TELKOM
NIKA
, Vol. 11, No. 6, June 20
13, pp. 3463
~ 347
2
e-ISSN: 2087
-278X
3463
Re
cei
v
ed
Jan
uary 28, 201
3
;
Revi
sed Ap
ril 23, 2013; Accepted Ap
ril 30, 2013
Mathematical A
n
alysis and Application on Mechanical
Image o
f
Hybrid Wavelet Transform
Algorithm
Fuze
ng Yan
g
*1
, Qiong Liu
1,2
, Meng
y
u
n Zhang
1
, Yuanjie Wang
1
,
Yingjun Pu
1
1
Colle
ge of Me
chan
ical a
nd El
ectronic En
gin
eeri
ng, No
rth
w
est A&F Univer
sit
y
, Yan
g
li
ng, P.R. China;
2
Colle
ge of Info
rmation En
gin
e
e
rin
g
, Yang
lin
g
Vocation
al
& T
e
chn
i
cal C
o
ll
eg
e, Yangl
in
g, P.R. Chin
a.
*Corres
p
o
ndi
n
g
author, e-ma
i
l
:
y
f
z07
01@
16
3.com
A
b
st
r
a
ct
T
o
overc
o
me t
he s
hortco
m
i
n
gs suc
h
as
sig
n
ifica
n
tly d
e
-no
i
sing
effect a
n
d
easi
l
y l
o
si
ng t
he
detai
l
s
of the
i
m
a
g
e
character
i
stics
of th
e ex
istin
g
i
m
age
d
e
-n
o
i
sing
meth
ods,
an
i
m
age
d
e
-
noisi
ng
a
l
gor
ithm
base
d
o
n
the
h
y
brid w
a
ve
let transfor
m
w
a
s p
r
opos
ed.
T
he
a
l
gorit
hm inte
gr
ated the
adv
an
tages of w
a
ve
l
e
t
de-n
o
isi
ng ret
a
inin
g i
m
age
de
tails featur
es a
nd W
i
e
ner
filte
r
obtai
nin
g
the
opti
m
al
s
o
l
u
ti
on, an
d took t
h
e
imag
es proc
es
sed by w
a
ve
let
transform
an
d
W
i
ener filter
a
s
ma
le a
nd fe
ma
le of th
e in
itial p
o
p
u
lati
on.
T
he
steps of the al
gorith
m
ar
e as
follow
s
: map
p
i
ng fro
m
i
m
a
g
e
space to co
din
g
spac
e, iterating to par
en
ts
throug
h selecti
on, crossover
and mutatio
n
operati
on
unt
il the offsprin
g me
etin
g the
constraints w
a
s
obtai
ne
d, red
u
c
ing th
e su
pe
rior offspri
ng t
o
i
m
a
ge s
pac
e, gai
ni
ng th
e
appr
oxi
m
at
e
opti
m
a
l
sol
u
tio
n
.
T
heoretic
al a
n
a
lyses w
e
re
ma
de
on th
e
core of t
he al
gorith
m
,
co
din
g
,
crossover a
nd mutati
on. T
h
e
alg
o
rith
m w
a
s app
lie
d to agri
c
ultura
l mac
h
i
nery parts i
m
a
ge de-
nois
i
n
g
such as pl
oug
h and d
i
sk har
row
.
T
he results sh
ow
ed that it h
ad the
a
d
va
ntages
of hig
h
p
eak sig
n
a
l
to n
o
ise rati
o (PS
NR), obvi
ous
edg
e
character
i
stics, goo
d visi
on
e
ffect, and so o
n
.
T
he resu
lt o
f
the pres
ent
w
o
rk
imp
l
i
ed t
hat the
prop
os
ed
alg
o
rith
m is an
effective an
d feasibl
e
exp
l
or
ati
on.
Ke
y
w
ords
: hy
brid w
a
vel
e
t transform, i
m
a
ge
de-n
o
isi
ng,
ma
thematica
l
an
al
ysis, mach
in
er
y ima
g
e
Copy
right
©
2013 Un
ive
r
sita
s Ah
mad
Dah
l
an
. All rig
h
t
s r
ese
rved
.
1. Introduc
tion
With the po
p
u
lari
zation
of digital scan
ne
rs
a
nd
came
ras, imag
es
are be
comin
g
the mo
st
comm
only used inform
atio
n carrie
r in h
u
man life an
d the main way to get information fro
m
the
outsid
e
worl
d
.
But in the
proce
s
se
s of i
m
age
acqui
si
tion, tran
smi
s
sion,
scan
nin
g
, sto
r
age,
et
c, it
is often di
stu
r
bed
by varie
t
ies of noi
se
s, l
eading to
a
decre
ase
in
image q
uality and affe
ctin
g
visual effect
s of subsequ
e
n
t image pro
c
e
ssi
ng. It
is necessa
ry for image de
-no
i
sing in o
r
de
r to
obtain
high
q
uality digital i
m
age,
red
u
ci
ng the
noi
se
disturban
ce
i
n
ima
g
e
s
a
n
d
better reflect
i
ng
the ori
g
inal i
n
formation
ca
rried
by the i
m
age.
Ho
w to eliminate
th
e noi
se
rea
s
onably h
a
s
b
een
one
of the
m
a
in
subj
ect
s
of the
re
sea
r
ch fiel
d in
im
age
pro
c
e
s
si
ng, so that
th
e ima
ges n
o
t only
maintain th
e
origin
al inte
grity of the informati
on b
u
t al
so
rem
o
ve u
s
eless info
rma
t
ion, and
ada
pt
to the human
observation.
