TELKOM
NIKA
, Vol.14, No
.1, March 2
0
1
6
, pp. 187~1
9
4
ISSN: 1693-6
930,
accredited
A
by DIKTI, De
cree No: 58/DIK
T
I/Kep/2013
DOI
:
10.12928/TELKOMNIKA.v14i1.2739
187
Re
cei
v
ed Se
ptem
ber 28, 2015; Revi
se
d De
ce
m
ber
30, 2015; Accepted Janu
ary 16, 201
6
Image Restoration Algorithm Based on Artificial Fish
Swarm Micro Decomposition of Unknown Priori Pixel
Dan Sui*
1,2
, Fang He
2
1
School of Infor
m
ation Sci
enc
e and T
e
chno
l
o
g
y
, W
uhan
U
n
iversit
y
of T
e
chno
log
y
, H
ube
i
4300
70, Ch
ina
2
School of Softw
a
r
e En
gin
eer
i
ng, An
yan
g
No
rmal Univ
ersit
y
, An
y
a
ng 4
550
00, Hen
an, Ch
i
n
a
*e-mai
l
: 462
88
112
9@q
q
.com
A
b
st
r
a
ct
In this p
aper,
w
e
put forw
ard a
new
method
to h
o
lo
grap
hic rec
o
n
s
truct image
that pri
o
r
infor
m
ati
on, modu
le match
i
n
g
an
d
e
dge structur
e infor
m
ati
on
is un
know
n. T
he
prop
osed
i
m
a
g
e
hol
ogra
p
h
i
c re
storation
alg
o
r
i
thm c
o
mbi
nes
artifi
cial fish
sw
arm micro
deco
m
positi
o
n
and br
ightn
e
ss
co
mp
en
sa
ti
on
. Th
e
tra
d
i
ti
ona
l
m
e
th
od
u
s
es su
b
s
pa
ce fe
ature inf
o
rmati
on of
multi
d
i
m
ensi
ona
l searc
h
meth
od, it is faile
d to achi
ev
e t
he fine stru
cture infor
m
ati
on of imag
e texture template
matchi
ng a
n
d
the
effect is not well. T
heref
or
e, it is difficult to holo
g
rap
h
ic r
e
construct the
unknow
n pix
e
ls. T
h
is w
eakness
obstructs the app
licati
on of i
m
a
ge re
stor
ati
on to ma
ny fields. T
herefor
e, w
e
builds a
structure texture
cond
uctio
n
mo
del f
o
r the
pri
o
rity det
ermin
a
t
ion
of t
he
blo
ck that to
be r
epa
ired, th
en
w
e
use s
ubsp
a
c
e
feature infor
m
a
t
ion multid
i
m
e
n
s
ion
a
l searc
h
meth
od to
the
confid
enc
e up
dates of
unkn
o
w
n
pixel. In order
to mai
n
tain
the
conti
nuity
of d
a
mag
ed r
e
g
i
on
in
i
m
a
ge, th
e
artificial
fish
s
w
arm a
l
g
o
rith
m
dec
o
m
pos
iti
o
n
mo
de
l is co
mbi
ned w
i
th the i
m
age br
ig
htness
compens
atio
n strategy of edg
e featur
e. T
he
simulati
on res
u
lt
sh
o
w
s th
a
t
i
t
h
a
s
a
g
o
o
d
vi
su
a
l
e
ffe
ct in
im
ag
e
re
sto
r
a
t
io
n
o
f
a
p
r
i
o
ri
u
n
k
no
wn
p
i
xel
,
re
co
ve
ry tim
e
and
computati
on co
sts are less,
the stabil
i
ty and
conver
genc
e p
e
rformanc
e is i
m
pr
ove
d
.
Ke
y
w
ords
: artificial fish swarm
,
im
age restor
ation, subs
pac
e, brightn
e
ss compe
n
satio
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
Image resto
r
ation play
s a
n
impo
rtant role in
the fiel
d of digital i
m
age p
r
o
c
e
s
sing. Th
e
obje
c
tive of recove
ry is th
at can be
achieved
for th
e re
storatio
n and p
r
e
s
ent
of loss
of image
informatio
n a
nd valu
able i
n
formatio
n. Evidently, a
pri
o
ri u
n
kno
w
n
pixel imag
e restoration i
s
o
f
great
significance in imag
e rep
r
od
uctio
n
and info
rm
ation re
cove
ry. Image restoration
，
whi
c
h
accords in a
certai
n criteri
a
and alg
o
rit
h
m, fills
and
sho
w
s image
information t
hat is lost an
d
remove
d thro
ugh the visio
n
prin
ciple
s
o
f
human.
In real life, age-o
l
d image nee
d resto
r
ation.
In
the field of target dete
c
ti
on, fuzzy re
mote se
n
s
in
g
image information and l
a
cking in a certain
angle
charact
e
risti
c
s al
so
need
im
ag
e
restoration. T
herefo
r
e, th
e
image
re
sto
r
ation te
chn
o
l
ogy
have impo
rt
ant appli
c
ati
ons i
n
the
prote
c
tion
of
cultural relics
paintin
gs,
medical im
age
informatio
n p
e
rception, vid
eo effe
ct de
sign, imag
e o
b
ject
re
cog
n
ition an
d remo
te are
a
s. In t
h
e
study of im
a
ge resto
r
atio
n, it is
difficu
lt to
rep
a
ir i
m
age
pixel
with u
n
kno
w
n texture fe
a
t
ure.
