TELKOM
NIKA
, Vol.11, No
.1, Janua
ry 2013, pp. 411
~41
6
ISSN: 2302-4
046
411
Re
cei
v
ed Au
gust 16, 20
12
; Revi
sed
De
cem
ber 2, 20
12; Accepted
De
cem
ber 1
1
,
2012
A Wavelet-based Algorithm for Vehicle Flow
Information Extraction
Zhi Qiao*
1
, Li-duan Liang
2
, Lei Shi
3
, Ling-ling Li
4
1,3
Department of Electrical a
n
d
Electron
ic En
gin
eeri
ng, He
n
an Un
iversit
y
o
f
Urban Co
nstruction,
Ping
din
g
sh
an, Chin
a
2,4
Department of Mathematics
and Ph
ysi
cs,
Hen
an Un
ivers
i
t
y
of Urb
an Co
nstruction, Pin
gdi
ngsh
an, Ch
i
n
a
*Corres
p
o
ndi
n
g
author, e-ma
il: richard
q
ia
o5
109
@
y
ah
oo.co
m
1
, liangli
d
u
a
n
198
1@
ya
ho
o.cn,
shile
iston
e4@
gmail.c
o
m, lili
n
g95
04
06@
163
.com
A
b
st
r
a
ct
T
h
is pa
per
pr
opos
ed a
n
i
m
p
r
oved
alg
o
rith
m a
ppl
ie
d
in v
i
deo
intel
lig
ent
traffic control s
ystem for
vehicl
e det
ecti
on. T
he accur
a
cy of origi
n
a
l
alg
o
rith
m,
w
h
ich is base
d
on
the co
m
par
isio
n of contrast a
n
d
lu
min
anc
e dist
ortion of prese
n
t ima
ge w
i
th backgr
oun
d, reduces gr
eatly und
er bad w
e
ather bec
ause
of
false d
e
tectio
n
cause
d
by
no
ises i
n
ca
ptur
ed i
m
ag
es
. In this p
aper w
e
chose
Da
ub
e
c
hies w
a
ve
let
as
moth
er w
a
velet
to add a 2-d
i
mensi
on w
a
vel
e
t process bef
or
e the alg
o
rith
m, just after the ima
ge is ca
ptur
ed
,
to de-no
ise e
a
ch captur
ed i
m
a
ge. W
e
used F
P
GA-
bas
ed eq
uip
m
ent
s to test
the alg
o
rith
m, and
the
exper
iment pro
v
ed hi
gh
er perf
o
rm
anc
e of improve
d
alg
o
rith
m, esp
e
cia
lly u
nder b
ad w
eat
her.
Key
w
ords
: vehicl
e flow,
contrast dist
ortion, lu
m
i
nance distorti
on, Daube
ch
ies wa
velet, de-
noisi
ng
Copyrig
h
t
©
2013
Univer
sitas Ahmad
Dahlan. All rights res
e
rv
ed.
1. Introduc
tion
To solve the
proble
m
of urban
con
g
e
stion
a
nd o
b
stru
ction, video intellige
n
t traffic
control syste
m
is well-kn
o
w
n as the be
st appr
oa
ch, in which the length of red
and green lig
hts
time are varia
b
le throu
gh vehicl
e flow. There a
r
e 3
su
bsyste
ms in video intellige
n
t traffic contro
l
system,
whi
c
h a
r
e ve
hi
cle-dete
c
ting
sub
s
yste
m, vehicle
-
cou
n
ting sub
s
ystem and ti
me-
controlling
su
bsyste
m. Veh
i
cle-dete
c
ting
sub
s
ystem,
whi
c
h is m
a
in
ly for detectin
g
the vehicl
e in
a given are
a
,
is the most important part in vi
deo
intelligent traffic control system, and
it
determi
ne
s the efficiency
of whole sy
stem.
