TELKOM
NIKA
, Vol.14, No
.4, Dece
mbe
r
2016, pp. 13
62~136
7
ISSN: 1693-6
930,
accredited
A
by DIKTI, De
cree No: 58/DIK
T
I/Kep/2013
DOI
:
10.12928/TELKOMNIKA.v14i4.3181
1362
Re
cei
v
ed
De
cem
ber 2, 20
15; Re
vised
Augus
t 16, 2
016; Accepte
d
Septem
ber 1, 2016
Interleaved Reception Method for Restored Vector
Quantization Image
I
m
a
n
E
l
aw
a
d
y*
1
, Abdelmounaim Moulay
Lakhdar
2
, Mustapha
Khelifi
3
1
Departme
n
t of Electrical En
gi
neer
ing, T
ahri Moham
m
ed U
n
iversit
y
, Bech
ar, Algeri
a
/ lab
:
CAOSEE
2,3
Department of Electrical En
gin
eeri
ng, T
ahri Mohamm
ed U
n
iversit
y
, Bech
ar, Algeri
a
BP 417 Ro
ute
Kena
dsa, Béch
ar 080
00, Alg
e
r
ia, +
213 49 2
3
89 93
*Corres
p
o
ndi
n
g
author, e-ma
i
l
: imane
la
w
a
d
y
2@gma
il.com
A
b
st
r
a
ct
The transmissi
on of i
m
a
ge co
mpr
e
sse
d by vector
qua
nti
z
a
t
i
on pro
duc
e w
r
ong b
l
ocks i
n
receiv
e
d
imag
e w
h
ich
a
r
e co
mp
letely
different to th
e
origi
n
a
l
o
ne w
h
ich
makes th
e restorati
on
p
r
ocess too
diffi
cult
beca
u
se w
e
d
o
n
’
t h
a
ve
any
i
n
formatio
n
a
b
o
u
t the or
igi
nal
block. As
a sol
u
tion w
e
prop
o
s
e a tra
n
smissi
on
techni
qu
e that save the
maj
o
rity of
pixels
in the ori
g
i
nal
block by b
u
i
l
di
ng new
bl
ocks
doesn'
t conta
i
n
nei
ghb
orh
ood
pixels
fro
m
th
e
ori
g
in
al
bl
ock
w
h
ich i
n
cr
eas
e
the
pro
bab
ility
of restor
atio
n.
Our pro
positi
o
n is
base
d
on
dec
ompos
ition
an
d interl
eav
ing.
F
o
r the simul
a
tion w
e
use
a bin
a
ry sy
mmetric c
han
ne
l
w
i
th
different BE
R
and
in
the
rest
oratio
n pr
oces
s w
e
use
si
mp
le
me
di
an fi
lter
just to
ch
eck
the effici
ency
of
prop
osed appr
oach.
Ke
y
w
ords
: De
compos
ition, In
terleav
ing, BS
C Cha
nne
l, Media
n
F
ilter, Vector Quanti
z
a
t
io
n
Copy
right
©
2016 Un
ive
r
sita
s Ah
mad
Dah
l
an
. All rig
h
t
s r
ese
rved
.
1. Introduc
tion
The redu
ctio
n of bit rate
rece
ntly gets l
o
t of
attention espe
cially
with the bi
g a
m
ount of
data that con
s
ume th
e ba
ndwi
d
th usag
e and
co
m
p
u
t
ation re
sou
r
ce
s. Many te
chni
que
s hav
e
been p
r
op
ose
d
to solve this proble
m
as
shown in the a
r
ticle
s
[1-3].