Therefore, th
e prio
ri un
kn
own pixel
re
storati
on al
gorithm with so
me cutting
-e
dge an
d practical
signifi
can
c
e i
s
the focu
s of
sch
olars [1].
Curre
n
tly, resea
r
ch on i
m
age resto
r
a
t
ion is
still in
its infancy, the key tech
nologi
es
related
alg
o
ri
thm are n
o
t
perfe
ct, man
y
of them
are
used i
n
im
a
ge
re
storatio
n focusi
ng
o
n
the
establi
s
hm
en
t of a non-technical texture
image re
storation model
by enhan
ci
ng
the contra
st of
the image, a
nd then extracting hi
ghlig
hts mod
e
l an
d texture ch
ara
c
teri
stics
of the image
for
image
re
stora
t
ion. Among t
hem, the literature [2] p
r
op
ose
s
a
cla
s
si
c resto
r
ation
algorith
m
ba
sed
on texture
hig
h
lights, thi
s
al
gorithm
sele
cts a
pixel o
n
t
he e
dge
of th
e dam
age
d a
r
ea
at first, a
n
d
then u
s
ing th
e template m
a
tchin
g
si
ze
and fa
st matching meth
od t
o
rep
a
ir the
d
a
mage
d area
. It
is
simple
a
n
d
practi
cal, b
u
t
the a
c
curacy is
not
hi
gh.
The
literatu
r
e [3] p
r
op
ose
s
a
resto
r
atio
n
method ba
se
d on image chara
c
te
rist
i
c
s preprocessin
g
prior m
odel
for an image
of the damage
d
area
and it
doe
s not ch
oose the mo
st simila
r blo
ck m
a
tchi
ng
to compl
e
te
the rep
a
ir. T
h
is
algorith
m
en
han
ced the visual bette
r g
a
ins,
but hug
e comp
utatio
n and co
mpl
e
x algorithm
are
the limitation
s
of th
e met
hod. Th
e lite
r
ature [4]
u
s
es
affine Ma
rkov
ra
ndom
pixel-level
to fill
image re
sto
r
ation, the re
pair is fea
s
ib
le on a
singl
e object, but
algorithm
s consume a lot
o
f
Evaluation Warning : The document was created with Spire.PDF for Python.
ISSN: 16
93-6
930
TELKOM
NIKA
Vol. 14, No. 1, March 2
016 : 187 – 1
9
4
188
memory spa
c
e in the mo
del and com
p
lex. The literature [5] pro
poses a
n
image completi
on
algorith
m
b
a
sed o
n
fragme
n
ts, kno
w
n
a
s
a
pa
rt of
th
e ima
ge train
i
ng
set to
inf
e
r th
e u
n
kno
w
n
parts
of the iterative app
roximation of an un
kno
w
n
regio
n
and
a
daptive co
mp
osite ima
ge
of
debri
s
.
Ho
we
ver, this
algo
rithm i
s
si
mil
a
r
whe
n
lo
oking for de
bri
s
with a
full
search
metho
d
,
resulting
i
n
sl
ow re
storatio
n
s
pee
d, an
d
greatly limits its p
r
a
c
ti
cal
appli
c
ation. T
he literature [
6
]
gives a fa
st restoration al
g
o
rithm, althou
gh to so
me e
x
tent, accele
rating the re
st
oration
rate,
but
it affects re
storative effect
s. The literat
ure [7
] pro
p
o
s
e
s
a linea
r additive of gradient an
d the
gradi
ent of the loga
rithmic
function to d
e
c
ide to
repai
r the block, int
r
odu
cin
g
the
spa
r
se de
gre
e
of large
amo
unt of cal
c
ula
t
ion, thus lea
d
ing to
lo
w e
fficiency. The
literature [8]
use
s
sub
s
pa
ce
feature info
rmation multid
imensi
onal
search m
e
tho
d
for ima
ge restoration. B
u
t it has faile
d to
template mat
c
hin
g
and joi
n
t decom
po
si
tion, image re
storatio
n re
su
lt is bad.
To solve the
s
e pro
b
lem
s
, this pa
per
pro
poses a
re
sto
r
ation alg
o
rith
m based on
artificial
fish d
e
comp
o
s
e
and
p
r
iori
of u
n
kno
w
n
holog
rap
h
ic
pixels, a
nd it
extra
c
t wea
k
info
rmatio
n
by
decompo
sin
g
the image t
o
obtain
a priori. Thi
s
al
g
o
rithm u
pdat
e the pri
o
ri o
f
unkn
o
wn pi
xel
confid
en
ce b
y
the blo
c
k
prio
rity gra
d
e
determinat
io
n, buildin
g a
r
tificial fish d
e
com
p
o
s
e fi
ne
image
re
storation mod
e
l to achieve re
storatio
n
of t
he detail
s
of
the image
a
nalysi
s
an
d the
positio
n of th
e pixel poi
nts. Finally, the
simulatio
n
re
sults
sh
ow th
e su
pe
rior pe
rforma
nce of
the
algorith
m
.
2. Conduc
tio
n
Model and
Priorit
y
Dete
rmination
2.1. Image Textur
e Infor
m
ation Tran
smission
Model Based on Intuitionistic Fuz
z
y
Sets
Before
we d
e
s
ign
an ima
g
e
re
storation
algorith
m
wit
h
a pri
o
ri u
n
known pixel, we give a
model
struct
ure
de
sign
a
bout the
ima
ge textur
e
inf
o
rmatio
n con
ductio
n
.
