h
ey
are
2x
2,
4
x
4
,
8
x
8
,
or
1
6
x
1
6
)
.
Aft
e
r
t
h
at
a
co
deb
o
o
k
wi
t
h
b
e
st
est
i
m
a
t
i
on of
bl
ocks
i
s
c
o
nst
r
uct
e
d a
n
d i
nde
xe
d.
Fin
a
lly, th
e
o
r
i
g
in
al im
ag
e b
l
o
c
ks ar
e su
b
s
ti
tu
ted
b
y
th
e ind
e
x of
b
e
st esti
mate co
d
e
f
r
om
th
e cod
e
book
.
The basic
pri
n
ciple of ve
ctor qua
n
tization is to
m
a
tch each input v
ect
or
with a code-ve
c
tor in the
code
b
o
o
k
s
o
t
h
at
t
h
e di
st
o
r
t
i
on bet
w
ee
n t
h
e i
n
p
u
t
v
ect
or and the chosen code
-
v
ect
or i
s
m
i
nim
u
m
[14]
.
Qu
an
tizatio
n
is an
irrev
e
rsib
le p
r
o
cess so
t
h
ere is no
way to find
th
e
o
r
i
g
inal v
a
lu
e fro
m
t
h
e qu
an
tized
valu
e
[1
6]
. The
di
ffe
rence
bet
w
ee
n
t
h
e i
n
p
u
t
and
out
put
si
g
n
al
s
of t
h
e
qua
nt
i
z
e
r
ge
nerat
e
t
h
e
qua
nt
i
z
i
ng e
r
r
o
r
,
or
qua
nt
i
z
i
n
g
noi
s
e
[
17]
.
Fi
gu
re
1.
Vect
or
q
u
a
n
t
i
zat
i
o
n
p
r
oce
d
ure
3.
GENER
A
TIO
N
O
F
CO
DEBOOK
The o
b
j
ect
i
v
e
of co
deb
o
o
k
desi
gn i
s
t
o
m
i
nim
i
ze the
com
b
ined
recons
tru
c
tion
erro
r ov
er a
represen
tativ
e train
i
ng
d
a
ta set. If th
e train
i
n
g
set
is rep
r
esen
tativ
e, th
e d
e
sig
n
e
d
cod
e
b
o
o
k
will
m
i
n
i
mi
ze th
e
reco
nst
r
uct
i
o
n
err
o
r
of
i
n
put
dat
a
h
o
w
e
ve
r,
except
fo
r t
h
e case o
f
ve
ry
sm
al
l
dim
e
nsi
on. C
o
deb
o
o
k
si
ze,
and/
or
t
r
ai
ni
ng
set
s
ex
ha
ust
i
v
e sea
r
ch
f
o
r
t
h
e set
of
codeb
ook
v
ectors th
at m
i
n
i
miz
e
d
th
e to
tal error is
i
n
t
r
act
abl
e
p
r
o
b
l
e
m
[18]
. A
g
l
obal
co
de
bo
o
k
has
o
n
e c
ode
bo
o
k
f
o
r a cl
as
s of i
m
ages. T
h
e co
de
bo
o
k
i
s
bei
n
g
deri
ved
fr
om
vect
o
r
s o
f
al
l
im
ages i
n
the class. It is less ove
rhead as
c
o
m
p
ared to local code
book due
t
o
whi
c
h i
t
ha
s l
o
wer
pe
rf
orm
a
nce.
ENCODER
DECO
DER
Ou
t
p
ut
i
n
dex
I
npu
t
vector
Out
put
vecto
r
Decode
r
code
b
o
o
k
Enc
ode
r
code
b
o
o
k
Indices
C
ode
-vect
ors
000
001
010
.
111
000
001
010
.
111
Searc
h
Loo
kup
Channel
Evaluation Warning : The document was created with Spire.PDF for Python.
I
J
ECE
I
S
SN
:
208
8-8
7
0
8
Th
e N
o
ise Redu
ctio
n o
v
er Wi
reless Cha
n
n
e
l
Using
Vecto
r
Quan
tiza
tion
C
o
mp
ressi
o
n
and
…
(
I
m
a
n El
a
w
ady)
13
2
3.
1.
B
a
si
c SO
M Al
g
o
ri
thm
Ko
h
one
n pr
o
p
o
se
d
a
Sel
f
-
O
r
g
ani
zat
i
o
n
M
A
P
(S
OM
) f
o
r u
n
su
pe
rvi
s
e
d
ne
ural
net
-
wo
rk
i
n
19
8
0
[
1
9]
.
In
itially SOM
is train
e
d with th
e rando
m
l
y in
itialized
we
ig
h
t
v
ectors and
g
e
n
e
rate an
i
n
itial co
d
e
bo
ok
. Th
e
in
itial Co
d
e
boo
k Gen
e
ratio
n
alg
o
rith
m
is su
mmarized
as fo
llo
ws [20
]
:
Step
1:
Transform
the facial im
ages
i
n
d
a
taset to
in
t
e
n
s
ity v
a
riation
v
ect
o
r
s, and
co
m
b
in
e all d
a
ta to
g
e
th
er
in
to
o
n
e
train
i
ng
set.
Step
2
:
Sp
ecify th
e
size o
f
t
h
e cod
e
bo
ok
to
N
an
d in
itialize th
e co
d
e
v
ect
o
r
s
b
y
usin
g
con
tin
uous in
ten
s
ity
vari
at
i
o
n vect
o
r
s.
Step
3:
Select a ne
w training
vect
or from
the traini
ng set.
St
ep
4:
Fi
nd
t
h
e
best
m
a
t
c
hi
ng c
o
de
vect
o
r
cl
ose
s
t
t
o
t
h
e
t
r
ai
ni
ng
v
ect
or.
Step
5
:
Mo
v
e
th
e
b
e
st-match
in
g
and
i
t
s n
e
ighb
or
hood
co
d
e
v
ector
s t
o
w
a
rd
s th
e tr
ai
n
i
ng
v
ector
.