In imag
e
co
mpre
ssion,
q
uantization i
s
us
ed to
red
u
ce
the
num
ber of
symb
ols
and
hen
ce le
ss th
e amo
unt of i
n
formatio
n (compressi
o
n
)
need
s to b
e
e
n
co
ded. In
da
ta tran
smissi
on,
a seri
ou
s pro
b
lem face
d the image en
co
ded with
vect
or qua
ntizatio
n. Since the degradatio
n wil
l
be in fo
rm o
f
blocks. M
a
ny of publi
s
hed p
ape
rs
try to solve t
h
is p
r
obl
em
by com
b
inati
on
betwe
en the
VQ and
different tra
n
sfo
r
ms a
s
sho
w
n in arti
cle
s
[4-6]. The
s
e
combi
nation
s
of
those relative
ly basic meth
ods utili
ze the
favorable ch
ara
c
teri
stics
of each meth
od [7].
As propo
sitio
n
to solve thi
s
p
r
oble
m
, we are
goin
g
t
o
ma
ke
some
modificatio
n
i
n
origi
nal
image
by con
s
tru
c
ting
ne
w blo
c
ks th
at d
oesn't
cont
ai
n
any neig
h
b
o
rho
od pixels
this modification
give us a ch
a
n
ce to ke
ep
some pixel
s
in the or
iginal
block whi
c
h
make the restoration process
easi
e
r because we decrease the pr
obability of getting
wrong block.
Our
pape
r i
s
orga
nized a
s
follo
ws. In
Section 2
a
n
overvie
w
o
f
vector q
u
a
n
tization
comp
re
ssion.
In se
ction 3
rep
r
e
s
entati
on of di
gital
cha
nnel m
o
d
e
l whi
c
h i
s
b
i
nary symm
etric
c
h
annel. Sec
t
ions
4 to 5 des
c
ribe the main s
t
ep
s
of our propos
ition us
ing the decompos
i
tion and
interleavin
g. Section 6 the
n
com
b
ine
s
the re
sult
s usi
ng differe
nt BER with thre
e image
s Le
na
,
boat and G
o
l
dhill. Finally, Section 7 p
r
e
s
ent
s our
con
c
lu
sion
s an
d some p
r
o
p
o
s
i
t
ions.
2. Vector Qu
antization
Shanno
n first sugge
sted
that enco
d
in
g a seq
uen
ce of sample
s from a so
urce ca
n
provide
bette
r re
sult tha
n
encoding i
n
dividual sam
p
les i
n
term
s of comp
re
ssi
on effici
en
cy
[8].Image data com
p
ression using vect
or quantization (VQ)
has
received a lot of attention. Since
VQ ha
s
a
hig
h
codin
g
effi
ciency
and
simple
de
code
r a
r
chitectu
re
, it is ve
ry
su
itable fo
r lo
w-bit
rate appli
c
ati
ons. Th
e gen
eral VQ alg
o
ri
thm has three
main step
s [9]:
1.
First the ima
g
e
is partition
e
d
into blocks
whi
c
h are usually 2×2, 4
×
4, 8×8, and 1
6
×1
6.
2. After the
division
into
blocks, a
co
debo
ok
whi
c
h re
presents the b
e
st
est
i
mation of
all
the
blocks of the image is
con
s
tructed
and in
dexed.
Evaluation Warning : The document was created with Spire.PDF for Python.
TELKOM
NIKA
ISSN:
1693-6
930
Interlea
ved
Rece
ption Met
hod for Resto
r
ed
Ve
ctor Q
uantization Im
age (Im
an
Elawad
y)
1363
3.
Finally, the image bl
ocks are
substituted
by an index of best
estimation
code from the
cod
ebo
ok.
Figure 1. Vector quanti
z
ati
o
n
The ba
si
c pri
n
cipl
e of vect
or qu
antization ba
sed im
age
comp
re
ssion te
chni
qu
es i
s
to
match
each i
nput vecto
r
with a
cod
e
-v
ector in
the
codeb
oo
k. Th
e disto
r
tion b
e
twee
n the i
nput
vector a
nd th
e cho
s
e
n
co
d
e
-vecto
r is mi
nimum [8
]. Quantization is
an irreversibl
e
pro
c
e
s
s (th
e
re
is no way to find the origi
nal value fro
m
the
quanti
z
ed valu
e) [1
0]. The difference betwee
n
the
input and o
u
tput sign
als of
the
quantizer becom
es the
quantizi
ng er
ror, o
r
quanti
z
ing n
o
ise [11].