We prop
ose a
m
a
g
e
texture info
rmation tran
sfer m
odel
in
this
pap
er [9], and
cho
o
se
a
blo
c
k to
repair fro
m
m
any
blocks aroun
d the edg
e pi
xels and
dete
r
mine it
s pri
o
rity. Let
denot
e the imag
e texture
informatio
n p
e
r u
n
it time, a
nd a
s
sume t
h
e imag
e textu
r
e
sub
s
p
a
ce
of fuzzy set
s
as
a cond
ucti
on
function:
(1)
Among the
m
,
is th
e h
eat
flux of fuzzy
set p
e
r
unit time in im
age
texture, sh
owing
poor vi
sual i
m
age e
dge i
n
formatio
n in
dicate
s c
ond
uctivity. Assu
ming the g
r
a
d
ient directio
n
along the e
d
g
e
of the image informatio
n is:
,
;
,
;
/
(2)
Utilizatio
n th
e obje
c
tive functio
n
to ze
ro unifo
rmly
ergo
dic
prop
erties, a
nd g
e
t image
texture inform
ation flow de
nsity vector i
s
:
(3)
Among them,
,
are unit direction vecto
r
. Based on intuitionisti
c
fuzzy sets struct
ure of
image texture information
transmi
ssio
n
model, we
can obtain the center
of the partial
derivative through th
e obj
ective fun
c
tio
n
of zero
eve
n
traversal f
e
ature
s
a
nd lo
gical
differen
c
e
variable
scal
e feature
s
. With the prop
a
gation di
re
ct
ion of the ho
ri
zontal a
nd ve
rtical
con
d
u
c
tive
sub
-
regio
nal division, we can
fin
d
an
i
m
age with
th
e cu
rrent hig
hest p
r
io
rity to be repai
re
d
optimal samp
le blo
ck inta
ct
from a la
rge
sampl
e
a
r
ea,
get image tex
t
ure info
rmati
on cond
uctio
n
model
struct
ure
de
sig
n
. In the
current
blo
c
k h
a
s b
een
re
paired
by
see
k
in
g
obje
c
tive fun
c
tion
zero unifo
rm
traversal feature
s
and
differential
lo
gic varia
b
le
scale feature
s
to get partial
derivatives center
and
make
de
riva
tive zero.
T
he state
eq
uation of th
e intro
d
u
c
tio
n
of
intuitionis
t
ic
fuzz
y s
e
ts
is
de
sc
r
i
be
d
as
fo
llo
ws
:
,;
Gx
y
t
0
()
(
,
)
(,
)
l
i
m
[
]
x
uu
u
u
x
t
px
t
x
x
(,
;
)
(,
;
)
(,
;
)
[
(
,;
)
(
,;
)
]
xy
p
xy
t
u
xy
t
G
xy
t
Gx
y
t
i
G
x
y
t
j
Evaluation Warning : The document was created with Spire.PDF for Python.
TELKOM
NIKA
ISSN:
1693-6
930
Im
age Resto
r
ation Algorith
m
Based on Artificial
Fish
Swarm
Micro De
com
positio
n… (Dan Sui)
189
(4)
In the above
formula,
repre
s
ent
s the
confide
n
ce level of the edge pixel
s
, and
rep
r
e
s
ent
s co
nfiden
ce leve
l of an u
n
kno
w
n a
pri
o
ri th
e pixel,
is the inform
ation transmission
to be repai
re
d multidimen
sion
al spe
c
tral peak,
is th
e noise
comp
onent. Throu
gh the buildin
g
of the introdu
ction
of imag
e texture i
n
fo
rmation
tran
smissi
on
mod
e
l of intuitioni
stic fu
zzy sets,
multidimen
sio
nal sp
ectru
m
peak
se
arch metho
d
is use
d
to con
s
tru
c
t
the inform
ation
cha
r
a
c
teri
stic of image texture st
ru
cture
spa
c
e
by the
minimax eig
envalue
shu
n
t
into the image
noise su
bspa
ce an
d sig
nal
sub
s
pa
ce [1
0], thus laying a solid fou
n
dation for rep
a
iring.
2.2. Priorit
y
Determination on Unrepaired Block
In the b
u
ildin
g of the
stru
cture
of imag
e
te
xture i
n
formation tran
smissi
on
mod
e
l, it nee
d
to
r
e
pa
ir
q
u
i
ck
ly to
d
e
t
er
min
e
pr
io
r
i
ties
.
Ba
s
e
d
on th
e
literature [5], this p
ape
r p
r
opo
se
s ne
utron
spatial
chara
c
teri
stic information m
u
ltid
imensi
onal
search
metho
d
to d
e
si
gn
sub
s
p
a
ce m
odel
stru
cture. Th
e model dia
g
ra
m is shown in Figure 1.
Figure 1. Det
e
rmin
ation m
odel ba
se
d o
n
mult
i-dime
n
s
ion
a
l su
bspa
ce re
pai
r blo
c
k se
arch
Figure 1
sho
w
s
den
otes the da
mage
d
area
(white a
r
ea
).
∅
re
presents inta
ct re
gion
(gray area).
rep
r
e
s
ent
s
an ed
ge li
ne
regi
on a
n
d
the da
mag
e
d
area
of the inta
ct.
rep
r
e
s
ent
s th
e pixel to
be
rep
a
ire
d
in
.
ex
presse
d i
n
point
s in
th
e center of t
he pixel
block to be repaired the collectio
n.