Step
6
:
Rep
eat
fro
m
Step
3
u
n
til
th
e map
con
v
e
rg
es.
3.
2.
B
a
si
c L
B
G Al
gori
t
hm
The de
si
g
n
o
f
opt
i
m
al
vect
or qua
nt
i
zers
wer
e
pr
op
ose
d
an
d
ext
e
nsi
v
el
y
st
udi
e
d
by
Li
n
d
e
,
B
u
zo
, a
n
d
Gray using a clustering approac
h
,
and
is referred
to
as th
e LBG al
g
o
rith
m
[2
1
]
.
Th
is algo
rithm is a
gene
ral
i
zat
i
on
of t
h
e
Ll
oy
d
-
M
a
x desi
gn al
go
ri
t
h
m
for sc
al
ar q
u
ant
i
zat
i
on
[2
2]
. T
h
e L
B
G al
go
ri
t
h
m
and
ot
he
r
vari
at
i
o
ns o
f
t
h
i
s
al
go
ri
t
h
m
are base
d u
p
o
n
m
i
nim
i
zi
ng a di
st
ort
i
o
n m
e
asure
whi
c
h re
p
r
esent
s
t
h
e
pen
a
l
t
y
or
cost
ass
o
ci
at
ed
wi
t
h
t
h
e m
a
pp
i
ng.
T
h
e LB
G
al
go
ri
t
h
m
for c
ode
b
o
o
k
ge
ner
a
t
i
on i
s
as
f
o
l
l
o
ws:
[
2
3]
St
ep
1.
Gi
ve
n a
n
a
r
bi
t
r
ary
c
o
deb
o
o
k
,
enc
o
de eac
h i
n
p
u
t
vector ac
cording t
o
t
h
e
nearest
-
nei
g
hbor criterion.
Use a d
i
stan
ce m
e
tric to
co
mp
are all th
e i
n
p
u
t
v
ect
ors
to
the e
n
code
d
vectors, a
n
d the
n
s
u
m
these
err
o
rs
(di
s
t
a
nc
es) t
o
pr
ovi
de
a di
st
ort
i
o
n
m
easure.
If t
h
e di
st
ort
i
o
n i
s
eno
u
gh sm
al
l (l
ess t
h
an a
p
r
ed
efi
n
ed thresh
o
l
d
)
, th
en
quit. If
no
t,
g
o
to
th
e step 2.
St
ep
2.
For eac
h c
ode
bo
o
k
ent
r
y
,
c
o
m
pute the Euclidean cent
r
oi
d of all th
e i
n
p
u
t
vect
ors e
n
c
o
ded i
n
t
o
t
h
at
specific c
o
debook
vector.
Step
3:
use t
h
e c
o
m
puted ce
ntroids a
s
the new
code
book,
and go
to step 1.
4.
TRANSMISSION
CHANNEL (GILBERT-ELLIOTT)
Du
ri
n
g
t
h
e 1
9
80s
, t
h
e em
ergence o
f
di
gi
t
a
l
com
m
uni
cat
i
o
n base
d o
n
di
g
i
t
a
l
t
echnol
o
g
i
e
s pr
om
ot
ed
t
h
e seco
n
d
ge
n
e
rat
i
on
of m
o
b
i
l
e
co
m
m
uni
cat
i
on sy
st
em
an
d i
t
s
st
anda
rdi
zat
i
on de
vel
o
p
m
ent
,
t
h
e ap
pl
i
cat
i
o
n
o
f
d
i
g
ital techn
o
l
o
g
i
es no
t
on
ly in
creased
syste
m
cap
acity, bu
t also
m
a
d
e
wireless busin
ess
qu
ality m
o
re
reliab
l
e [24
]
.
Th
e Gil
b
ert-El
lio
tt
m
o
d
e
l, o
f
ten
u
s
ed
fo
r t
h
e
m
o
d
e
lin
g
o
f
a d
i
screte ch
ann
e
l with
m
e
mo
ry, is
sim
p
l
y
a
M
a
rk
ov c
h
ai
n
of t
w
o st
at
es
: a state "GOOD" a
n
d "BAD" a
n
d that
within eac
h state, the channel
beha
ves
as a B
S
C
o
f
B
E
R
‘
Є
G
’ in ca
se
of t
h
e state "GOOD" and
‘
Є
B
’ in c
a
se of the
state "BAD"
.
We co
n
s
i
d
er, in
th
e
fo
llowing, th
r
ee
differe
n
t channels that
param
e
te
rs are
gi
ve
n i
n
T
a
bl
e
4. C
h
an
nel
1
prese
n
t
t
h
e
m
o
re bad
t
r
an
sm
i
ssi
on c
o
n
d
i
t
i
ons
wi
t
h
a
n
ave
r
age
d
u
rat
i
on
o
f
fa
di
n
g
(cha
n
n
el
i
n
t
h
e st
at
e
"BAD"
) four t
i
m
e
s greater than that of channels 2 a
n
d
3. It rem
a
in
s 8
0
% o
f
th
e ti
m
e
in
th
e state "B
AD".
Howev
e
r, ch
ann
e
l 3
rem
a
in
s o
n
l
y 20
% of the ti
m
e
in
th
is s
t
ate. No
te th
at th
e BER o
f
th
e state "BAD" is 1
0
%
for the
three c
h
annels c
o
nsidered and t
h
e m
ean BERs
of
t
h
ese c
h
a
nnels
are re
spectivel
y of the
orde
r
of 8%
fo
r c
h
an
nel
1,
5%
fo
r c
h
a
nnel
2 a
n
d
2 %
f
o
r
chan
nel
3.
Table
1.
Param
e
ters of t
h
e c
o
nsidere
d
c
h
anne
ls
Channel 1
Channel 2
Channel 3
Є
G
BER f
o
r the sta
t
e “
GOOD”
0.001
0.
001 0.001
Є
B
BE
R for
the state “BAD”
0.
1
0.
1
0.