Image d
e
-n
oi
sing
ha
s b
e
e
n
a b
a
si
c a
n
d
impo
rtant concern
over
i
m
age pro
c
e
s
sing and
detectio
n
of
defect
s
, also
pre
r
eq
uisite
con
d
ition
s
i
n
the cou
r
se
of image
a
nalysi
s
, featu
r
e
extraction
an
d pattern
re
cognition [1, 2
]. For a long
time, people
pre
s
ent a
nd
develop diffe
ren
t
de-n
o
isi
ng al
gorithm
s a
ccordin
g to the
image
and
n
o
ise
statisti
ca
l cha
r
a
c
teri
sti
cs
and
spect
r
al
distrib
u
tion. The cla
s
sic i
m
age de
-noi
sing al
gorit
h
m
s are: Wie
ner filterin
g, median filteri
n
g
algorith
m
, ne
ighbo
rho
od
a
v
erage
alg
o
ri
thm, and
so
on. Sin
c
e th
e nin
e
teen
ei
ghties,
wavel
e
t
transfo
rm
s h
a
ve been
su
ccessfully used in im
age
de-n
o
isi
ng, a
nd its de-noi
sing effe
cts
are
better than t
he tradition
al
method, an
d that ca
u
s
e
s
a wid
e
attention of scholars [3]. Man
y
method
s are
to achieve th
e purpo
se of
de-n
o
is
i
ng b
a
se
d on the
optimal thre
shold of wavel
e
t
coeffici
ents
o
f
image filteri
ng. For exam
ple,
in 20
00
Wal
k
er an
d
Che
n
p
r
opo
sed ad
aptive t
r
ee
wavelet
shri
n
k
ag
e metho
d
[4], they transformed th
e
sign
al with
noise for o
r
th
ogon
al wavel
e
t,
then got the de-n
o
ised si
g
nal by
the co
efficient thre
shold ope
ratio
n
, the de-noi
sing effe
cts
were
quite goo
d, in 2000
Jalo
b
eanu, Th
oma
s
and
Rod
r
ig
uez et al imp
r
oved the th
resh
old meth
od
and p
r
e
s
ent
ed the tran
slation invari
a
n
t wavelet d
e
-noi
sin
g
me
thod [5, 6], Che
n
an
d Bui
prop
osed
to
multiple
wa
velet thre
sh
o
l
d de
-n
oisin
g
of ima
g
e
noise by
m
e
rgin
g
adja
c
e
n
t
coeffici
ents
a
ppro
a
ch, the
effects
we
re
better tha
n
singl
e wavelet, supe
rio
r
to the tra
d
itional
Evaluation Warning : The document was created with Spire.PDF for Python.
e-ISSN: 2
087-278X
TELKOM
NIKA
Vol. 11, No. 6, June 20
13 : 3463 – 3
472
3464
method
s [7],
in 2002, Katkovnik, Kgiaza
rian an
d Asto
la descri
bed
a novel app
roach ba
sed
on
the interse
c
ti
on of
co
nfide
n
ce
interval
s
(ICI)
rule
to solve
a pro
b
le
m
of windo
w size
(ban
dwi
d
th)
sele
ction
for filtering
an i
m
age
sig
nal
given
wi
th
a noi
se, th
e
ada
ptive tra
n
sforms with
the
adju
s
ted thre
shol
d pa
rame
ter perfo
rm b
e
tter than
the
adaptive wa
velet estimat
o
rs [8], Vidya
e
t
al used a
Wa
velet Tran
sform (WT) m
e
th
od with
so
ft thre
shol
d for
signal d
e
-n
oisi
ng [9], Hua
n
g
et
al propo
se
d
Self-ad
aptive Decompo
s
ition Lev
el De-noi
sing
Method
B
a
sed
o
n
Wavelet
Tran
sfo
r
m [10]. Some researche
r
s had
propo
se
d d
e
-noi
sin
g
me
thods that combine
wave
let
transfo
rm an
d other filterin
g method
s [11, 12].
But these me
thods a
r
e mo
stly to an image us
in
g a method of imag
e pro
c
e
ssi
ng [13-16],
there a
r
e fe
w com
b
ining
m
u
ltiple metho
d
s a
nd ma
ki
n
g
full use of their
re
spe
c
tive advanta
g
e
s
of
de-n
o
isi
ng. In last fe
w years, ou
r work i
s
b
een
sup
porte
d
by National
Natural Scie
nce
Found
ation o
f
China with
referen
c
e 30
9716
90. We
attempt to make a trai
n o
f
thought, with
referen
c
e to
t
he tho
ught
of
biolo
g
ical hy
bridi
z
ation
b
r
eedin
g
a
nd t
he g
eneti
c
al
gorithm
ide
a
s. In
this pa
per,
we cross
wav
e
let tran
sform and
Wien
er
filterin
g re
spe
c
tively as male an
d fe
male
hybrid with di
fferent advan
tages of de
-n
oisin
g
, extract their de-n
o
ising domin
ant gene
s, achie
v
e
better imag
e
de-n
o
isi
ng eff
e
ct by usi
ng
prop
osed
hybrid wavelet trans
form
algorithm[13]. But
we
use
no
math
ematical
the
o
ry an
alysi
s
at all to p
r
ov
e this
algo
rit
h
m, only a
p
p
l
y to agri
c
ult
u
re
image d
e
-n
oi
sing. So it is
very
necessa
ry for mathe
m
atical a
naly
s
is of
codi
ng compl
e
tene
ss
an
d
algorith
m
co
nverge
nce, a
nd the
appli
c
ation
s
ran
g
e
are exp
a
n
ded from a
g
r
icultu
ral
pro
duct
image to
ag
ri
cultural m
a
ch
inery im
age
i
n
orde
r to
ob
tain a
better i
m
age.
We
d
e
scrib
e
a
no
ve
l
approa
ch to the pro
b
lem o
f
mechani
cal
image de-
no
ising, and it is prove
d
to be a benefi
c
i
a
l
exploratio
n.
2. Rese
arch
Metho
d
Hybrid
wavel
e
t tran
sform
algorith
m
b
e
g
i
ns
with t
w
o
p
a
rent
s
(al
s
o
called th
e p
o
p
u
lation
s)
of re
spe
c
tive
advantag
e ge
nes, eve
r
y po
pulation i
s
m
ade u
p
of a
certain
numb
e
r
of individ
ual
s
obtaine
d by gene code. Th
erefo
r
e, the coding ta
sk fro
m
phenoty
pe to
genotype mappin
g
nee
ds
to be accom
p
lishe
d at the outset. After the gen
erat
io
n of the initial populatio
n, hybridization can
be carried
o
u
t. In the proce
s
s of hy
bridi
z
ati
on,
consi
deri
ng th
e two
pa
rent
al advanta
g
e
s
,
according
to
the
prin
cipl
e
of
survival
of t
he fittest,
better an
d
better
ne
w i
ndividual
s
wil
l
be
prod
uced fro
m
generation
to generatio
n. In every
generatio
n, indi
viduals a
r
e selecte
d
acco
rding
to individual fi
tness value i
n
the p
r
oble
m
domai
n, then cro
s
sover
and mutatio
n
are m
ade
am
ong
popul
ations,
new p
opul
ation is g
ene
rat
ed at last. Th
e last po
pula
t
ion is con
s
id
ered
as
opti
m
al
and the ap
proximate optimal soluti
o
n
after de
codin
g
operation.