3. Transmiss
i
on Chann
e
l
The bi
na
ry sy
mmetric
chan
nel (BS
C
) is
defined
by th
e chan
nel
dia
g
ram
sho
w
n i
n
Figu
re
2, and its ch
a
nnel matrix is
given by Equation (1
):
P
YX
⁄
1
p
p
p1
p
(
1
)
The ch
ann
el has two input
s (x
1
= 0,
x
2
= 1) and two o
u
tputs (y
1
= 0
,
y
2
= 1). The cha
nnel
is
sym
m
etri
c becau
se
the
prob
ability
of receiving a
1
if a 0 is sent i
s
the
sa
me a
s
the
pro
babil
i
ty
of re
ceiving
a
0 if a
1 is se
nt. This
com
m
on tran
sition proba
bility is d
enoted
by
p [12]. The
erro
r
events are also inde
pen
de
nt of t
he data bits [13]. This is the simpl
e
st model of a chan
nel wit
h
errors, yet it captu
r
e
s
mo
st of the comp
lexity
of the
gene
ral p
r
obl
em [14]. The
cap
a
city of this
cha
nnel give
n by Equation
(2):
C1
H
p
in
(
2
)
With the bina
ry entropy fun
c
tion given by
Equation (3
):
H
p
p
l
o
g
p
1
p
log
1p
(
3
)
Figure 2. Binary symmetri
c
ch
ann
el
Evaluation Warning : The document was created with Spire.PDF for Python.
ISSN: 16
93-6
930
TELKOM
NIKA
Vol. 14, No. 4, Dece
mb
er 201
6 : 1362 – 136
7
1364
4. Decompo
s
ition
We
su
gge
st
decompo
sin
g
the
origi
nal i
m
age
into
16
imag
es an
d
recon
s
tru
c
tin
g
a
ne
w
image with t
hose image
s as it is illustrated in Fi
gure 3. Whi
c
h ma
ke th
e pixels in the
recon
s
tru
c
ted
image not ne
ighbo
rho
od pi
xels.
Figure 3. De
compo
s
ition of
image
As sho
w
n o
n
Figure 3
we
take an exa
m
ple of 4 blocks
size 4×4 (16 pixel
s
for ea
ch
block).
The
first st
ep
of de
compo
s
ition
wi
ll produ
ce
a
n
e
w
blo
c
k si
ze
4×4
co
ntains 4
pixels fro
m
each o
r
igin
al block. T
he n
e
xt step of
d
e
com
p
o
s
ition
pro
d
u
c
e a
n
e
w bl
ock
si
ze 4
×
4
contai
ns
1
pixel from ori
g
inal blo
c
k.
5. Interleav
in
g
In gen
eral
th
e interl
eavin
g is on
e of t
he mo
st i
n
te
restin
g te
chn
i
que
s u
s
e
d
i
n
lot of
appli
c
ation
s
like stora
ge,
error co
rre
c
tion,
an
d multidimen
sio
nal
data structure.
We use
interleavin
g i
n
col
u
mn
s he
re to en
han
ce the qu
ality of the image
after de
com
p
osition
as
sh
own
in the Figure 4:
Figure 4. Interleaved im
ag
e
Gene
rally in transmissio
n of image co
mpre
ssed by
the vector quantization the noise
attack th
e wh
ole blo
c
k. Th
e use of the restoration
in t
h
is
ca
se be
co
mes u
s
el
es
s
or mo
re dif
f
i
c
u
lt
.
The kno
w
led
ge of blo
ck’
s
pixels po
se
s
a big challe
n
ge. The id
ea
here i
s
to rea
rra
nge o
r
cre
a
te
blocks with
n
o
neig
hbo
rh
o
od pixel
s
fro
m
the o
r
ig
in
al
image. T
h
is
will de
crea
se
the proba
bility of
obtainin
g
noi
sy blo
c
ks i
n
the re
ceive
d
image,
a
nd i
n
crea
se th
e
resto
r
atio
n p
r
obability for t
he
noisy pixel
s
. To achieve this aim
we u
s
e the d
e
co
mpositio
n an
d interle
a
ving
as sho
w
n in
the
Figure 5.