In the block p
r
io
rity determi
n
a
tion, firstly,
update the e
dge
pixels a
nd
use multi-dimen
s
ion
a
l sub
s
pa
ce fe
atur
e
inf
o
rmatio
n
sea
r
ch m
e
thod.
M
u
ltidimen
sion
al
sea
r
ch su
bsp
a
ce featu
r
e in
fo
rmation iterative equatio
n is:
(5)
(6)
In the above
formula,
repre
s
e
n
ts th
e iteration n
u
mbe
r
,
is the total
numbe
r
of iteration
s
,
is
the pixel val
ue,
is the
u
pdate
sp
eed,
it rep
r
e
s
ent
s the
comm
on cha
r
acteri
stic fe
ature
sub
s
pa
ce
informat
ion
and the texture structu
r
e o
f
the two part
s
.
The
size
of the imag
e to
be re
pai
red i
s
a
s
su
med
a
s
, the size
of block
is
,
and
achi
eve the p
r
iority to be re
paire
d pixel locati
o
n
determination by the
above iterative search.
2.3. Upda
te
of a Priori Unkno
w
n
Pixe
l Confiden
ce
In ord
e
r to
m
a
intain the
continuity of r
epairi
ng of th
e dam
age
d a
r
ea
of imag
e, it must
update p
o
int of repaire
d confiden
ce. Th
e update
d
gui
deline
s
is:
12
12
1
1
1
12
12
12
2
2
2
12
(,
)
(
1
)
0
(,
)
(
1
)
0
xx
fx
x
r
x
NN
xx
gx
x
r
x
NN
p
p
p
(1
)
(
)
(
)
1
(,
)
(
,
)
(,
)
nn
n
ux
y
u
x
y
u
x
y
(
)
()
()
1
(,
)
(
,
)
(,
;
)
nn
n
st
u
x
y
M
ux
y
N
ux
y
d
1
,
2
,
.
..,
nT
T
()
(,
)
n
ux
y
mn
p
s
s
Evaluation Warning : The document was created with Spire.PDF for Python.
ISSN: 16
93-6
930
TELKOM
NIKA
Vol. 14, No. 1, March 2
016 : 187 – 1
9
4
190
(7)
After
is
rep
a
i
r
ed,
confid
en
ce
of ori
g
inal
is
all up
date
d
confiden
ce
of
. The averag
e numb
e
r of t
he num
be
r of edge
pixels
i
n
ea
ch
calcul
ating is
.In the tradition
al
method, it use
s
AFSA efficient globa
l search
of the Update i
n
formatio
n to obtain fuzzy
membe
r
ship
and cl
uste
r centers
iterativ
e update exp
r
ession i
s
:
(8)
is the
divers
ity fac
t
or,
is
the fuz
z
y
membershi
p
matrix for t
he parameter
neigh
borhoo
d
information.
Becau
s
e th
e
maximum nu
mber
of
y
is
n
o
more than
. The hig
h
e
s
t
comp
utationa
l com
p
lexity of updat
ed p
r
iori u
n
kno
w
n
pixel co
nfide
n
ce
after
rep
a
iring
t is
.
Thro
ugh
up
d
a
ting the
p
r
io
ri u
n
kno
w
n
pi
xel
info
rmati
on, the
un
kn
own
pixel
s
b
e
com
e
kn
own
informatio
n, so as to provide a prio
ri information for im
age re
sto
r
atio
n
3. The
Introduc
tion o
f
Artificial
Fish
Imperc
eptible De
c
o
mposition Algorithm a
n
d
Impro
v
ed Image Re
sto
r
ation Algorith
m
The ima
ge te
xture structu
r
e inform
ation
con
d
u
c
tion
model a
nd th
e blo
ck to b
e
repai
red
and pri
o
rity determi
nation
algorithm that prop
ose
d
above ha
s failed to achi
eve template
matchin
g
and
joint d
e
com
positio
n, so it
is diffi
cult to
have
a
goo
d result of im
age
re
storation.
This pap
er u
s
e
s
a
r
tificial fish swarm micro
de
co
m
positio
n to p
i
xel feature
s
and
co
mbin
es
brightn
e
ss co
mpen
sation t
o
improve the
priori u
n
kno
w
n pixel re
pai
r perfo
rma
n
ce.
3.1. Artificial
Fish S
w
a
r
m
Micro Deco
mposition M
odel
In ord
e
r to
im
prove th
e a
ccura
cy of ima
g
e
re
sto
r
ation,
we
nee
d to
p
r
ocess th
e
structural
feature
s
of t
he ima
ge inf
o
rmatio
n so
that the
mi
cro features of
the imag
e h
a
ve an
effect
ive
pre
s
entatio
n and extra
c
tio
n
. This pap
e
r
use
s
arti
fici
al fish swarm global se
a
r
ch al
gorith
m
for
image featu
r
e
tiny, combini
ng the g
r
eat clusteri
ng feat
ure
s
of fuzzy
set to obtain t
he targ
et of the
evaluation fu
nction. The i
m
age g
r
ay value time se
rie
s
is:
(9)
Rep
r
e
s
ente
d
on the
gray p
i
xel image
fe
ature ve
cto
r
,
is th
e nu
mb
er of
pixel. We u
s
e
artificial fi
sh
micro featu
r
e
decomp
o
siti
on meth
od.