On the view of macro
scopic, algo
rith
m
requi
re
s the followin
g
step
s: mappi
ng from
image
spa
c
e
to codi
ng
spa
c
e a
nd redu
ct
ion from
co
ding spa
c
e to i
m
age
spa
c
e,
this is
so
-call
ed
codi
ng an
d decodin
g
. In image spa
c
e, two paren
tal image
s a
r
e mainly ob
tained. Whil
e
the
spe
c
ific sele
ction,
crosso
ver a
nd m
u
tation o
peration a
r
e
com
p
leted i
n
co
ding
sp
ace,
the
operation is
iterated to parent
s throu
gh fitness
v
a
lue
s
usin
g the sele
ction
,
crossove
r and
mutation op
e
r
ator
acco
rdi
ng to respective advant
ag
es of the t
w
o parents,
u
n
til the offsp
r
ing
meeting the
constraints i
s
obtai
ne
d. In the end, redu
cing the
sup
e
r
ior offspri
ng
to image spa
c
e,
filial generation image
s ha
ving the domi
nant
gen
e of two pa
rent
s are gaine
d.
The im
pleme
n
tation of
hybrid
wavel
e
t
transfo
rm
alg
o
rithm i
s
: firstly enco
d
ing
the two
image
s n
eedi
ng hyb
r
idi
z
ati
on o
peration
into two
pop
u
l
ations re
co
rd
ed a
s
P1
(0
)
and P2
(0
), th
e
sup
e
rscript 0
is the initial populatio
n, and so
on. Ne
w pop
ulati
on P1(0*
)
is obtained aft
e
r
hybridi
z
ation,
then mutatio
n
is ope
rate
d
on P1(0*
)
in
orde
r to get
more comp
a
t
ible popul
atio
n
P1(0**
); at la
st, P1(0
**) i
s
con
s
id
ere
d
as in
itial
po
pulation
of n
e
xt gene
ratio
n
P1(1) until
the
optimal po
pul
ation P1(n) i
s
obtaine
d. Se
e detail
s
in
lit
eratu
r
e [13].
The
core of t
he alg
o
rithm
are
codi
ng, cro
ssover a
nd m
u
tation, the
key
are th
e com
p
letene
ss
of codi
ng
a
nd
t
he
conve
r
ge
n
c
e
of cro
s
sover
and mutation.
2.1. Coding
The algo
rith
m sele
cts float encodi
ng
method, ca
l
c
ulates a
c
cord
ing to gray scale ima
g
e
pixel value.
Con
s
id
erin
g
a gray image
of 256
pixel
s
x 25
6 pixel
s
, the
corre
s
pondi
ng n
u
m
e
rical
matrix a
r
e
d
enoted
a
s
(
256
,
2
5
6
)
M
. The el
eme
n
ts
in
(
256
,
2
5
6
)
M
are
di
spl
a
ce
d by
ro
w,
divides into
section
s
conta
i
ning a fixed numbe
r
(su
c
h as 20
), ea
ch seg
m
ent wi
ll be a new
row
Evaluation Warning : The document was created with Spire.PDF for Python.
TELKOM
NIKA
e-ISSN:
2087
-278X
Mathem
atical
Analysi
s an
d
Application o
n
Mech
ani
cal
Im
age of Hybrid…
(Fu
z
en
g Yang)
3465
vec
t
or
(
1
,
20)
i
r
, here
[
256
256
20
1
]
3277
i
, then these
vectors
(
1
,
20)
i
r
will form a
new mat
r
ix
(
3277,
20)
P
in colum
n
s.
1
3277
(1
,
2
0
)
(
327
7,
2
0
)
.
.
.
(
1
,
20)
r
P
r
(1
)
That is to
say, a matrix of pixels
(256
,
2
5
6
)
M
is conve
r
ted i
n
to a ne
w
matrix
(
3277,
20)
P
formed by row vecto
r
s
(
1
,
20)
i
r
in colu
mn
s,
1
,
2
,
...
,
3
2
7
7
i
. At this
time, the
origin
al imag
e is en
co
de
d into a correspon
ding
matrix co
ntai
ning 32
77 i
ndividual
s. It is
descri
bed i
n
geneti
c
algo
rithms that th
e init
ial pop
u
l
ation ca
n be
thought of
a colle
ction
P
contai
ning
32
77 in
dividual
s, the
po
pula
t
ion si
ze
is 3
277, a
nd t
h
e
individu
al
(o
r
chromo
som
e
)
length in the
popul
ation is
20.
2.2. Selection
Selection i
s
u
s
ed to dete
r
mine the indi
viduals n
eedi
ng crossove
r
or mutation b
a
se
d on
ran
k
ing
ran
k
i
ng-sele
ction[
17]. Firstly, sort the
individual
s acco
rdi
ng to
the fitness; se
co
nd
ly,
determi
ne a thre
shol
d valu
e by sele
ction
ra
te. The s
p
ec
ific
method is
as
follows
:
(1)
Cal
c
ulate
all the individual fitness value
()
Vi
()
()
1
()
|
(
)
|
n
ii
k
Vi
r
m
r
(2
)
(2) Sort
the i
ndividual
fitness valu
e fro
m
sm
all to
b
i
g (or from
b
i
g to
small
)
,
the results
a
r
e
recorded a
s
'
V
.
(3)
Determi
n
e threshold value ts acco
rding to the
selectio
n ra, the fitness valu
e greate
r
tha
n
ts
will be required to the following
operation. Threshol
d cal
c
ulation methods are as follows:
'
(
),
(
*
(
1
))
ts
V
k
k
j
ra
(3
)
Here: k is thresh
old value
corre
s
p
ondin
g
to the chro
moso
me lo
ca
tion;
'(
)
Vk
is the staini
ng
fitness value
corre
s
p
ondin
g
to the locati
on k.
If
'(
)
Vk
t
s
, it will need cro
s
sover a
n
d
mutation op
eration fo
r the sele
cted in
dividual
s.
2.3. Cros
sov
e
r
The p
u
rp
ose of cro
s
so
ver op
eratio
n is to
ge
nerate
ne
w individual
s by gen
e
recombi
natio
n. We firstly identify locu
s position
of inferio
r
gen
e
in every indi
vidual, and t
hen
hybridi
z
e the
gene
dire
ctly in ord
e
r to
qu
ickly o
b
tain e
x
cellent n
e
w i
ndividual
s. T
he ge
ne valu
e
s
are different from so-calle
d
bad gen
es to
other gen
es i
n
the same
chrom
o
some.
(1) Calcul
ate
ab
solute
de
viation by ea
ch l
o
cu
s
gen
e value
s
rela
tively to the mean val
u
e
s
of
gene o
n
chromosome, th
at is cal
c
ul
ated loci
m
gen
etic fitness
()
()
i
um
.
The meth
od
is a
s
follows
:
()
()
()
()
|
(
)
|
ii
i
um
r
m
r
(4)
(2) So
rt
()
()
i
um
from s
m
all to big, the result is
rec
o
rded as
()
'
i
u
.
(3)
Determine
locu
s positio
n of needing t
o
cro
s
sove
r
operatio
n, it is the position o
f
inferior gen
e.
Fix a thresho
l
d value
s
accordin
g to the hybrid rate
hra
. The fitness val
ue is gre
a
ter
than the
value of the gene was thou
ght to be inferior g
ene.