1
P
GB
: Probability of transition fro
m
th
e "G
OOD
" to state
"BAD
"
0.005
0.005
0.0012
5
P
BG
: Probability of transition fro
m
th
e"BA
D
"
to state
"GOOD
"
0.0012
5
0.005
0.005
P
G
: P
r
obability th
at the channel is in
state "GOO
D"
0.2
0.5
0.8
P
B
: Probability tha
t
the channel
is in
state “
b
ad”
0.8
0.5
0.2
m
e
an BE
R
0.
0802
0.
0505
0.
0208
Aver
age length in
bits of a fadin
g
(
r
e
side
nce tim
e
in
the state "
B
AD")
800
200
200
5.
IND
I
CES
RE
CO
VER
Y
Thi
s
t
e
c
hni
que
de
pe
nds
on c
o
l
l
ect
i
ng som
e
veri
fi
cat
i
on i
n
f
o
rm
at
i
on (bl
o
c m
ean and com
p
l
e
xi
ty
vari
at
i
o
n)
i
s
e
m
bedded i
n
t
o
t
h
e i
n
di
ces
of
t
h
e V
Q
e
n
c
o
d
e
d dat
a
. Fi
rst
,
we di
vi
de
t
h
e
i
ndi
ces
m
a
t
r
i
x
(c
ode
vectors) into s
u
b
blocs
of siz
e
w×
w. for ea
ch i
ndice
we calcu
l
ate th
e abso
lu
te
differe
n
ce bet
w
een it
vector
mean value
a
n
d the m
e
dian m
ean va
l
u
e of nei
g
h
b
o
r
i
n
g
co
de vect
or
. Aft
e
r
t
h
at
, we
kee
p
t
h
e
m
a
xi
m
u
m
abs
o
l
u
t
e
di
ffe
r
e
nce fo
r
eac
h w×w
su
b bl
o
c
.
Evaluation Warning : The document was created with Spire.PDF for Python.
I
S
SN
:
2
088
-87
08
IJEC
E V
o
l
.
6, No
. 1, Feb
r
uar
y
20
1
6
:
13
0 – 13
8
13
3
Figu
re
2.
Ve
rif
i
cation in
fo
rm
ation
In rece
ption we
have
the rece
ived
i
ndi
ces m
a
t
r
i
x
t
h
at
c
ont
a
i
ns so
m
e
erroneous indices.
We calculate
in reception the sa
m
e
information in
em
ission. The
n
we c
o
m
p
are it with
the origi
n
al inform
ation rec
e
ived
whic
h is the maxim
u
m
for w×w sub
bloc
. If the calculated
value is
higher than the
m
a
x
we re
place it with the
m
e
di
an
val
u
e o
f
nei
g
hb
ori
n
g
i
ndi
ces
.
6.
NOISE
DETE
CTIO
N
Seve
ral
se
que
n
ces are
ge
ne
rat
e
d
d
u
ri
n
g
t
h
e i
m
pul
se det
ect
i
o
n
p
r
oced
u
r
e.
The
fi
rst
i
s
a s
e
que
nce
o
f
ori
g
i
n
al
i
ndi
ce
s
m
a
t
r
i
x
and i
t
s
vect
or m
eans, x(i
,
j) a
n
d M
x
(i
,j
) res
p
ect
i
v
el
y
.
(i
,j) i
s
posi
t
i
on
of i
n
di
ce, i
t
can b
e
1
≤
i
≤
M, 1
≤
j
≤
N
where M and
N are the
num
ber
of t
h
e indi
ces
in horiz
ontal a
n
d ve
rtical direction
resp
ectiv
ely, Th
e second
is a v
a
riation
m
a
trix
, f(Mx(i,j))
is
use
d
to indicat
e whethe
r
the indice at (i,j) in noisy
indices detecte
d
as noisy or noise free. T
h
e third is
a seque
nce of rece
ived indi
ces matrix and its
means,
x’
(i,j) a
n
d M
x
’
(
i,j)
respecti
v
ely
[9]
.
Fig
u
re
3
.
Blo
c
k
d
i
agram
o
f
propo
sed Filter
1.
Let
s
t
a
ke
a (
2p+
1
)
×(
2p+
1
)
wi
n
d
o
w
a
r
ou
n
d
M
x
(i
,
j
)
m
eans M
x
(i
+
k
,
j
+l
)
whe
r
e
-p
≤
(k
,l)
≤
p,
an
d p
≥
1
.
W
h
er
e
)
,
(
1
)
,
(
1
j
i
x
q
q
j
i
Mx
q
q
k
k
(1
)
x
k
i
s
t
h
e
g
r
ay
l
e
vel
o
f
c
o
de
ve
ct
or
pi
xel
.
q×
q
i
s
t
h
e
bl
oc si
ze
o
f
qua
nt
i
f
i
e
d
p
i
xel
s
.
2
.
Find
Med
i
an v
a
lu
e of
t
h
is
w
i
nd
ow
m
(
i,j
)
m
(
i,j)=M
edian
[
M
x
(i+
k
,
j
+l)]
(3
)
3.
Find a
b
s
o
lut
e
differe
n
ce
be
tween
M
x
(i,
j
)
and
m
(
i,j), a
n
d
assign
)
,
(
)
,
(
))
,
(
(
j
i
m
j
i
x
j
i
Mx
f
(3
)
4
.
F
i
n
d
t
h
e ma
x
i
mu
m v
a
l
u
e
o
f
f
f
o
r
1
≤
(i,j)
≤
w
5
.
Send
j
u
st t
h
is m
a
x
i
m
u
m
v
a
lu
es with
orig
i
n
al ind
i
ces
m
a
tri
x
,
x(i
,
j
)
.
we
have
N
V
=M
×
N
/
w
×
w
val
u
es.
Thi
s
is n
e
g
lig
ib
le.