The a
r
ea
of
the se
arch center of
ima
ge
segm
entation
is:
(10)
Combi
ned F
u
zzy set theory and u
s
ed
fuzzy memb
ership fun
c
tio
n
that extended from
Atanassov, it gives i
n
tuitionisti
c
fuzzy
set
s
. In fuzzy
set
s
sca
l
e sp
ace, we use a
combi
nation
of coa
r
se an
d
fine se
arch
(sea
rch ste
p
i
s
1)
ba
sed
on
the be
st pha
se Q
u
ick
sea
r
ch
block mat
c
hi
ng algo
rithm.
Assu
ming th
e spe
ed of a
r
tificial fish m
i
gration i
n
the sea
r
ch ima
g
e
micro feature
point is:
(11)
In the formula,
and
represe
n
t the
ar
tificial fish in the search i
m
age mi
cro
feature poi
nt and artifici
al fish migration
ra
te in the moment
and
,
sho
w
s th
e pulse
freque
ncy of
artificial fish
swarm i
n
th
e se
arch im
age mi
cro fe
ature p
o
int, on be
half of
the
()
=
(
)
I
yC
p
p
y
p
()
Iy
y
0(
)
1
Cp
p
b
1/
1
1/
1
(1
(
1
)
)
(
)
(1
)
(
1
(
1
)
)
n
m
ik
k
k
k
i
n
m
ik
k
ux
x
v
u
ik
u
2
s
2
()
Ot
s
y
12
{,
,
,
}
n
X
xx
x
n
12
{,
,
,
}
c
vv
v
v
{}
ik
uu
1*
()
.
tt
t
id
id
id
d
i
vv
x
x
f
t
id
v
1
t
id
v
i
1
t
t
i
f
Evaluation Warning : The document was created with Spire.PDF for Python.
TELKOM
NIKA
ISSN:
1693-6
930
Im
age Resto
r
ation Algorith
m
Based on Artificial
Fish
Swarm
Micro De
com
positio
n… (Dan Sui)
191
artific
i
al fish
the
i
gro
up
artificial fish
swarm, it ge
ner
ally exp
r
e
s
sed Daigo
n
g
individual f
i
sh
swarm, which
the expressi
on
can b
e
expre
s
sed a
s
:
(12)
and
indicate
s artifici
al fish
pulse fr
eq
ue
ncy ra
nge,
expresse
s a
s
pixels i
n
the
a pri
o
ri un
kno
w
n
re
pai
r interval ran
g
e
unifo
rm
ly di
stribution. Ima
ge e
nha
ncem
ent ba
se
d o
n
a
two-di
men
s
io
nal plane
zeros di
screte
method, and
then feature
extracted to
be micro m
a
trix
sea
r
c
h
st
ep
i
s
a
, a
nd g
e
t a sa
mple
blo
ck
that is no
t nece
s
sa
rily the be
st mat
c
hin
g
blo
ck
of
, we u
s
e AFSA i
m
pleme
n
tatio
n
detail
s
d
e
com
p
o
s
ition
to obtain
a micro
decompo
sitio
n
results a
s
:
(13)
In the above formul
a, by AFSA, we can
obtai
n goo
d sample blo
c
k area to search, using
expre
s
s the
block
as
kn
o
w
n to
be
rep
a
ired. In
su
m
m
ary, we
obt
ain p
r
e
c
ise
fine
image de
co
m
positio
n mod
e
l and alg
o
rit
h
m desi
gn p
r
oce
s
s is sho
w
n in Figu
re
2.
Figure 2. Flow Ch
art
In summa
ry, the artificial
fish swarm
algor
ith
m
de
comp
ositio
n model p
r
ovid
e a prio
ri
cha
r
a
c
teri
stic texture information for the
ev
entual re
storation of the
image pixel lo
cation.
3.2. Impro
v
e
d
Image Res
t
ora
t
ion Alg
o
rithm
Based
on
the
image
texture information
co
ndu
ction
model
and
th
e artifici
al fish swa
r
m
algorith
m
de
comp
ositio
n
model, we
combi
ne u
n
i
t
e the edge
feature p
o
i
n
ts bri
ghtne
ss
comp
en
satio
n
algo
rithm to achieve a
prio
ri un
kno
w
n pixel imag
e
informatio
n repair. Alg
o
rit
h
m
key technol
o
g
ies
are de
scrib
ed
as foll
ows: In t
he a
r
tificial fish swarm alg
o
rith
m de
comp
osi
t
ion
model, we e
x
press the fish mig
r
ation
rate and th
e
n
have to take the spatia
l position of the
artific
i
al fis
h
trans
form definition is
:
(14)
i
f
mi
n
m
ax
m
i
n
()
.
i
f
ff
f
r
a
n
d
mi
n
f
max
f
r
and
p
'
p
p
'a
r
g
m
i
n
(
,
)
q
pp
q
d
(,
)
p
q
d
p
1
tt
t
id
id
id
vv
v
Evaluation Warning : The document was created with Spire.PDF for Python.