Th
e method to determi
ne th
e threshold i
s
as
follows
:
Evaluation Warning : The document was created with Spire.PDF for Python.
e-ISSN: 2
087-278X
TELKOM
NIKA
Vol. 11, No. 6, June 20
13 : 3463 – 3
472
3466
'
()
()
,
(
*
(
1
)
)
i
su
k
k
n
h
r
a
(5)
Her
e
,
s
is fixed thre
shol
d value; k i
s
thresh
old value
corre
s
po
ndi
ng to the lo
cation of gen
e
;
()
'
i
u
(k) is the po
sition k corre
s
po
ndin
g
to the
ch
romo
so
me fitness v
a
lue; n is ge
ne numb
e
r
contai
ning in
chromo
som
e
;
hra
is
hybrid rat
e
.
(4) Hyb
r
id the
gene
who
s
e
fitness value i
s
g
r
eate
r
tha
n
the th
re
shol
d value
s. Th
e metho
d
i
s
a
s
follows
:
()
(
)
()
()
ii
rm
R
m
(6)
Her
e
,
()
()
i
rm
is g
e
n
e
m on
ch
romosome i
from Parent
One P1;
()
()
i
Rm
is gene m
on
chromo
som
e
i from Parent
Two P2.
2.4. Muta
tion
The offsprin
g
is mutate
d af
ter hybri
d
op
eration; thi
s
i
s
ge
ne
s on
a
ch
romo
so
me
cha
nge
with a very
small probabili
ty. It is a local ran
dom
se
arch which m
a
ke
s the
algo
rithm itself al
so
has the capability of local
rand
om search. At the same time,
new individuals are generat
ed
throug
h varia
t
ion. It is the basi
s
of gua
ranteein
g
the diversity of p
opulatio
n, an
d it can redu
ce
prem
ature
co
nverge
nce of algorith
m
.
In order to reduce the vari
ability of
random, we use the followi
ng
methods:
(1)
Cal
c
ulate
absol
ute value b by gene value
s
on chromo
so
me chai
n rel
a
tively to mea
n
deviation, tha
t
is calculatio
n of gene fitness value,
()
()
()
|
(
)
|
ii
i
bm
r
m
r
(7)
Her
e
,
()
i
bm
is gen
e fitness valu
e of calculate
d
gene lo
cu
s.
(2) So
rt the calcul
ated ab
solute value of the deviati
on from small to
big, the resul
t
is marke
d
a
s
b'. Determi
n
e
the thresh
old
by the variation ra
te, meta
morp
ho
sis th
e gene
which
is greate
r
th
an
this thre
shol
d
.
The way is the sam
e
as d
e
termini
ng th
e locatio
n
of disso
c
iation g
ene.
(3) Mut
a
tion operation, the
method is a
s
follows:
()
()
(
)
()
(
)
(1
)
1
()
(
1
)
((
1
)
(
1
)
)
2
i
ii
ii
rm
m
rm
rm
m
n
rm
rm
m
n
1
(8)
2.5. Termination
Rule
If the popula
t
ion gene
rate
d by the algorithm sa
tisfi
ed the co
nst
r
aints, the al
gorithm
stop
s. The p
opulatio
n is
optimal, it is con
s
ide
r
e
d
as the a
p
p
r
o
x
imate optimal solutio
n
af
ter
decodin
g
ope
ration. De
co
d
i
ng is the inve
rse o
p
e
r
ation
for codi
ng.
2.6. Algorithm
Pr
ocess
Step 1
initial popul
ation T
a
ke the ima
g
e
s proc
esse
d
by wavelet tran
sform a
n
d
Wiene
r
filter as mal
e
and femal
e
of the initial populat
io
n
of a hybrid wavelet tra
n
sform. They
are
recorded
as
“pare
n
t one
”
and “pa
r
ent t
w
o”, the
s
u
p
e
rscript 0 rep
r
esents th
e i
n
itial pop
ulation,
and so on.
Step 2
Codin
g
Code
the t
w
o im
age
s to
be
cros
se
d according
to the
metho
d
s given
in
se
ction 2.1, a
nd form two p
opulatio
ns.
Step 3
Sele
ction Ma
ke
the
sele
ction
ope
ration i
n
a
c
co
rdan
ce
with
section
2.2, an
d the
n
determi
ne the
individuals n
eedin
g
crossover or mutati
on.
Step 4
Cro
ssover Ma
ke th
e gene
s reco
mbination to
gene
rate ne
w individual
s a
c
cordi
n
g
t
o
sect
io
n 2.
3
.
Step 5
Mutati
on Make the mutation ope
ration acco
rdi
ng to se
ction
2.4.
Evaluation Warning : The document was created with Spire.PDF for Python.
TELKOM
NIKA
e-ISSN:
2087
-278X
Mathem
atical
Analysi
s an
d
Application o
n
Mech
ani
cal
Im
age of Hybrid…
(Fu
z
en
g Yang)
3467
Step 6
Jud
g
m
ent Jud
ge the
hyb
r
id o
peratio
n
i
s
t
e
rmin
ated
or not
acco
rdi
ng to
the
terminatio
n rule given in se
ction 2.5. If it is
termin
ated, the approximate op
timal solution
is
output and th
e hybrid op
eration stop
s;
o
t
herwi
se, tra
n
sferring to ste
p
4.
3. Results a
nd Analy
s
is
3.1
. Theoretic
al analy
s
is
3.1.1. Coding
Completen
ess
Theorem 1
(
Bolzano's theorem)
If
{}
m
P
is a bo
und
ed
sequ
en
ce i
n
n
R
, there a
r
e
conve
r
ge
nt sub
s
e
que
nce
s
in
{}
m
P
.
{}
m
P
is called Cau
c
h
y
sequ
en
ce.
If
0,
,
N
,,
nN
mN
there
nm
PP
.
Theorem 2
(
Cauc
h
y
criterion)
Necessary and
sufficient co
ndition
s for
conve
r
g
ence of
sub
s
e
que
nce
s
{}
m
P
is that
{}
m
P
is Ca
uchy sequ
en
ce.
Numeri
cal m
a
trix
(
256
,
2
5
6
)
M
corre
s
p
ondin
g
to a
g
r
ay ima
ge i
s
conve
r
ted to
a ne
w
matrix
(
3277,
20)
P
which
is formed by row ve
ctors
(
1
,
20)
i
r
in colum
n
s, here
1
,
2
,
.
..,
3277
i
. So we c
a
n treat the matrix
(
3277,
20)
P
as a spa
c
e
colle
ct
ion co
nstitut
ed by
(
1
,
20)
i
r
,
c
r
ed
ite
d
as
12
2
0
{
(
,
,
...,
)
|
,
1
,
2
,
.