6. i
n
the
recei
ver, Calculate
)
,
(
'
)
,
(
'
))
,
(
'
(
j
i
m
j
i
x
j
i
x
f
(4
)
Wi
t
h
:
m
’(i,j)=M
e
dia
n
[M
x’
(i+k,
j
+l)
]
(5
)
Input noisy ind
i
ces
Switch
Filtered
indices
Switching
mec
h
anis
m
base
d o
n
no
ise
det
e
ct
io
n
Median filterin
g
No
Filtering
Start
Calculating the
mean value for
each vector in
the codebook
Calculating the
complexity va
lue
f
o
r
eac
h
vect
o
r
in
t
h
e co
debo
o
k
Information of
co
debo
o
k
Evaluation Warning : The document was created with Spire.PDF for Python.
I
J
ECE
I
S
SN
:
208
8-8
7
0
8
Th
e N
o
ise Redu
ctio
n o
v
er Wi
reless Cha
n
n
e
l
Using
Vecto
r
Quan
tiza
tion
C
o
mp
ressi
o
n
and
…
(
I
m
a
n El
a
w
ady)
13
4
7
.
If f(Mx’(i,j
)) > m
a
x
[f(Mx(i,j))]
x
’
(i,j) is
detected
as
n
o
i
sy th
en
t
h
e esti
mated
v
a
lu
e of x’(i,j
) will
be m
odi
fi
ed
as
otherwise
j
i
x
erroneous
is
j
i
x
if
j
i
m
j
i
x
)
,
(
'
)
,
(
'
)
,
(
'
)
,
(
(6
)
After all th
at we m
a
k
e
th
e sa
m
e
alg
o
r
ith
m b
u
t
we
rep
l
ace th
e o
r
i
g
in
al in
d
i
ces m
a
trix
mean
Mx
(i,j)
b
y
orig
i
n
al indices m
a
trix
com
p
lex
i
t
y
Cx
(i,j).
C
x
(i
,
j
)=m
a
xk(
xk
(i
,
j
)
)
-
m
i
nk(
xk
(i
,
j
)
)
k=
1,.
.,
qx
q
(7
)
7.
R
E
SU
LTS AN
D ANA
LY
SIS
Fo
r t
h
e sim
u
la
tio
n
we
u
s
e fou
r
im
ag
es (lena, Bo
at, Go
ldhill, Pep
p
e
rs) si
ze 5
12x
512
, in
th
e gray
lev
e
l, th
e b
l
o
c
k
size is 4x4
, and
un
iv
ersal cod
e
boo
k si
ze is 256
x
1
6
,
th
e qu
ality of
recon
s
tru
c
ted imag
e
wi
t
h
o
u
t
t
r
a
n
sm
i
ssi
on
usi
n
g
L
B
G c
ode
b
o
o
k
and
S
O
M
c
ode
bo
o
k
i
s
rep
r
ese
n
t
e
d a
s
s
h
o
w
n
bel
o
w:
Tab
l
e
2
.
Th
e qu
ality o
f
reco
nstru
c
ted im
ag
e (d
B
)
L
B
G codebook
SOM
codebook
lena 30.
384
8
29.
703
4
Boat 28.
994
8
28.
096
1
Goldhill
29.
416
0
29.
032
5
Pepper
s
25.
879
8
25.
139
6
Fo
r tran
sm
issi
o
n
o
v
e
r wireless ch
ann
e
l the p
a
ram
e
ters
o
f
Gilb
ert-Ellio
tt
m
o
d
e
l are selected
as
indicated in ta
ble 1. The simulation
results
prese
n
t: received im
age, filtere
d, rec
o
vere
d i
ndi
ces, and rec
ove
re
d
in
d
i
ces with
filterin
g
.
Figure
4. Received and
restore
d
im
age (lena
)
(left with:
LB
G c
ode
book. R
i
ght
with
: SOM codebook)
10
-1
.
6
10
-1
.
5
10
-1
.
4
10
-1
.
3
10
-1.
2
10
-1.
1
16
18
20
22
24
26
28
30
m
ean B
E
R
PS
N
R
l
ena LB
G
rec
e
i
v
ed
f
i
l
t
ered
rec
o
v
e
ry
in
d
rec
o
v
e
ry
in
d
& f
i
l
t
e
r
10
-1
.
6
10
-1
.
5
10
-1
.
4
10
-1
.
3
10
-1.
2
10
-1.
1
19
20
21
22
23
24
25
26
27
28
m
ean B
E
R
PS
N
R
le
n
a
S
O
M
rec
e
i
v
ed
fi
l
t
e
r
e
d
rec
o
v
e
ry
in
d
rec
o
v
e
ry
in
d
& f
i
l
t
e
r
Evaluation Warning : The document was created with Spire.PDF for Python.
I
S
SN
:
2
088
-87
08
IJEC
E V
o
l
.
6, No
. 1, Feb
r
uar
y
20
1
6
:
13
0 – 13
8
13
5
Figure
5. Received and
restore
d
im
age (Boat) (left
with: LB
G c
ode
book. R
i
ght
with
: SOM codebook)
Figure
6. Received and
restore
d
im
age (Goldhill) (left
with:
LBG c
ode
book. Right with
: SOM
code
book)
Figure
7. Received and
restore
d
im
age (Pe
p
pers) (l
eft with: LBG
c
ode
book.
Ri
ght with:
SO
M code
book)
Before we commen
t
in
th
ese resu
lts we sho
u
l
d
see
th
e
b
a
ck
gro
und
s of
each
technique for
code
book
creat
i
o
n
an
d t
h
e rel
a
t
i
ons
hi
p
bet
w
ee
n c
ode
vect
o
r
s i
n
t
h
e
c
ode
w
o
r
d
.
10
-1
.
6
10
-1
.
5
10
-1
.
4
10
-1
.
3
10
-1
.
2
10
-1
.
1
17
18
19
20
21
22
23
24
25
26
m
ean B
E
R
PSN
R
Bo
a
t
L
B
G
re
c
e
i
v
e
d
f
ilt
e
r
e
d
re
c
o
v
e
ry
in
d
re
c
o
v
e
ry
in
d
& f
i
l
t
e
r
10
-1
.
6
10
-1
.
5
10
-1
.
4
10
-1
.