ISSN: 16
93-6
930
TELKOM
NIKA
Vol. 14, No. 1, March 2
016 : 187 – 1
9
4
192
We u
s
e a
r
tificial fish swa
r
m traversal
s
earch m
e
tho
d
s to find the
image featu
r
e micro
points in th
e p
r
o
c
e
s
s o
f
edge
featu
r
e
brig
htne
ss
wea
k
poin
t
s for b
r
ight
ness. After
the
comp
en
sated
image freq
ue
ncy ban
d, the enhan
cem
e
n
t
degree of ex
pre
ssi
on is:
(15)
It can
be
se
en from the
above
equ
ation that to
fin
d
a
be
st mat
c
hin
g
bl
ock
of a
prio
ri unkno
wn numbe
r of pixel repair
need to trav
erse
times. We trave
r
sed
throug
h the
micro
sea
r
ch
for the
ori
g
i
nal pixel
s
y u
n
kn
own info
rmation in
co
nfiden
ce afte
r the
resto
r
atio
n of an assignm
ent, and the
n
we got
th
e texture structural information maxi
mum
likeliho
od e
s
timation pa
ram
e
ters ite
r
ative
formula:
(16)
Evenly distrib
u
ted withi
n
a
singl
e artifici
al fish b
eari
n
g individu
al p
i
xels on th
e texture
spa
c
e
are
a
is rep
r
e
s
ente
d
by the attenu
ation char
act
e
risti
cs, i
s
a
consta
nt on th
e pap
er valu
e
of
0.37. Then,
we u
s
e the fi
nite differen
c
e met
hod
discreti
zation m
e
thod ba
se
d
on mathem
atical
morp
holo
g
y theory
of top
o
logi
cal the
o
r
y, a prio
ri u
n
kn
own pixel
image i
n
formation to re
pair
s
o
me of the
covarianc
e
matrix is
:
(17)
In the ab
ove
formula,
is th
e samplin
g v
a
rian
ce
of the
data m
a
trix,
is th
e nu
mbe
r
of sna
p
shots.
Assume
stati
s
tically in
dep
ende
nt
betwe
en succe
s
sive sn
ap
shot
s.
Individual m
a
ps
throug
h artifi
cial fish to e
a
ch
point in t
he imag
e mi
cro fe
ature p
o
ints of the
search
spa
c
e.
We
use th
e artifi
cial fish be
h
a
vior a
s
the
obje
c
tive fun
c
tion to
solv
e the p
r
oble
m
in imag
e
micro
feature
poi
nts se
arch
optim
ization
p
r
o
c
e
ss.
The
fi
nal
iteration
of th
e alg
o
rithm
a
nd the
cycle
can
be run im
ple
m
ented on G
P
U to achiev
e the purp
o
se
of image re
storation.
4. Simulatio
n
Experiment and Result
Anal
y
s
is
In order to
test and validate the method
ba
sed on artificial fish swarm micro
decompo
sitio
n
an
d ima
g
e
brightn
e
ss
co
mpen
sation
p
r
iori
un
kn
own
pixel
repai
r
p
e
rform
a
n
c
e,
we
con
d
u
c
t a
si
mulation
exp
e
rime
nt. Experime
n
tal sa
mples taken
from a l
a
rg
e
image
data
base
Crimi
n
isi
Co
w, Rabbit
an
d Wall fo
r th
e test
imag
e
sampl
e
s.
In t
he p
r
io
ri u
n
known im
age
pixel
image
re
stora
t
ion pe
rforma
nce te
sting, t
he dime
nsi
o
n
s
a
r
e: 9
×
9, 1
1
×1
1, in o
r
de
r to dem
on
strate
the algorith
m
and traditio
nal algo
rithm
in two
grou
ps of image
sampl
e
s to repair the visual
effects, sam
p
ling of the block template:
9×9, cu
rsory
search
step:
. We use this algorithm an
d
sub
-
spa
c
e
search
algo
rithm in [8]
with a pri
o
ri
unkno
wn pix
e
l impri
n
ting
to com
p
a
r
a
t
ive
s
i
mulation. Vis
ual tes
t
s
a
mple
s
are 3 to 5 respec
tively.\
(a) O
r
igin
al image
(b) O
b
je
ct re
mov
ed
image
(c) Ou
r meth
od
(d) met
hod in
[8]
Figure 3. Co
w
10
[
1
e
xp(
.
)
]
t
ii
rr
t
'
p
22
(/
9
)
On
m
a
s
1
.
tt
ii
A
A
22
0
,,
xs
s
n
RR
I
0
Evaluation Warning : The document was created with Spire.PDF for Python.
TELKOM
NIKA
ISSN:
1693-6
930
Im
age Resto
r
ation Algorith
m
Based on Artificial
Fish
Swarm
Micro De
com
positio
n… (Dan Sui)
193
(a) O
r
igin
al image
(b) O
b
je
ct re
mov
ed
image
(c
) Ou
r meth
od
(d)
sub
s
p
a
c
e
method
Figure 4. Rab
b
it
(a) O
r
igin
al image
(b) O
b
je
ct re
mov
ed
image
(c
) Ou
r meth
od
(d)
sub
s
p
a
c
e
method
Figure 5. Wal
l
Figure 3
-
5
sh
ows the
remo
ving and
re
pa
iring effe
ct
of three gro
u
p
s
test
sa
mple
s image,
the sim
u
latio
n
re
sult
s
we
re an
alyze
d
a
s
follo
ws: ma
rkin
g
circle
of
Figu
re
3 a
n
d
Figu
re
4
can
be
found se
riou
s
stru
ctu
r
al
ru
pture and discontin
uity
after
the re
pair usin
g
the sub
s
pa
ce
t
r
aditio
nal
resto
r
atio
n al
gorithm
chan
ge in
p
r
io
ri u
n
kn
own
pixel
area, and co
mbined results, we ca
n se
e in
Figure 5, usi
ng the algo
rithm of this pa
per have g
o
o
d
visual effects after the restoration of the
image. Th
e reason i
s
ofte
n that t
he
use of this algo
rithm
can
be
introdu
ce
d a
r
tificial fish
swarm
micro d
e
com
positio
n p
r
io
r kn
owl
edg
e,
updatin
g a
p
r
iori
un
kn
own pixel
co
nfiden
ce i
n
im
a
ge
repai
r, resto
r
ation p
r
iority
in favor
of a
stron
g
stru
ct
ural i
n
form
ation dete
r
min
e
d
to be
the n
e
xt
repai
r bl
ock
dire
ction, a
n
d
the
u
s
e of
traditional met
hod
s
cann
ot
effectively
cal
c
ulate un
kno
w
n
confid
en
ce in
formation pix
e
l. Therefore
Figure
3-
5 to be repai
re
d is p
r
eferabl
y visually rep
a
ir
image.