..,
20
}
ni
Vx
x
x
x
R
i
. Suppo
se fo
r an
y vector
11
2
0
(
,
...,
)
rx
x
,
21
2
0
(
,
...,
)
ry
y
in
n
V
.
Firstly, prove
n
V
ca
n be
con
s
i
dere
d
a
s
a m
e
tric
sp
ace.
Define
the di
stan
ce
betwe
en
1
r
and
2
r
as:
2
1
2
12
1
2
12
1
(,
)
(
,
)
(
)
n
ii
i
dr
r
r
r
r
r
r
r
r
r
(9)
then
12
1
2
(,
)
dr
r
r
r
satisfie
s the following
con
d
ition
s
is
calle
d Euclid
ean metri
c
in
n
V
,
Symmetry:
12
2
1
(,
)
(
,
)
dr
r
d
r
r
; Non
n
egative
:
12
1
2
(,
)
0
,
(
,
)
0
dr
r
d
r
r
is equivale
nt to
12
rr
;
Triangle inequality:
12
3
,,
n
rr
r
V
, it is
the cons
tant that
12
1
3
3
2
(,
)
(
,
)
(
,
)
dr
r
d
r
r
dr
r
, and th
e
necessa
ry an
d sufficie
n
t condition
s of
e
s
tabli
s
hme
n
t of the equatio
n is
12
3
,,
rr
r
are o
n
the sa
me
straig
ht line.
Secon
d
ly, pro
v
e the point seque
nces in
n
V
are Cau
c
hy seque
nces.
12
2
0
{
(
,
,
...,
)
|
,
1
,
2
,
.
..,
20
}
ni
Vx
x
x
x
R
i
, beca
u
se
i
x
is from the
gra
y
image pixel
matrix,
i
x
are
real
num
bers bet
wee
n
0 and
255, f
o
r any
seq
u
e
n
ce
11
2
0
(
,
...,
)
n
rx
x
V
is a
bou
nded
s
e
quenc
e
, that is
to s
a
y
{}
i
r
is a boun
ded
seque
nce in
n
V
.
Learning fro
m
Theo
rem 1
(
Bolzano'
s
theore
m
), the
r
e is a co
nvergent sub
s
e
q
u
ence in
{}
i
r
.
{}
i
r
is called Cauchy
sequence, fulfill the
following condition: if
0,
,
N
,,
nN
mN
then
nm
rr
. This sho
w
s that, when the
para
m
eters i
n
n
r
are fixed,
n
r
is a Cau
c
hy sequ
en
ce
. Learnin
g
from Theo
rem
2(Ca
uchy
crite
r
ion
)
, there is a
0
r
,
0
()
m
rr
m
.
Last, prove
n
V
is
Complete
metric
s
p
ac
e.
In fac
t, in inequality
nm
rr
, if let
n
,
0
m
rr
. That expl
ains that
m
r
uni
formly
conve
r
ge
s to
0
r
. By mathematical a
nalysi
s
,
0
r
is a
contin
uou
s fun
c
tion
over a
n
interval. For any
0
n
rV
, and when
mN
,
0
m
rr
, that is
0
()
m
rr
m
. Tha
t
explains th
a
t
n
V
is a
compl
e
te met
r
ic
spa
c
e. Th
erefo
r
e, t
he coding o
peration is complet
ed. QED.
Evaluation Warning : The document was created with Spire.PDF for Python.
e-ISSN: 2
087-278X
TELKOM
NIKA
Vol. 11, No. 6, June 20
13 : 3463 – 3
472
3468
3.1.2. Algorithm
Conv
ergence
There are the
related defini
t
ion and theo
rem.
Defini
tion 1:
If
'
{(
)
}
0
PM
C
x
x
,
()
M
Cx
rep
r
e
s
ents p
o
int se
quen
ce
gene
rated by the
c
r
oss
o
ver
and
mu
ta
tion
op
e
r
a
t
ors
,
{.
}
P
rep
r
esents the
p
r
oba
bility of a
ra
ndom
eve
n
t
{.}
, the
individual
'
x
is called reachable from
x
through cro
s
sove
r and mutatio
n
.
Theorem 3
:
Let a Geneti
c
Algorithm fulfill the followin
g
con
d
ition
s
[18]:
(1) T
he pop
ul
ation se
que
n
c
e
(0)
,
(
1
)
,
.
.
.
PP
is monoto
ne, i.e.
t
:
m
i
n
t
1
|
t
1
P
t
1
mi
n
t
|
t
P
t
,
aa
aa
(10)
(2)
,,
aa
I
a
is
r
e
ac
ha
ble
fr
o
m
a
by mean
s of mutation and reco
mbination. T
hen:
{l
i
m
(
)
}
1
t
Pa
P
t
(11)
The proof of the co
nverg
e
n
c
e of the algo
rithm.
Firstly, prove
that two
ran
d
o
m individ
ual
s
'
x
and
x
in in
di
vidual spa
c
e,
'
x
is
r
e
ac
ha
b
l
e fr
o
m
x
throug
h cro
s
sover and m
u
tation.
In orde
r to p
r
ove this, th
at is to p
r
ove
'
{(
)
}
0
PM
C
x
x
, here
()
M
Cx
re
pre
s
ent
s poi
nt
seq
uen
ce
ge
nerate
d
by th
e cro
s
sover a
nd mutatio
n
operators,
{.
}
P
represent
s the
probability
of a rand
om
event
{.}
. Supposi
ng that e
v
ent A(X) re
pre
s
ent
s the
individual
s gene
rated b
y
cro
s
sove
r o
p
e
rato
r, event
B(X)
re
pre
s
ents th
e in
di
viduals ge
nerated by
muta
tion op
erato
r
,
m
expre
s
ses th
e num
ber
of inferio
r
ge
ne l
o
cu
s d
e
te
rmi
ned by
crossover o
per
ator, k expresse
s the
numbe
r of v
a
riant g
ene
locu
s d
e
term
ined by m
u
tation op
erat
or, he
re
,
1
,2
,
.
.
.
,2
0
mk
.
Therefore,
'
''
'
'
''
'
'
'
{
(
)
}
(
(
)
)
(
(
)
)
((
(
)
)(
(
)
))
(
(
))
(
(
))
(
(
))
(
(
(
)
)
|
(
)
)
)
20
20
20
20
PM
C
x
x
P
A
xx
P
B
xx
P
A
xx
B
x
x
P
A
xx
P
B
xx
P
A
xx
P
B
xx
A
x
x
mk
m
k
m
m
1
,
2
,
...,
2
0
m
so,
0
m
0
20
m
that is
'
{(
)
}
0
PM
C
x
x
Con
s
e
quently
, two ra
ndom
individual
s
'
x
and
x
in indivi
dual
spa
c
e,
'
x
is rea
c
habl
e
from
x
throug
h crossover
an
d mutation.The
n
p
r
ove po
pulation
seq
uen
ce
(
1
)
,
(
2
),
...