3
10
-1
.
2
10
-1
.
1
18
19
20
21
22
23
24
25
26
m
ean B
E
R
PSN
R
B
oat
S
O
M
re
c
e
i
v
e
d
f
ilt
e
r
e
d
re
c
o
v
e
ry
in
d
re
c
o
v
e
ry
in
d
& f
i
l
t
e
r
10
-1
.
6
10
-1
.
5
10
-1
.
4
10
-1
.
3
10
-1
.
2
10
-1
.
1
16
18
20
22
24
26
28
m
ean
B
E
R
PSN
R
G
o
l
dhi
l
l
LB
G
re
c
e
i
v
e
d
f
i
l
t
er
ed
re
c
o
v
e
ry
in
d
re
c
o
v
e
ry
in
d
& f
i
l
t
e
r
10
-1
.
6
10
-1
.
5
10
-1
.
4
10
-1
.
3
10
-1
.
2
10
-1
.
1
18
19
20
21
22
23
24
25
26
27
m
ean B
E
R
PSN
R
G
o
ld
h
i
ll S
O
M
re
c
e
i
v
e
d
f
ilt
e
r
e
d
re
c
o
v
e
ry
in
d
re
c
o
v
e
ry
in
d
& f
i
l
t
e
r
10
-1
.
6
10
-1
.
5
10
-1
.
4
10
-1
.
3
10
-1
.
2
10
-1
.
1
16
18
20
22
24
26
28
m
ean
B
E
R
PS
N
R
P
e
pper
s
L
B
G
re
c
e
i
v
e
d
f
i
l
t
er
ed
re
c
o
v
e
ry
in
d
re
c
o
v
e
ry
in
d
& f
i
l
t
e
r
10
-1
.
6
10
-1
.
5
10
-1
.
4
10
-1
.
3
10
-1
.
2
10
-1
.
1
18
19
20
21
22
23
24
25
26
27
m
ean B
E
R
PS
N
R
P
e
p
per
s
S
O
M
re
c
e
i
v
e
d
f
ilt
e
r
e
d
re
c
o
v
e
ry
in
d
re
c
o
v
e
ry
in
d
& f
i
l
t
e
r
Evaluation Warning : The document was created with Spire.PDF for Python.
I
J
ECE
I
S
SN
:
208
8-8
7
0
8
Th
e N
o
ise Redu
ctio
n o
v
er Wi
reless Cha
n
n
e
l
Using
Vecto
r
Quan
tiza
tion
C
o
mp
ressi
o
n
and
…
(
I
m
a
n El
a
w
ady)
13
6
As m
e
nt
i
one
d
i
n
[
2
5]
an
d
[
2
6]
t
h
e m
a
i
n
di
ffe
rence
bet
w
e
e
n LB
G a
n
d S
O
M
al
g
o
ri
t
h
m
s
i
s
i
n
t
h
e
or
der
of t
h
e c
ode
vect
o
r
i
n
d
e
xes. T
h
e LB
G al
go
ri
t
h
m
does n
o
t
de
fi
ne
any
or
der i
n
t
h
e code
b
o
o
k
;
t
h
e
code
w
o
r
d
s f
o
r t
h
e m
odel
vect
or ca
n be sel
e
c
t
ed arbi
t
r
a
r
i
l
y
. On t
h
e ot
h
e
r
h
a
nd t
h
e co
de
bo
ok t
r
ai
ne
d usi
n
g t
h
e
SOM algo
rithm h
a
s an
in
tern
al o
r
d
e
r,
ad
j
a
cen
t cod
e
boo
k en
tries. Th
is is
due t
o
the fa
ct that SOMs use t
h
e
m
e
thod of com
p
etitive learning [27]. Fo
r m
o
re details the articles [28] a
nd [29] provide m
o
re inform
ation
abo
u
t
di
ffe
re
nc
es i
n
c
o
deb
o
o
k
creat
i
o
n
bet
w
e
e
n LB
G a
n
d
S
O
M
al
g
o
ri
t
h
m
s
.
That’s why the sim
u
lation results
show that
SOM provi
d better res
u
lts for recei
ved image beca
use
ev
en we g
e
t an error in ind
i
ces th
e erron
e
ou
s in
d
e
x
will b
e
clo
s
e to th
e
orig
in
al
o
n
e
.
After th
e restoratio
n
p
r
o
cess
th
e co
m
b
in
atio
n
b
e
tw
een
ind
i
ces recov
e
ry an
d
stand
a
rd
med
i
an
filter
provide the bes
t
result howe
v
e
r
we can’t be
nefit on th
is im
provem
ent achieved in recei
ve
d im
age using
SOM
code
b
o
o
k
fo
r r
e
st
orat
i
o
n p
r
oc
ess. Th
e col
l
e
c
t
ed i
n
f
o
rm
at
i
on usi
ng
LB
G c
ode
b
o
o
k
fo
r i
n
di
ces rec
ove
ry
wi
l
l
be with hi
gh precision
because there is big diffe
re
nce
bet
w
een c
ode
vect
ors s
o
all
colle
cted inform
ation ca
n
speci
fy
t
h
e ri
g
h
t
bl
oc
k f
o
r re
cove
ry
p
r
oces
s
ho
we
ver i
n
S
O
M
co
deb
o
ok
t
h
e col
l
ect
ed i
n
f
o
rm
at
i
on can
’t
gi
ve
us
or
det
ect
a
s
p
eci
fi
c
bl
oc
k i
n
t
h
e
rec
o
very
pr
ocess
.
Also
we sh
ou
l
d
n’t forg
et th
at th
e q
u
a
lity o
f
recon
s
tru
c
ted i
m
ag
e u
s
in
g
LBG is b
e
tter th
an
SOM.