Evaluation Warning : The document was created with Spire.PDF for Python.
TELKOM
NIKA
ISSN:
1693-6
930
Interlea
ved
Rece
ption Met
hod for Resto
r
ed
Ve
ctor Q
uantization Im
age (Im
an
Elawad
y)
1365
Figure 5. Pro
posed flowch
art
After applyin
g
the
de
com
p
osition
an
d in
terleav
ing
(4,128) a
code
b
ook will
be
g
enerate
d
to encode the 16 im
ages af
ter transmi
s
si
on the
de
coded
will reconstruct th
e 16 i
m
ages then
we
invert the interleaving a
nd
comp
ose the data fina
lly we make a restoration u
s
ing
simple medi
a
n
filter the obtained re
sult
s are
pre
s
ente
d
in the next se
ction
6. Results a
nd Analy
s
is
For
simul
a
tio
n
we
u
s
e g
r
a
y
level image
s, si
ze
512
×5
12, co
ded
in
8 bits, tra
n
smitted in
binary
symm
etric
cha
nnel
, comp
resse
d
by ve
ctor quantization
with blo
c
k size
4×4 a
n
d
cod
ebo
ok g
e
nerate
d
by LBG algorithm.
For the
re
sto
r
ation we use
a standa
rd m
edian filter.
Original image
Encod
ed ima
ge
PSNR=29.1
5
3
dB
PSNR = 2
5
.0
23dB at
BER=
1
0
-2
PSNR = 2
8
.8
90dB
PSNR = 2
7
.4
85 dB at
BER=
1
0
-2.5
PSNR = 2
9
.5
34dB
PSNR = 2
8
.5
37 dB at
BER=
1
0
-3
PSNR =29.8
3
dB
Figure 6. Restored lena im
age at different BER
using
median filter.
(Left: Degraded. Right:
Re
store
d
)
Evaluation Warning : The document was created with Spire.PDF for Python.
ISSN: 16
93-6
930
TELKOM
NIKA
Vol. 14, No. 4, Dece
mb
er 201
6 : 1362 – 136
7
1366
Original image
En
c
o
de
d
imag
e
PSNR = 2
7
.3
73 dB
PSNR = 2
4
.0
36 dB at
BER=
1
0
-2
PSNR =27.0
25 dB
PSNR = 2
5
.9
12dB at
BER=
1
0
-2.5
PSNR = 2
7
.5
24 dB
PSNR = 2
6
.8
19 dB at
BER=
1
0
-3
PSNR = 2
7
.7
10 dB
Figure 7. Restored boat image at
different BER using
median f
ilter (Left: Degraded. Right:
Re
store
d
)
Original image
Encod
ed ima
ge
PSNR = 2
8
.1
72 dB
PSNR = 2
4
.4
54 dB at
BER=
1
0
-2
PSNR = 2
7
.6
28 dB
PSNR = 2
6
.4
08 dB at
BER=
1
0
-2.5
PSNR = 2
8
.1
46 dB
PSNR = 2
7
.5
27 dB at
BER=
1
0
-3
PSNR = 2
8
.2
41 dB
Figure 8. Restored Goldhill
image at different BER
using median filter (Left: Degraded. Right:
Re
store
d
)
The simul
a
tio
n
result
s sho
w
re
ceived
i
m
age
s (Le
n
a
,
boat
a
nd G
o
ldhill)
in different
BER
(10
-2
, 10
-2.5
and 10
-3
).
The propo
se
d approa
ch i
s
ba
sed o
n
image de
co
m
positio
n and i
n
terleavin
g in
orde
r to
spread th
e n
o
isy pixels in
different blo
c
ks
whi
c
h kee
p
som
e
origi
n
al pixels in th
e noise blo
c
k. In
Evaluation Warning : The document was created with Spire.PDF for Python.