The algorithm ba
se
d on contra
st and luminance
distortio
n
[1] is an accu
rate
algorithm for vehicl
e dete
c
tion. That is, compa
r
ing the cont
rast a
nd
luminan
ce di
stortion of p
r
esent imag
e, which
is captured with certain freque
nci
e
s, with
backg
rou
nd image. If the
2 param
eters pass throug
h a
thresh
old
from large to small, there is a
car p
a
ssin
g into the desig
nated area, and then if
they pass through an
othe
r thresh
old from
small to larg
e, there is a
car pa
ssing
away from the are
a
. Wh
en the 2 pro
c
e
s
ses h
app
en
su
ccessively, vehicle-cou
n
t
ing sub
s
yste
m w
ould co
u
n
t up by one. The backg
round imag
e is
update
d
every fixed interval of time. Howeve
r,
whe
n
weathe
r gets too bad, such a
s
sn
ow,
rain
storm a
n
d
sand
storm, the ca
pture
d
image woul
d
be full of noises. The
r
efore
,
sometime
s the
contrast di
st
ortion a
nd lu
minan
ce di
st
ortion
m
a
y pass the 2 p
r
oce
s
se
s but
there i
s
no
car
passin
g
, cau
s
ing error to th
e result.
In this paper, we focus
on the impro
v
ement of this algo
rithm. We add a wavelet
pro
c
e
ssi
ng [2
] after ca
pture of ba
ckgro
und ima
ge a
nd present i
m
age fo
r
de
-noisi
ng, befo
r
e the
pro
c
e
ss of th
e algorithm b
a
se
d on co
ntrast and
lumi
n
ance disto
r
tio
n
. Daube
chi
e
s wavelet (db
N
)
has a g
ood e
x
pansi
on abili
ty [3] so we choo
se Daub
e
c
hie
s
(db8
) a
s
mother wave
let.
The pap
er was organi
zed
as follows. Section 2 de
scribe
d the algorith
m
ba
sed on
contrast and l
u
minan
ce di
stortion, which
was or
igi
nal algorith
m
. Section 3 analyzed Daube
chi
e
s
wavelet tra
n
sformation,
ste
p
s of getting
filter
coeffici
e
n
ts and
wave
let de-n
o
isi
n
g
,
and structu
r
e
of improved a
l
gorithm. Re
sults of simula
tion and
com
pari
s
on exp
e
riment were shown in Section
4. Finally we made con
c
lu
sion in Se
ctio
n 5.
Evaluation Warning : The document was created with Spire.PDF for Python.
ISSN: 23
02-4
046
TELKOM
NIKA
Vol. 11, No
. 1, Janua
ry 2013 : 411– 4
1
6
412
2. The Algori
t
hm Bas
e
d o
n
Contras
t
a
nd Luminan
ce Distortion
To start the
whol
e syste
m
, the came
ra sho
u
ld be
adju
s
ted so that it can ca
pture the
image of ap
p
r
op
riate a
r
ea.
Gene
rally the length of th
e are
a
is a
s
same a
s
that of a ca
r, and
the
width is a
s
sa
me as the roa
d
. That is sh
o
w
n in Figu
re
1.
Figure 1.
Ch
oo
s
i
ng
a
r
ea
Defini
tion
is backgroun
d image an
d
is pre
s
ent ima
g
e
, and ea
ch i
m
age is
with N
pixels. The g
r
ay value of each pixel i
s
written as
(1)
(2)
The varia
n
ce of
and
are
(3)
(4)
Therefore, th
e contrast di
stor
tion of an image is
writt
en as
(5)
And the lumin
ance disto
r
tio
n
is
(6)
In equation
(6),
and
are
the averag
e value of
and
.To avoid
that th
e
denomi
nato
r
of either
or
is 0,
and
are a
dded.
and
.
is dynami
c
range of pixe
l grayscal
e (e.g. if
the image i
s
8-bi
t grayscale i
m
age,
is 255
).
Gene
rally an
d
.