Th
ere are m
a
n
y
articles p
r
opo
se a so
l
u
tio
n
as sho
w
n
in
[30
]
we can
enhan
ce th
e
q
u
a
lity o
f
im
ag
e b
y
u
s
ing
Savi
t
z
ky
-
G
ol
ay
pol
y
n
o
m
i
al
; i
n
art
i
c
l
e
[3
1
]
t
h
e enha
nce
m
ent
can be a
c
hi
eve
d
by
usi
ng
o
p
t
i
m
a
l
Koh
o
n
en
to
po
log
i
cal m
a
p
to
d
e
term
in
e th
e op
ti
m
a
l co
d
e
boo
k
and
also
avo
i
d
t
h
e prob
lem
o
f
"d
ead
u
n
its" th
at can
arise
for ex
am
p
l
e with
th
e LBG al
g
o
rith
m
,
an
o
t
her sugg
estion
to
so
l
v
e th
is
p
r
o
b
l
em
is
m
e
n
t
i
o
n
e
d
in
articles [32
]
and
[
3
3]
by
a
p
pl
y
i
ng
hi
era
r
c
h
i
cal
SOM
.
Th
e
sim
u
lat
i
o
n
resu
lts prove th
at
we can enh
a
n
ce the
q
u
a
lity of
receiv
ed im
ag
e u
s
ing
SOM
co
d
e
bo
ok
also prov
id
e an efficien
t
restoratio
n
tec
h
n
i
q
u
e
u
s
i
n
g no
ise
redu
ction
an
d m
e
d
i
an
filter. Our
su
gg
estion
l
o
ok
sim
p
le an
d easy co
m
p
ared
with
o
t
h
e
r sugg
estion
s
wh
ich m
a
k
e
u
s
in
terested
to m
a
k
e
lo
t of
researc
h
base
d on
p
r
op
ose
d
s
o
l
u
t
i
on.
8.
CO
NCL
USI
O
N
In this pa
pe
r we provide a
suggestion tha
t
can
i
m
prove
the quality of receive
d image without
m
i
nim
i
ze t
h
e c
o
m
p
ressi
o
n
rat
i
o o
r
m
a
xim
i
ze t
h
e
ban
d
w
i
d
t
h
usa
g
e as
s
h
o
w
n i
n
[
2
8]
an
d
[
29]
.
O
u
r
s
u
g
g
e
s
t
i
on
depe
n
d
i
n
g
on
t
h
e rel
a
t
i
o
n
s
hi
p bet
w
ee
n p
r
o
duce
d
i
ndi
ces
i
n
t
h
e c
ode
vec
t
ors.
We al
s
o
t
e
st
t
h
e effi
ci
e
n
cy
of
r
e
stor
atio
n
i
n
th
is case to
b
e
n
e
f
it fr
o
m
th
e
m
a
x
i
m
u
m
i
m
p
r
ov
em
en
t ach
iev
e
d
.
Th
e
u
s
e o
f
SO
M co
d
e
b
ook
p
r
ov
id
es
g
ood q
u
a
lity o
f
receiv
e
d
im
ag
e co
m
p
ared
w
ith
LBG
cod
e
bo
ok
.
We also d
i
scov
ered
that th
e
co
m
b
in
atio
n
between
ind
i
ces recov
e
ry and
stan
d
a
rd
m
e
d
i
an
filter p
r
ov
id
e th
e b
e
st resu
lt
s. Howev
e
r, we stil
l
have a
pr
o
b
l
e
m
i
n
rest
orat
i
o
n p
r
oce
ss t
o
s
p
eci
fy
t
h
e col
l
ect
ed dat
a
i
n
S
O
M
co
deb
o
o
k
.
The s
o
l
u
t
i
o
n c
a
n be
by
usi
n
g m
e
t
hod
s
sh
o
w
n
i
n
[
3
0
]
, [3
1]
,
[
32]
a
n
d
[
3
3]
o
r
by
usi
n
g
t
r
a
n
sf
or
m
s
t
h
at
can a
n
al
y
ze t
h
e i
m
age t
o
i
m
p
r
ov
e th
e quality o
f
recon
s
tru
c
ted
im
ag
e an
d also h
e
l
p
u
s
to
co
llect in
fo
rmatio
n
with mo
re precision
.
REFERE
NC
ES
[1]
Jun
Chen,Tob
y
Berger .Robust Distributed
Sour
ce
Coding
.
I
E
EE Transactions on Information
Theory
. August
2008; 54(8); 338
5-3398.
[2]
S.K. Muttoo, S
u
shil Kumar .Robust Source Coding Steg
anog
raphic
Techn
i
q
u
e Using Wavelet Transforms.
International Jo
urnal of Inform
ation Techno
log
y
Bharat
i Vidya
p
eeth’s Institute of
Computer Applications
and
Management (
B
VICAM)
. New D
e
lhi, December,
2009; 1(2); 91-9
6
.
[3]
Michelle
Effros. Robustness to
Cha
nnel V
a
riation in Source C
oding for
Trans
m
ission Across
Nois
y
Chann
e
ls.
IEEE Internatio
nal C
onference on
Acoustics
, Speech
,
and Sign
al
Processing. 21-
24 Apr 1997;
4;
2961-2964.
[4]
Mannava Sriniv
asa Rao, Boppana Swa
ti Laksh
m
i, Dr.Panakala
Rajesh Kuma
r . Wireless Image Transmission over
Nois
y
Chann
e
ls
Using Turbo C
odes and De
-noi
sing Filters
.
International Jo
urnal of
Engin
eerin
g Research
and
Applica
tions
. September- October 2012; 2(5)
; 193
6-1941.
[5]
Laura
Toni, Yee Sin Chan, Pamela C.
Cosm
an, L
a
urenc
e
B.
Milst
e
in .Ch
a
nne
l Co
ding for Progres
sive Images in
a
2-D Time-Frequency
OFDM Block W
ith Chan
nel Estim
at
ion
Errors.
IEEE Transactions on Image Processin
g
.
November 2009; 18(11); 2476
-24
90.
[6]
Wei Xiang, Stev
en S. Pietrobon
and S. Adrian Barbulescu
.
Iterative Source-Channel D
ecoding
fo
r Robust Image
Transmi
ssion
. Australi
an Com
m
unica
tion
Theor
y
W
o
rkshop Pro
ceed
ings. 2003
;
05-07.