,
(
)
pp
p
t
is
monotoni
c, th
at is for
t
, any solution i
n
(1
)
pt
is non i
n
ferio
r
or at lea
s
t n
o
t worse th
an
an
y
solution in
()
pt
. Learning from optimum
maintainin
g
st
rategy of
mutation a
nd sel
e
ctio
n
operator, offspring p
opul
ation se
que
nce prod
uced by this alg
o
rithm
is mon
o
tonic.
That is for
t
any s
o
lution in
(1
)
pt
is non inferior or at lea
s
t not
worse tha
n
any solutio
n
in
()
pt
.
Last, learning
from Theo
re
m 3, the algorithm is conve
r
gen
ce.
3.2
The image d
e
-noising a
p
plication
for
Agricultural machiner
y
p
a
rts
Evaluation Warning : The document was created with Spire.PDF for Python.
TELKOM
NIKA
e-ISSN:
2087
-278X
Mathem
atical
Analysi
s an
d
Application o
n
Mech
ani
cal
Im
age of Hybrid…
(Fu
z
en
g Yang)
3469
High
-definitio
n image
s of
ploug
h and
di
sk
har
ro
w
we
re sele
cted in
the experi
m
ent. The
image resolu
tions of plou
gh and di
sk harro
w we
re
respe
c
tively 2133
×19
79
and 22
72
×18
64.
After
addin
g
Gau
ss white
noise
with m
ean
0,
va
rian
ce 0.0
5
to im
age
s, we
u
s
e
d
a vari
ety of de-
noisi
ng m
e
th
ods to
verify the a
c
curacy
and validity
o
f
the algo
rith
m. Then im
a
ges
of di
sk
h
a
rrow
were redu
ce
d
to 100
0×820
and 4
0
0
×
32
8
.
And addi
ng
Gau
s
s white
noise with
m
ean 0, va
rian
ce
0.05 to image
s, we viewed
effects of this
algorithm o
n
different re
sol
u
tion image
s.
Based o
n
hybrid wavelet transfo
rm alg
o
ri
thm, co
din
g
freque
ncy i
m
age which is after
wavelet tran
sforming
wa
s recorded a
s
P1, wiene
r filtering ima
ge
codi
ng was reco
rde
d
as P
2
;
para
m
eters
were listed in
table 1. As sho
w
n in
tabl
e 1, the optional len
g
th o
f
a chrom
o
so
me
were
re
spe
c
t
i
vely 4, 8, a
nd 1
6
; the
selectio
n
rate and hybridi
z
ation
rate we
re
0.3
an
d step
length
wa
s 0.3, and the
rat
e
s
were set to 0.3, 0.
6, an
d 0.9; the mu
tation rate
wa
s 0.1 a
nd ste
p
length was 0
.
2, the rate wa
s set to 0.1, 0.3,
and 0.5(the mutati
on rate ex
ce
eding 0.5
will
be
redu
ce
d to a rand
om search algorith
m
);
and the hybri
d
gene
ration
were 2, 5, and10.
Four pa
ramet
e
rs were
fixed and
only
o
ne pa
ram
e
ter wa
s alte
re
d
in every
exp
e
rime
nt.
Programmin
g
by setting the initial value
and ste
p
len
g
th, 27 imag
es after hyb
r
i
d
izatio
n ca
n be
obtaine
d thro
ugh alte
ring
o
ne pa
ram
e
ter every time. Select ima
ge
with the be
st visual
effect an
d
the highe
st p
eak
sign
al-to
-
noise ratio (P
SNR) as the
optimal sol
u
tion.
Table 1. Para
meters incl
ud
ed in algo
rith
m (plou
gh)
Original Image
Size
Chromo
- s
o
me
Length
Chromo
- s
o
me
Number
Selection
Rate
Cr
ossover
Rate
Mutation
Rate
Gene
ration of
Hy
b
r
id
2133×1979
4
16384
0.3/0.6/0.9
0.3/0.6/0.9
0.1/0.3/0.5
2/5/10
2133×1979
8
8192
0.3/0.6/0.9
0.3/0.6/0.9
0.1/0.3/0.5
2/5/10
2133×1979
16
4096
0.3/0.6/0.9
0.3/0.6/0.9
0.1/0.3/0.5
2/5/10
a. Origin
al
Image
b. Nois
y
Imag
e
c.
Medi
an filter Image
d. Neig
hb
orho
od aver
agi
ng
method for d
e
-noisi
ng im
ag
es
d.
W
a
velet transf
o
rm for de-
noisi
ng th
e ima
g
e
e.
H
y
brid
w
a
ve
let
transform for
de-n
o
isi
ng of i
m
age
Figure 1. De-noisi
ng Resul
t
s of the Plough Image
Evaluation Warning : The document was created with Spire.PDF for Python.
e-ISSN: 2
087-278X
TELKOM
NIKA
Vol. 11, No. 6, June 20
13 : 3463 – 3
472
3470
3.2.1.
Comparis
on
h
y
brid
w
a
v
e
let
tran
sfor
m algorithm
w
i
th
conv
entional d
e
-n
oising
method
s
In order to
compa
r
e th
e
algorith
m
with th
e
conven
tional d
e
-n
oising m
e
thod
s,
we
d
e
-
noise the
ima
ges of pl
oug
h
and
di
sk ha
rrow u
s
ing
G
a
ussian
filter, t
he
Wien
er filter, the
medi
an
filter, and the averag
e filter at the same time, de-n
o
isi
n
g effects a
r
e sho
w
n in Fig
u
re 1
c
to Figu
re
1f and Figu
re
2c to Figu
re 2
f
.
a.
Origin
al Imag
e
b.
Noi
s
y Image
c.
Median Filte
r
image
d. Neig
hb
orh
ood
Averag
ing
Method for d
e
-noisi
ng
Images
e.
W
a
velet T
r
ansform for de-
Noisi
ng the Im
age
f.
Hy
brid Wavelet
T
r
ansform for
de-n
o
isi
ng of Image
Figure 2. De-noi
sin
g
Re
sults of the Disk Ha
rrow Ima
g
e
Median
filter,
neig
hbo
rho
od ave
r
agi
n
g
me
tho
d
a
nd oth
e
r
co
nventional
d
e
-noi
sin
g
method
wea
k
noi
se to some extent, but the w
hol
e image i
s
fuzzy, the ed
ge feature is not
appa
rent, an
d de-n
o
isi
n
g
effect of the backg
ro
u
n
d
is not goo
d; while u
s
in
g the pro
posed
method, the
noise is si
gnif
i
cantly
red
u
ced, and it kee
p
s ma
ny det
a
ils of image, the edg
e feature
is clea
r. To analysi
s
the advantag
es
and dis
adva
n
tage
s of variou
s de-noi
sing method
s,
th
e
nume
r
ical val
ues are give
n in tabl
e 2.