In orde
r to quantitative analysis the rep
a
ir quality an
d perfo
rman
ce based on a
sample
template
size
9×9,
coa
r
se
sea
r
ch
step i
s
u
s
ed to
cal
c
ulate th
e SNR comp
ari
s
o
n
metho
d
to test
the repai
r tim
e
an
d im
age
restoration
after
com
p
a
r
is
o
n
of th
e d
a
ta
obtaine
d te
st
results sho
w
n
in
Table 1. The
results fro
m
Table 1 are:
(1) Usin
g thi
s
algo
rithm
an
d conventio
n
a
l sub
s
pa
ce
repair alg
o
rith
m ne
ed to
be
re
paired
with the re
pai
r time image
size incre
a
se
s, and the
co
mplexity of the algo
rithm d
epen
ds o
n
th
e
image
si
ze i
n
crea
se
s by
contrast
calculation
algo
ri
thm complexi
ty. The re
sul
t
s of the
s
e
two
prop
ertie
s
is
con
s
i
s
tent wi
th the actual
sit
uation an
d theory to prove
the effectivene
ss of
the
algorith
m
.
(2)
Usi
ng this algorithm, th
e error of si
g
nal
to noise ratio of repai
red image
s a
r
e small
e
r,
and m
a
intain
at less tha
n
6%. It indicat
e
s that th
e p
r
opo
se
d al
go
rithm is mo
re
than
sub
s
p
a
c
e
algorith
m
s to
ensure the q
u
a
lit
y of the restore
d
imag
e.
(3) It can be
see
n
from the time of tho
s
e tw
o Repai
r algorith
m
s, the large
r
the
size of
the repai
r im
age, the gre
a
t
er time ratio R of QSOM
B algorithm to repai
r, but as the size of the
resto
r
atio
n of image be
co
mes la
rge
r
a
nd its val
ue tend to 42. Selecte
d
wh
en
the R value is 3
alway
s
le
ss than the
R-value
4. From an
alysi
s
, the alg
o
rith
m in thi
s
p
aper ha
s b
e
tter
conve
r
ge
nce and sta
b
ility.
In summa
ry, our al
gorith
m
can e
n
sure t
he vi
sual of
repaired ima
g
e
effectively than the
traditional
su
bsp
a
ce al
gori
t
hms, a
nd it
costs sma
lle
r t
i
me to
rep
a
ir
and i
m
prove t
he
stability a
nd
Evaluation Warning : The document was created with Spire.PDF for Python.
ISSN: 16
93-6
930
TELKOM
NIKA
Vol. 14, No. 1, March 2
016 : 187 – 1
9
4
194
conve
r
ge
nce
to achieve
holog
rap
h
ic.
Therefore,
th
e artificial fish image rest
oration al
go
rithm
prop
osed in this pa
per i
s
a
more excelle
nt
performan
ce of image
restoration alg
o
rithm
s
.
Table 1. The
Perform
a
n
c
e
Image Dataset
Our m
e
thod
Subspace
R=T2/ T1
(U-V)
/SNR
(%)
Computing
time:
T
1
(S)
Repaired image
SNR:U(dB
)
Computing
timeT2(S)
Repaired image
SNR
：
V(
d
B
)
Co
w
(512×384
)
208.66
22.512
1697.233
18.789
9.01
↑
2.05
Rabbit
(402×336
)
20.595
33.561
174.388
33.187
8.47
↑
1.12
Wall
(262×350
)
12.28
30.407
85.967
30.489
7
↓
0.30
5. Conclusio
n
We
pre
s
e
n
t a
n
alg
o
rithm
with prio
ri
un
kn
own
pixel tha
t
can
be
re
ali
z
ed
on
the
re
covery
of missi
ng inf
o
rmatio
n, whi
c
h have im
po
rtant appli
c
ati
ons in th
e field of image ta
rget re
co
gniti
on
and rem
o
te sensi
ng dete
c
tion.