[7]
Ch. Rama Krish
n
a Redd
y
,
M.Padmaja,
Dr.
K
.
Sri Rama Krishna,
K.
Prasuna
.Imag
e
Transmission Using Adaptive
Channel Coding
for Electro
n
i
c
Toll Coll
ect
ion S
y
s
t
em
.
Interna
tional Journal o
f
Electronics
&
Communication
Technology (
I
JECT)
. June 2011;
2 (2); 60-65
.
[8]
Jianfei C
a
i, Ch
ang Wen Chen. Robust Joint
Source-Ch
annel Coding for Image Tr
ansmission Over Wireles
s
Channels.
IEEE Transactions
on Circuits
and
Sys
t
ems for Video
Technolog
y
. September 2000; 10
(6); 962-966
.
[9]
A. Moulay
Lakh
dar, M. Beladgh
am
,
M.
Ana
n
e,
Be
n Aissa
.
M N oise
reduction of
VQ Encoded Medical.
Computin
g
and Control
App
licat
ions (
CCCA)
IEEE Con
f
eren
ce
. Tun
i
sia M
a
rch 2011; 1-6
.
Evaluation Warning : The document was created with Spire.PDF for Python.
I
S
SN
:
2
088
-87
08
IJEC
E V
o
l
.
6, No
. 1, Feb
r
uar
y
20
1
6
:
13
0 – 13
8
13
7
[10]
Usama S.
Mohammed,
H.
A.
Ha
mada.
Im
age transmission
over
OFDM channel
with rate allocation scheme and
m
i
nim
u
m
peak t
o
aver
age
power
rat
i
o.
Journal of Telecommunica
tions
. May
2010; 2(2); 70-78
.
[11]
Cornelius
Hellg
e, Val
e
ntin
a P
u
l
l
ano, M
a
nu
el H
e
ns
el, Gi
ov
anni E. Corazza,
Th
omas Schierl,
Thomas Wiegand.
Mobile TV with long Time Interleaving and Fast
Zapping
.
IEEE International Co
nference on Multimedia and Exp
o
Workshops
. 2012.
[12]
Abdelmounaim Moulay
Lakhdar
,
Must
apha Khelifi, Mohammed Beladgh
am, Mohammed Ben
Ai
ssa,Abdesselam
Bassou. Im
age Vector Quanti
za
tion Codec Inde
xes Filtering
.
Serbian Journal of Electrica
l
Engin
eering
. June 201
2;
9(2); 263-277
.
[13]
Xiaofeng Wu, S
h
igang Hu,
Zhiming Li, Zh
ijun
Tang, Jin
Li
, Jin
Zhao. Compar
isons of Threshold
EZW and SPIHT
Wavelets Based
Imag
e Compression Methods.
TELKOMNIKA Indonesian
Journal
of Electrica
l Engineering
.
March 2014; 12
(3); 1895-1905
.
[14]
Ajol Kumar Ray,Tinku
Achar
y
a.
Information Technology
Princip
l
es and App
l
ications
. Prentic Hall of India,2004
.
[15]
N.M
.
Nas
r
abad
i, R
.
A. King.
Im
age coding
using
vector
q
u
antization: a review.
IEEE Transactions
on
Communications
. Aug. 1988; 36
(8); 957-971
.
[16]
Yun Q. Shi, Hui
f
ang Sun.
Image and Video Compression for Mul
timedia Engin
e
ering
. CRC Press, second edition
,
2008.
[17]
Zhou Wang
, A
l
an C
.
Bovik. M
ean squar
e
d
err
o
r love it or
leave it.
IEEE sign
al processing m
agazine
. Janu
ar
y
.
2009; 99-117.
[18]
Ronald G. Driggers.
En
cy
clop
edi
a
of Opt
i
cal
Eng
i
neer
ing
. Sep
1,
2003; 2(3).
[19]
Cheng-fa Tsai,
Yu-Jiun Lin.
Ad
vances in Image and Video Technology
. Pacific Rim Sy
mposium on I
m
age and
Video Techno
lo
g
y
(PSIVT). Tok
y
o
,
Japan
Januar
y
13-16, 2009.
[20]
Qiu Chen, Koji Kotani, Feifei Lee, Tad
a
hiro
Ohmi
. A Codeb
ook Design Me
thod for Robust VQ-Based Fac
e
Recognition Alg
o
rithm.
J. So
ftw
are Engin
eering
&
A
pplications
,
2010; 3(2); 119-
124.
[21]
Yoseph L, Andres B, Robert M.G. An
algorith
m for vector quantizer design
.
I
EEE T
r
ans, on Communications
.Januar
y
1980; 2
8
(1); 84-95.
[22]
Al Bovik.
Handbook of Image a
nd Vid
e
o
Proces
sing
. Acad
emic
Press, 2000.
[23]
Scott E
Umbaugh.
Computer
Imaging: Digital Im
age Ana
l
ysis an
d Processing
. C
RC Press. Jan 27
, 2005
.
[24]
Jianbin Xue, Songbai Li.Tra
nsmission Performance Res
earch
of
Digital
Modulation Signals in
AWGN Channel.
TELKOMNIKA Indonesian Journ
a
l
of Electrical
Engineering
. Februar
y
2013; 11(
2); 991-997
[25]
Abdelkader Amraoui, Pierre Du
four.
Robust Vector Quantizatio
n for No
isy Channels Using a Self-Organizing
Pseudo-Map
. R
obust Speech R
ecogni
tion for U
nknown Com
m
u
nica
ti
on Chann
e
ls Pont-a-Mousson, Franc
e
April
17-18, 1997; 13
5-138.
[26]
M. Ettaou
il, Y. Ghanou, K. El M
outaouakil, M. Lazaar
.Image Medica
l Compression b
y
A new Architectur
e
Optimizatio
n Model for the Koh
onen Networks.
International
Jo
urnal of Computer Theory and Engineering
. April
2011 ; 3(2)
;204:210.