We
ca
n s
ee
that the pe
ak sign
al to
noi
se
ratio
(PSNR)
values we
re 164.64 and
1
62.03dB
fo
r ploug
h
an
d
d
i
sk ha
rro
w using
hyb
r
id wa
velet
tran
sform
algorith
m
, it is si
gnificantl
y
higher tha
n
that
of co
nventional
d
e
-noi
sin
g
me
thods,
relativ
e
to
wavelet tran
sform, PSNR i
s
also rai
s
ed.
3.2.2.
The Applica
t
ion Effec
t
of
Differen
t Re
solution Ima
g
e
In orde
r to view the ap
plication effects
of different re
solutio
n
imag
es, this alg
o
ri
thm wa
s
applie
d to the
narro
wing
disk
harro
w im
age, the ima
ge re
sol
u
tion
were 10
00
×8
20an
d 40
0×3
28,
and add
ed
Gau
ss
white
noise with
mean 0, vari
ance 0.05 to images. At the sam
e
time we
comp
ared wit
h
the re
soluti
on of 2272
×
1864 di
sk ha
rro
w imag
e a
s
sh
own in F
i
gure 3, a to c
were for noi
sy images, d to e were for d
e
-noi
sin
g
ima
ges.
From ta
ble 3
we
can
se
e
that with the
redu
cin
g
of resol
u
tion, the
pea
k si
gnal
to noise
ratio (PSNR)
of the processed ima
g
e
s
are also
decre
ase
d
. Combi
ned with figure 3, de-noi
si
ng
effect is obvio
usly de
cre
a
sed.
Evaluation Warning : The document was created with Spire.PDF for Python.
TELKOM
NIKA
e-ISSN:
2087
-278X
Mathem
atical
Analysi
s an
d
Application o
n
Mech
ani
cal
Im
age of Hybrid…
(Fu
z
en
g Yang)
3471
a.
2272
×1
864 n
o
isy image
b.
1000
×8
20 noi
sy image
c.
400
×32
8
noi
sy image
d. 2272
×1
864
d
e
-noi
sin
g
image
e. 1000
×8
20
de
-noi
sing
image
f. 400
×32
8
de-noisi
ng
image
Figure 3. De-noi
sing
Re
su
lts of Differen
t
Resol
u
tion
s
Table 2. Anal
ysis of the Re
sults fo
r PSNR
De-noising
Methods
Resolution
rati
o
Noisy
image
Median
F
ilter
Neighbor
hood
averaging
Wavelet
Transform
Hy
b
r
id
Wavelet
Transform
PSNR
Plough 2133×1979
146.65
159.69
157.69
155.40
164.64
Disk harro
w
2272×1864
145.27
158.54
159.25
156.17
162.03
Table 3. Anal
ysis of the Re
sults fo
r PSNR
resolution
ratio
2272×1864
1000×820
400×328
PSNR
disk
harro
w
noisy
image
145.27
145.33
145.20
de-noising image
162.03
160.30
157.11
4. Conclusio
n
(1) By the in
spiration of h
y
bridization b
r
eedi
ng, this
arti
cle
propo
ses ima
ge d
e
-noisi
ng
algorith
m
ba
sed on hyb
r
id
wavelet tra
n
sform. T
he a
n
a
lysis from m
a
themat
ical theory
sho
w
s that
codi
ng op
erat
ion is complet
e
, and the alg
o
rithm is
con
v
ergent.
(2)
The m
e
th
od is
applie
d
to mechani
cal image
de
-noisi
ng
su
ch
as pl
oug
h a
nd di
sk
harro
w, the p
eak
sig
nal to
noise ratio
(P
SNR)
of the d
e
-noi
se
d ima
ge were
164.
64 (pl
oug
h) a
nd
162.03 (disk
harro
w), it is the be
st in so
me tradi
tion
al
de-noi
sin
g
m
e
thod
s su
ch
as nei
ghb
orh
ood
averagi
ng me
thod (15
7
.69
and 15
9.25),
median filter(159.69 a
nd 1
58.54).
(3) T
he expe
rimental
resu
lts sho
w
that
im
age de
-n
oisin
g
algo
rithm ba
sed o
n
hybrid
wavelet tra
n
s
form
appli
e
d to mecha
n
ical field
h
a
s a
high
PSNR valu
e, appa
rent e
dge
cha
r
a
c
teri
stics, and g
ood
visual effect
and so
on. E
x
perime
n
tal data verifie
s
the validity and
feasibility of this algorithm.
(4) T
he exp
e
r
iment al
so
showed that
with t
he red
u
cin
g
of re
solution, the
algorithm
pro
c
e
ssi
ng
effect is al
so d
e
crea
sed,
so
it rem
a
ins to
be im
prove
d
. But for the
high
re
solutio
n
image, the al
gorithm
can b
e
use
d
effecti
v
ely.
Evaluation Warning : The document was created with Spire.PDF for Python.
e-ISSN: 2
087-278X
TELKOM
NIKA
Vol. 11, No. 6, June 20
13 : 3463 – 3
472
3472
Akno
w
l
e
dge
ment
The
re
sea
r
ch
presented
in
this
arti
cle
wa
s
sup
porte
d by the
Chi
nese fun
d
s:
“Nation
a
l
Natural S
c
ie
nce
Fo
und
ation of
Chi
n
a (309
7169
0
)”,
“Proj
e
ct
s in th
e
Nati
onal S
c
ien
c
e
&
Technology
Pillar Program during
the Twelfth Five-year Plan
Period (2011BAD
20B10-3)” and
“The Fi
rst Cultivate proje
c
t of Yanglin
g mode
rn
ag
ricultu
r
e inte
rnational
in
stitute (K201
1-1
0
)”.
Any opinion
s,
findings, a
n
d
con
c
lu
sion
s
expre
s
sed
in
this arti
cle a
r
e those
of the
authors an
d
d
o
not necessa
ri
ly reflect the views
of the
Northwes
t A&F Univers
i
ty.
Notation
(
1
,
20)
i
r
: A ro
w
v
e
ctor
constituted of
t
w
e
n
t
y
e
l
eme
n
ts, in this alg
o
rithm it repre
s
ents an in
divi
dua
l (o
r
chromos
o
me);
j
:
T
he number o
f
chromosome
s,
[
256
2
5
6
2
0
1
]
3277
j
;
i
:
T
he location
of the current chromos
o
me,
1
,
2
,
...,
ij
and i is an i
n
teg
e
r;
n
:
T
he gene n
u
m
ber of chrom
o
somes,
20
n
;
m
:
T
he locus loc
a
tion of the cur
r
ent chromos
o
me,
1
,
2
,
...