TELKOM
NIKA
ISSN:
1693-6
930
Interlea
ved
Rece
ption Met
hod for Resto
r
ed
Ve
ctor Q
uantization Im
age (Im
an
Elawad
y)
1367
this case, the pixel recov
e
ry
will be easy. The effi
ciency of
restoration show
s a good results
esp
e
ci
ally with BER=10
-2
by using
sim
p
le medi
an filter. Due to
q
uality of reco
nstru
c
ted i
m
a
g
e
us
ing
vec
t
or quantiz
ation c
o
mpress
ion we c
an’t
get the same improvement at B
E
R=
10^-2.5 and
BER=
1
0
-3
.
Th
e vecto
r
q
uan
tization
com
p
ressio
n d
epe
nds on
the
si
milarity of the
regi
on
s it ta
kes
some
sampl
e
s to
rep
r
e
s
ents e
a
ch re
gion
(code
bo
ok). In
case
of cha
ngin
g
this si
milarity
of
regio
n
s the
sele
ction
of
sampl
e
s will
be m
o
re di
fficult and
n
o
t perfe
ct
which
ma
ke t
h
e
recon
s
tru
c
ted
image d
egraded, ho
wev
e
r the resto
r
ation be
com
e
s e
a
sie
r
b
e
c
au
se
we h
a
v
e
some info
rma
t
ion about th
e nature of o
r
iginal bl
ock, so we sh
ould
balan
ce bet
wee
n
the qu
ality
of reconstruct
ed image and the efficiency
of the
restoration, howev
e
r our
approach i
s
still sim
p
le
and d
o
e
s
n’t
con
s
um
e lot
of processi
ng resource
s com
p
a
r
ed
with othe
r
re
sea
r
che
r
[15
-
16].
Some pro
p
o
s
itions is g
o
in
g to be provi
ded in con
c
l
u
sio
n
that ca
n make our
prop
ositio
n m
o
re
effic
i
ent.
7. Conclusio
n
The propo
se
d solutio
n
is
not perfe
ct a
s
sh
ow
n f
r
om
t
he simulat
i
o
n
re
sult
s
so
we mu
st
take ou
r ch
oi
ce bet
wee
n
the quality of image an
d t
he efficien
cy of restor
ation (we mu
st use
the
approa
ch
wisely). As prop
osition m
a
yb
e we
can
opt
i
m
ize in th
e d
e
com
p
o
s
ition
by using
artif
i
cial
intelligen
ce
a
l
so ve
ry so
p
h
isticated filters can
b
e
use
d
in th
e
data receiver to enh
an
ce
the
quality of image or may be
by chan
ging t
he way of co
mpre
ssion.
Referen
ces
[1]
Suqi
ng Z
han
g, Aiq
i
an
g W
a
n
g
. An Ima
g
e
Compress
io
n
Method
Bas
e
d
on
W
a
vel
e
t T
r
ansform
and
Neur
al N
e
t
w
or
k.
T
E
LKOMNIKA T
e
leco
mmunic
a
tion
Co
mputin
g El
ectron
ics an
d Co
ntro
l.
201
5; 13(
2):
587-
596.
[2]
Bo W
ang, Yubi
n Gao.An
Image Com
p
ressi
on Sch
e
me Base
d on F
u
zz
y
N
e
ural Net
w
o
r
k.
T
E
LKOMNIKA T
e
leco
mmunic
a
tion C
o
mputi
n
g Electron
ics a
nd Co
ntrol
. 20
15; 13(1): 1
37-
145.
[3]
M Masudur R
ahma
n
, Moha
mmad Motiur
Rahm
an.Efficie
n
t Image Co
mpressio
n
T
e
chni
que us
in
g
JPEG2000
w
i
t
h
Ad
aptive
T
h
resho
l
d.