[27]
Norbert Ádám
. A S
p
eech Anal
ys
is
S
y
s
t
em
Bas
e
d on Vector Q
u
anti
zat
ion Us
ing the LBG Alg
o
rithm
and S
e
lf
-
Organizing Map
s
.
Internationa
l
Journal of Com
puter and Information Technolog
y
. September 20
14 ; 03 (05); 952
-
957.
[28]
As
hraf R, Akba
r M
.
Adaptiv
e
Archite
cture
Ne
ural Nets
for
M
e
dica
l Im
age
Com
p
res
s
i
on.
IEEE In
ternation
a
l
Conferenc
e
on
E
ngineering
of
Int
e
llig
ent
.
S
y
stems
, 2006; 1-4.
[29]
Xu, Lixin, Liu,
W.Q., Venkates
h
, Svetha.
A tw
o stage vector quantization approach via self-or
ganizing map
. in
ICSP'
02 : 6th
In
ternational Conf
erence on
Signal Processi
ng pro
c
eedings I
EEE
Piscataway
N. J. 2
002;1;913-916.
[30]
Tomi Kinnunen, Teemu Kilpeläinen, Pasi
Fränti. Comparison of
Clustering Algo
rithms in Speaker Identif
ication
.
P
a
ttern
Recogn
it
ion L
e
tt
ers
ar
chi
v
e.
Els
e
vier
S
c
i
e
nce In
c.
New
Y
o
r
k
, N
Y
,
USA
.October 2011
; 32(1
3
); 1604-1617.
[31]
Mohamed Ettao
u
il,Mohamed Lazaar.
Compression of Medical Images us
ing Improved Koh
onen Algorithm.
Specia
l
Issue of Internationa
l
Journal of Co
mputer
Applications (
0975 – 8
887)
on Software Engineering
.
September 2012; 4 (6);111-1
17.
[32]
G Boopathi, Dr. S. Arockiasam
y
.
An Image Compression
Appr
oach using Wavelet Transform and Modified Self
Organizing
Map
.
IJCSI
International Journal
of
Computer Scien
ce
Issues. September 2011; 8
(5)
;
1694-0814.
[33]
Cheng-Fa Tsai, Chen-An Jhuang, Chih-Wei Liu.
Gray Image Compression
Us
ing New
Hierarchical Self-
Organizing Map Techniqu
e
.
The 3rd Intetnational Confer
ence on
I
nnovative Co
mputing Information
and Con
t
r
o
l
(ICICIC'
08)
IEEE, June 2008; 54
4.
BIOGRAP
HI
ES OF
AUTH
ORS
Im
an Elawad
y
was
born in Bechar, Alger
i
a
.
He
rec
e
ived th
e di
pl. Li
cens
e
from
the Univers
i
t
y
of Bechar
, Algeria,
in 2009 an
d the Master
d
e
gree
in dig
ital communication
s
y
stems from
University
of Bechar, Alger
i
a. in
2011. Actuelly
,
he prepare th
e doctoral degree
at University
of
Bechar
, Algeri
a. His
m
a
in interes
t
s
are I
m
age proces
s
i
ng trans
m
is
s
i
on, com
p
res
s
i
on,
electromagnetic field and an
tenna arr
a
y
.
Correspondence address:
Bechar University
,
Department o
f
Electronic,
Be
c
h
ar,
Al
geri
a,
Em
ail:
im
anel
aw
ad
y
2
@gm
a
i
l
.
c
o
m
Evaluation Warning : The document was created with Spire.PDF for Python.
I
J
ECE
I
S
SN
:
208
8-8
7
0
8
Th
e N
o
ise Redu
ctio
n o
v
er Wi
reless Cha
n
n
e
l
Using
Vecto
r
Quan
tiza
tion
C
o
mp
ressi
o
n
and
…
(
I
m
a
n El
a
w
ady)
13
8
A. Moulay
Lakh
dar got the
Engi
neering Degr
ee i
n
Telecom
m
unication in 2000
at t
h
e Institut
e
of
Tel
ecom
m
unicat
ions in Oran. Magister was the
s
econd degr
ee in
Signal and t
e
lecom
at Djillal
i
LIABES university
of Sidi Bel
Abbes in 2003.
Fr
om 2004
up p
r
esent, he worked in the Bechar
University
as lecturer
. Since M
a
y
2009, h
e
gradua
ted PhD Es
Sciences at
Sidi
Bel Abbes. He
do his
res
earch
act
ivit
y a
t
the B
echar Univ
ers
i
t
y
and Com
m
unications
, Arch
ite
ct
ure and M
e
di
a
Labora
t
or
y (CA
M
R) (Djilla
li
Li
abes Universi
t
y
)
.
His rese
arch
in
terests
are Im
ag
e tr
ansm
ission,
Image processing, and
digital transmission
performances. C
o
rrespondence
address: Bechar
Univers
i
t
y
,
Dep
a
rtm
e
nt of
E
l
ec
tr
onic,
Be
char
, Al
geria
,
Email: moulay
lakhdar78@
y
a
hoo
.fr
Khelifi Mustaph
a
was
born in Ai
n s
e
fra, Na
am
a.
He rec
e
ived
the
degree
in e
l
e
c
tri
cal
engin
eerin
g
from the Univ
er
sity
DR Moulay Tah
a
r Saida, Al
geria,
in 2009
and the Master
d
e
gree in
signal
and digital
communication from
University
of B
ech
ar
, Alger
i
a.in
2011. Actuelly
,
he prepar
es the
doctora
l degre
e
at Univers
i
t
y
of Bechar
, Alg
e
ria
.
His
m
a
in inter
e
s
t
s
are Im
age proc
es
s
i
ng,
compression, channel and
source
coding
.
Co
rrespondence address:
B
e
char Univers
ity
,
Department o
f
Electronic,
Be
c
h
ar,
Al
geri
a,
Email: Khelif
i_
m@y
a
hoo.fr
Evaluation Warning : The document was created with Spire.PDF for Python.