,
mn
and m is an inte
ger;
()
()
i
rm
:
T
he locus loc
a
tion
m
of the current chromos
o
me
()
i
r
;
()
i
r
:
T
he mean val
ues of all g
e
n
e
s
the current chromos
o
me
()
i
r
.
Referen
ces
[1]
He Q, W
ang L, Yu Z
,
Li G,
Gao L. Defect det
ectio
n
for microg
ear bas
ed
on imag
e pro
c
essin
g
a
n
d
supp
ort vector machi
ne.
Jour
nal of Jil
i
n Un
iv
ersity: Eng an
d
T
e
chnol Ed.
2
008; 38(
3): 565
-569.
[2]
Z
hu H, Jia
ng
P, Yang Y, M
a
C. Improve
d
defects d
e
tec
t
ion a
l
gor
ithm for
machi
ned
part
surface.
Transduc
er and Microsystem
Technologies
.
200
6; 25(1
1
): 66-69.
[3]
Starck JL, Ca
n
dès EJ, D
ono
h
o
DL. T
he cur
v
elet transf
o
rm for imag
e d
e
n
o
isin
g.
IEEE Transactions
on Image Proc
essin
g
. 200
2; 11(6): 670-
68
4.
[4]
W
a
lker JS, Ch
en YJ. Image d
eno
isin
g usi
ng
tree-bas
ed
w
a
velet sub
b
a
nd
correlati
ons a
n
d
shrink
ag
e.
Op
ti
ca
l
En
gi
nee
ri
ng
. 200
0; 39
(11): 290
0-2
9
0
8
.
[5]
Jalo
bea
nu
A, Blanc-F
ér
aud
L, Z
e
rubi
a J,
editors.
S
a
tell
i
t
e
i
m
ag
e dec
onvo
l
utio
n usi
ng
c
o
mpl
e
x
wavelet packet
s
. IEEE Proceedin
g
s Internati
ona
l Conf
erenc
e. 2000.
[6]
T
homas BA, Rodri
g
u
e
z JJ.
Editors.
Wav
e
let-based color image
denoising. IEEE
Proceedings
Internatio
na
l C
onfere
n
ce. 20
0
0
: I.
[7]
Che
n
G, Bui
T
.
Multi
w
ave
l
e
t
s den
oisi
ng
u
s
ing
ne
igh
bor
i
ng c
oefficie
n
ts.
Sign
al Pr
oce
ssing
Letters,
IEEE
. 2003; 10
(7): 211-2
14.
[8]
Katkovnik V, Kgiaz
a
ria
n
K, As
tola J. Adaptiv
e
w
i
nd
o
w
size i
m
age d
e
-no
i
si
ng bas
ed on i
n
tersection o
f
confid
enc
e inte
rvals (ICI) rule.
Journa
l of Mathe
m
atic
al Imag
ing a
nd Visi
on.
2002; 1
6
(3): 2
23-2
35.
[9]
Vid
y
a
H, T
y
a
g
i
B, Krishna
n V
,
Mallikar
ju
nap
pa K.
Rem
o
val
of Interferenc
es from Partia
l
Dischar
g
e
Pulses usi
ng W
a
velet
T
r
ansform.
T
E
LKOMNIKA Indo
nesi
an J
ourn
a
l
of Electrical
En
gi
neer
ing
. 20
11
;
3(1): 107-
11
4.
[10]
Hua
ng J,
Xie J
,
Li H, T
i
an G,
Che
n
X. Se
lf
-a
daptiv
e D
e
com
positi
on
Lev
el
De-n
oisi
ng M
e
thod
Base
d
on Wavelet T
r
a
n
sform.
T
E
LKOMNIKA Indon
esia
n Journ
a
l o
f
Electrical Eng
i
ne
erin
g
. 201
2; 10(5).
[11]
Chmelk
a
L, Ko
zumplik
J, Iee
e
.
W
a
vel
e
t-bas
ed W
i
e
ner fi
lte
r
for el
ectrocar
dio
g
ra
m si
gn
al
de
nois
i
ng
.
Co
mp
uters in
Cardi
o
l
ogy.
20
05; 32: 77
1-77
4.
[12]
Kazub
e
k M.
W
a
velet d
o
ma
in im
age
d
eno
is
ing
b
y
thr
e
s
hol
din
g
a
nd
W
i
ener fi
lterin
g.
Ieee S
i
g
n
a
l
Processi
ng Let
ters.
2003 Nov;
10(11): 32
4-3
26. PubMe
d
P
M
ID: W
O
S: 00018
61
363
00
00
3.
[13]
Yang
F
,
T
i
an
Y, Yang
L,
He
J. Agric
u
ltura
l
ima
ge
de-
noi
sing
al
gorit
hm bas
ed
on
h
y
brid
w
a
v
e
le
t
transform.
T
r
ansactio
n
s of the CSAE
. 201
1; 27(3): 172-
17
8.
[14]
Yang F
,
W
a
n
g
H, Yan
g
Q, W
ang Z
.
W
a
ve
let transfo
rm
a
nd its
app
licati
on i
n
the
proc
essin
g
of frui
t
imag
e.
T
r
ansa
c
tions of CSA
M
. 2005; 36(
00
5): 61-64.
[15]
Yang
F, Wang
Z, Han W, Ya
n
g
Q. Wavel
e
t trans
form-b
ased
muhisc
al
e e
d
g
e
d
e
tectio
n of
deh
iscen
c
e
furro
w
in C
h
in
e
s
e date ima
ge.
Transactions of
CSAE.
2005
; 21(6): 92-95.
[16]
Yang
F
,
W
ang
Z
,
Yang Q. Agr
i
cultur
al im
ag
e
enh
anc
ement
method
Base
d
on th
e
w
a
vel
e
t
transform
.
T
r
ansactio
n
s o
f
the CSAE
. 2004; 20(3): 1
24-
129.
[17]
Blickle
T
,
T
h
iel
e
L. A
com
pari
s
on
of se
le
ctio
n sch
emes
use
d
i
n
ev
ol
ution
a
r
y
al
gorithms.
Evoluti
ona
r
y
Co
mp
utation.
1
996; 4(4): 3
61-
394.
[18]
Back T
.
Evolution
a
r
y
alg
o
r
i
thms in the
o
r
y
a
nd pr
ac
tice: evoluti
o
n
strategies, evol
ution
a
r
y
progr
ammin
g
, gen
etic al
gorith
m
s. Oxfo
rd Uni
v
ersit
y
Press;
USA. 1996.
Evaluation Warning : The document was created with Spire.PDF for Python.