Intern
a
t
iona
l Jo
urna
l o
f
Imag
e Proc
es
sing (IJIP)
. 20
1
5
; 9(3): 1
66-
174.
[4]
Juan
li Hu, Jia
b
i
n
Den
g
, Jueb
o W
u
. Image Co
mpressio
n
Bas
ed on Improv
e
d
F
F
T
Algorithm.
Journal of
Netw
orks
. 201
1; 6(7): 104
1-1
048.
[5]
T
e
jas S Patel, Ravindr
a Modi, Ke
yur J
Pa
tel. Image
Compressi
on
Using DW
T
and Vector
Quantizati
on.
Internatio
na
l Journ
a
l of Inn
o
vative R
e
se
arch in C
o
mputer a
nd Co
mmu
n
icati
o
n
Engi
neer
in
g
. 2013; 1(3).
[6]
Moham
ed El
Z
o
rkan
y
.
A H
y
brid Imag
e Co
mpressio
n
T
e
chni
que Us
in
g Neur
al Net
w
o
r
k and Vector
Quantizati
on
W
i
th DCT
. Springer Inte
rn
atio
nal Pu
blis
hi
ng. 201
4: 233-
244.
[7]
HR W
u
, KR Rao. Digita
l
Vide
o
Image Q
ualit
y
and Perc
eptu
a
l
Codi
ng. T
a
ylor
& F
r
ancis
Group. 200
6.
[8]
Ajol Kum
a
r Ra
y, T
i
nku Achar
ya. Informati
on
T
e
c
hnol
og
y P
r
incip
l
es a
nd A
pp
lic
atio
ns. Prentice H
a
ll of
India. 20
04.
[9]
NM Nasra
bad
i
,
RA King. Image co
di
ng
usin
g vector
qua
ntizati
on: a revie
w
.
IEE
E
T
r
ans, on
Co
mmun
icati
o
ns
. 1988; 3
6
(8)
:
957-97
1.
[10]
Y Yun,
Q Shi,
Hu
ifang
Su
n. Image
a
n
d
V
i
de
o C
o
mpr
e
s
s
ion
for Mu
lti
m
edi
a En
gi
ne
erin
g. Sec
ond
editi
on. CRC P
r
ess. 2008.
[11]
Z
hou W
a
ng, Al
an C
Bov
i
k. Mean
squ
a
re
d e
rror lov
e
it
or l
eave
it.
IEEE
s
i
gnal process
i
ng m
a
ga
z
i
ne
.
200
9; 26(1): 98
-117.
[12]
H
w
ei
Hsu. An
alo
g
an
d di
git
a
l commu
nic
a
tions
.sch
aum’s.
Secon
d
ed
iti
on. McGra
w
-H
ill Ed
ucati
on.
200
3.
[13]
Milan S
onk
a, Vaclav H
l
av
ac,
Roger B
o
yle.
Image Proce
ssing An
al
ysis
and Mac
h
in
e
Vision. T
h
ird
editi
on. T
H
OMSON. 2008.
[14]
John Mi
ano. C
o
mpress
ed Image F
i
l
e
F
o
rma
ts. Addison W
e
sle
y
.
199
9.
[15]
Junp
ing W
a
ng
, Dapen
g Z
h
a
ng. Image D
e
noisi
ng B
a
sed
on Artificial
Bee Co
lo
n
y
a
nd BP Ne
ural
Net
w
ork.
TEL
K
OMNIKA Teleco
mmu
n
icati
o
n Co
mp
utin
g Electron
ics an
d Contro
l
. 201
5; 13(2): 61
4-
623.
[16]
K F
r
eema
n
, M
Re
icher.
A N
o
vel
Ap
proac
h
for Imag
e
De
noisi
ng
Usi
n
g
Digita
l
Fi
lters.
Internatio
na
l
Globa
l Journ
a
l
for Engin
eeri
n
g
Researc
h
. 201
5; 13(1): 21-2
4
.
Evaluation Warning : The document was created with Spire.PDF for Python.