In the whol
e
system the
backg
rou
nd i
m
age i
s
pe
ri
odically upda
ted, but it should
n
’t
inclu
de ca
rs. Thus a threshold (b
ackg
ro
und thre
sh
ol
d
)
sho
u
ld be set before. Th
at is, when it is
time for b
a
ckgrou
nd
updati
ng but
and
are belo
w
th
e thre
shol
d, there mu
st be
ca
rs in th
e
area a
nd b
a
ckgroun
d would not up
date. After
obtainin
g
ba
ckgro
und, th
e system st
arts
X
Y
{|
1
,
2
,
,}
i
x
xi
N
{|
1
,
2
,
,
}
i
yy
i
N
x
y
22
1
1
()
1
N
xi
i
x
x
N
22
1
1
()
1
N
yi
i
yy
N
1
22
1
2
xy
xy
C
c
C
2
22
2
2
()
(
)
xy
C
lum
x
yC
x
y
x
y
c
lum
1
C
2
C
2
11
()
CK
L
2
22
()
CK
L
L
L
1
0.
0
3
K
2
0.
01
K
c
lum
Evaluation Warning : The document was created with Spire.PDF for Python.
TELKOM
NIKA
ISSN:
2302-4
046
A Wavelet-b
a
s
ed Algo
rithm
for Vehicle Fl
ow Inform
atio
n Extra
c
tion (Zhi Qiao
)
413
initialization p
r
ocess, wo
rk
i
ng out the varian
ce
and averag
e value
of backgro
und image.
After that the
system
sta
r
ts real-tim
e samp
ling
process which sa
mples the im
age of
roa
d
and
then works
out the
and
of pre
s
e
n
t image, the
n
works o
u
t
and
accordi
ng
to
equatio
ns
(5
)
and (6).
Whe
n
the differen
c
e
s
of
p
r
e
s
en
t and ba
ckgro
und ima
ge a
r
e quite mo
de
st,
and
are ne
ar
to 1. And wh
en a car
run
s
into
the ch
osen area
com
p
letely, they are to the
minimum (ne
a
r to 0). Th
erefore, we ca
n
set
2 thre
sh
o
l
ds
and
throu
gh expe
rimen
t
and if
and
pass through
from large to small, there i
s
a car passi
ng into
the area. And then if
they pass th
rough
from small to
large,
the ca
r i
s
pa
ssing
away, and after th
e 2
pro
c
e
s
ses th
e
system
count
up by one.
The co
ntra
st distortio
n
in the algo
rithm is
releva
nt to the variance
s
of pixel gra
y
scal
e
thus the sha
dow effect co
uld be efficie
n
tly suppr
essed. And luminan
ce disto
r
tion is releva
nt to
the averag
e of pixel grayscale, so the a
c
cura
cy
of algorithm won’t
be affected by light chan
ges.
And the al
go
rithm is a
bout
contrast
and l
u
minan
ce
distortion, not p
o
int-to-point
correspon
den
ce,
thus the a
ccura
cy wo
n’t be affecte
d
b
y
came
ra sh
akin
g. Ho
wev
e
r, bad
weat
her
woul
d ca
use
noises in
capt
ured im
age, thus affe
cting the
accuracy seri
ou
sly.
3. Daube
chi
es Wav
e
let and the Impro
v
ed Algorithm
To improve
the accu
ra
cy of
algorithm
unde
r ba
d weath
e
r, a 2
-
dime
nsi
on
wavelet
pro
c
e
ssi
ng i
s
adde
d befo
r
e ori
g
inal
alg
o
rithm. In thi
s
pape
r, we choo
se
Dau
b
e
c
hie
s
wavelet
as
mother wavelet.
3.1. Daube
c
h
ies Wav
e
let
In orthog
onal
conditio
n
th
e Fou
r
ier T
r
ansfo
rm of
Dau
b
e
c
hie
s
(db8) [4]
wav
e
let filter
function
satisf
ies the eq
uati
o
n
(7)
If then
(8)
Hen
c
e
is a real co
efficient
polynomial, and
.
And sin
c
e
,
th
us
(9)
(10
)
From e
quatio
n (9), (11), (1
2) and
Rei
sz
theore
m
,
(11
)
x
x
y
y
c
lum
c
lum
a
b
c
lum
a
b
22
()
(
)
1
HH
()
in
n
n
Hh
e
8
1
()
[
(
1
)
]
(
)
2
ii
He
Q
e
()
i
Qe
2
()
(
)
()
ii
i
Qe
Qe
Qe
2
2
1
(1
)
c
o
s
/
2
2
i
e
2
2
28
(
)
[c
os
(
/
2)]
(
)
i
HQ
e
2
2
28
(
)
()
[
s
i
n
(
/
2
)
]
(
)
i
HQ
e
7
2
22
8
2
7
0
1
()
(
s
i
n
)
(
s
i
n
)
(s
i
n
)
22
2
2
ik
k
k
k
Qe
C
R
Evaluation Warning : The document was created with Spire.PDF for Python.