Thi
s
pap
er pro
p
o
s
e
s
a meth
od
ba
sed o
n
a
r
tificia
l
fish
re
sto
r
ation
micro de
com
positio
n with
prio
ri un
kn
o
w
n h
o
log
r
ap
hic pixel
s
. B
e
sid
e
s, b
u
ildi
ng st
ru
cture
of
image texture
information t
r
an
smi
ssi
on
model to r
e
p
a
ir pie
c
e of p
r
iority de
cisi
o
n
. Troug
h up
date
of a p
r
iori
un
known pixel
co
nfiden
ce, we
introdu
ce
micro im
age
de
compo
s
ition m
odel of
artifici
al
fish alg
o
rithm
and
com
b
ine
with the
edg
e feature poi
nt’s b
r
ightn
e
ss comp
en
sati
on alg
o
rithm
so
as to
a
c
hieve
a p
r
io
ri u
n
kn
own
pixel i
m
a
ge info
rm
atio
n repai
red. St
udie
s
sho
w
th
at the
algo
rith
m
in this pa
per
has a
goo
d visual effe
ct, saves r
epai
rin
g
time effecti
v
ely and improves the
stab
ility
and converg
ence. Our m
e
thod will a
pply in t
he image resto
r
ation algo
rith
m, fuzzy obj
ect
recognitio
n
a
nd feature ext
r
actio
n
and ot
her field
s
with
high value.
Referen
ces
[1]
T
ao
T
,
Xia
XY, Li L. Platform deali
ng
w
i
t
h
ani
m
a
ted ch
aracters bas
ed
on
the calcu
l
ation of dat
a
ble
ndi
ng.
Jour
nal of Hefe
i Un
iversity of T
e
ch
nol
ogy (Natur
al
Science)
. 20
1
4
; 31(1): 59-6
2
.
[2]
Z
hou Yat
ong,
Li L
i
n,
Xi
a Ke
w
e
n. R
e
se
arch
on
w
e
i
ghte
d
p
r
iorit
y
of e
x
em
plar-b
ase
d
im
a
g
e
i
n
p
a
inti
ng.
Journ
a
l of Elec
tronics
. 201
2; 29(1): 16
6-1
7
0
.
[3]
Xu
Z
ong
Ben,
Sun Ji
an. Imag
e in
pai
ntin
g
b
y
patch pr
op
ag
ation
usin
g p
a
tch
sparsit
y
.
IEEE
T
r
ansactio
n
on Image Proc
essin
g
. 201
0; 19(5): 115
3-1
1
6
5
.
[4]
T
a
z-Ho K
w
ok,
Hoi Sh
eu
ng, C
harli
e C
L
W
a
n
g
. F
a
st quer
y f
o
r e
x
em
plar-
b
a
s
ed ima
ge c
o
m
p
leti
on.
IEEE
T
r
ansactio
n
s o
n
Ima
ge Proce
ssing
. 20
10; 19
(12): 310
6-3
1
1
5
.
[5]
Drori I, Co
hen-
Or D, Yeshuru
n
H. F
r
agme
n
t-base
d
ima
ge
compl
e
tion.
AC
M T
r
ansactio
n
s
on Grap
hi
c
s
(TOG).
2003; 2
2
(3): 303-
31
2.
[6]
Borikar S, Bisw
a
s
K K, Pattana
ik S. F
a
st
alg
o
ri
thm for compl
e
tion of i
m
ages
w
i
th n
a
t
ural scen
e
s. In
CS T
e
chnical
Rep
o
rt. Univer
sit
y
of Ce
ntral
Florida. 2
0
0
4
.
[7]
Barlia
n He
nr
yr
anu Pras
etio,
Eko Setia
w
a
n
, Adhar
ul M
u
ttaqin.
Imag
e
Encr
yption
u
s
ing Sim
p
le
Algorit
hm on F
P
GA.
T
E
LKOMNIKA (T
elecommunic
a
tio
n
Co
mp
uting El
e
c
tronics an
d C
ontrol)
. 20
15;
13(4).
[8]
Crimin
isi A, P
é
rez P, T
o
ya
ma K. Re
gi
on
f
illi
ng
an
d o
b
ject r
e
mova
l
b
y
e
x
emp
l
ar-
base
d
im
ag
e
inp
a
inti
ng.
IEEE Transactions
on image proc
essing
. 200
4; 13(9): 120
0-1
2
1
2
.
[9]
Efros AA, F
r
eeman W
T
.
Image
qu
iltin
g fo
r texture synth
esis a
nd tra
n
s
f
er
.
In Procee
din
g
s of 2
8
t
h
Annu
al Co
nfer
ence o
n
Com
p
uter Graphics
and Inte
ractiv
e
T
e
chniq
ues. U
SA, ACM. 2001: 341-3
46.
[10]
Norh
adi
ja D
a
r
w
i
n
, An
uar Ah
mad. F
a
st Dat
a
Ac
qu
isitio
n
of Aeria
l
Imag
es Usin
g U
n
m
ann
ed A
e
ria
l
Vehic
l
e S
y
ste
m
.
Internation
a
l
Journ
a
l of Informatics an
d C
o
mmunic
a
tio
n
T
e
chno
logy
. 2
014; 3(3).
[11]
Z
hang F
F
.
Res
earch
of Image
Mosaic
Al
gor
ithm Base
d on
Neig
hb
orh
ood
Gradie
n
t.
Bulle
tin of Scie
n
c
e
and T
e
ch
no
log
y
. 2012; 10(
28)
: 61-63
．
[12]
Li Z
D, H
e
H J,
Yin Z
K. Ad
apt
ive Imag
e Inp
a
i
ntin
g Al
gorith
m
Based
on
P
a
tch Structure
Sparsit
y
.
Acta
Electron
ica Sin
i
ca
. 201
3; 41(3
)
: 549-55
4.
Evaluation Warning : The document was created with Spire.PDF for Python.