ISSN: 23
02-4
046
TELKOM
NIKA
Vol. 11, No
. 1, Janua
ry 2013 : 411– 4
1
6
414
Then we co
u
l
d obtain
from equation (7), (8) a
nd (1
1). No
w we
can wo
rk o
u
t
the coeffici
en
ts of Daub
echies
wavelet filter
as follo
wi
ng step
s:
Firstly, we sh
ould obtain
under
, and then we can tran
sfer
to
the function of
, so we obtain a univariate equati
on
with hig
h
e
r
o
r
de
r, and
we
set it a
s
and
. After th
at, all the roots
o
f this equ
ation coul
d b
e
obtained, a
nd then
we u
s
e
to obtain all the
.
Thirdly, as for th
e com
p
lex ro
ots, we pi
ck o
ne pai
r of ea
ch
2 pairs out a
n
d
one of ea
ch
pair a
s
for re
al root
s, we t
herefo
r
e o
b
ta
in
from
them. The
c
o
u
ld be obtai
ned from
, a
nd we co
uld
therefore o
b
tain
with
finally.
3.2. Wav
e
let-based Image
Processing
Wavelet imag
e pro
c
e
ssi
ng is 2-dim
e
n
s
io
n disc
rete wavelet trans
form [5], whic
h c
o
ns
is
ts
of line wavel
e
t transfo
rma
t
ion and col
u
mn wavelet tran
sform
a
tion
. First we sh
ould get low-
freque
ncy filter coeffici
ent
s
and high
-freque
nc
y filter coefficie
n
ts
through the
metho
d
above. As for image de
-n
o
i
sing b
a
sed o
n
Dau
b
e
c
hie
s
wavel
e
t [6], firstly we sh
ould de
co
mp
ose
the image
and obtain wa
velet coeffici
ents
with M
a
llat algorithm, which is described in
Figure 2.
Figure 2.
Mall
at algorithm
Here,
is lo
w-f
r
equ
en
cy filter and
is high
-f
requ
en
cy filter.
is do
wn-
sa
mpling by
two in wavel
e
t transform.
are scale coefficient
s and
are wavel
e
t coeffici
ents.
The image si
gnal is a ma
trix of pixels, and
each pi
xel is the gray value of
a point.
Gene
rally, ea
ch line an
d column of the matrix
coul
d be reg
a
rded
as 1-dime
nsi
on sig
nal [7]. We
start with lin
e pro
c
e
ssi
ng
, using Malla
t algorithm for de
com
posing each line
to get wavelet
coeffici
ents
and scale coefficient
s. Then we
use
soft-thre
shol
d de-noi
sin
g
method [8] fo
r
threshold d
e
-noisi
ng a
s
followin
g
(12
)
2
()
i
Qe
()
H
2
()
i
Qe
()
0
Rx
2
()
i
Qe
cos(
)
()
V
co
s
(
)
i
1
i
z
z
z
()
(
)
i
Qz
A
z
z
A
()
()
Qi
V
i
n
h
()
Qz
n
h
n
g
y
H
G
2
11
2
[]
,
[
]
,
[
]
,
[
]
jj
j
j
ck
c
k
c
k
c
k
1
[]
,
j
dk
12
[]
,
[
]
,
[]
jj
j
dk
d
k
d
k
s
g
n
(
)(
),
0,
ˆ
(,
)
{
tt
H
t
XT
t
Evaluation Warning : The document was created with Spire.PDF for Python.
TEL
K
noisi
n
sam
e
Nex
t
4. Si
m
origi
n
simu
l
are r
214.
3
to be
imag
e
in F
P
impr
o
with
o
follo
w
K
OM
NIKA
In equati
n
g, we u
s
e
e
pr
oc
ess
to
Therefor
e
it’s process
The blo
c
k
m
ulation an
Figure 2
a
n
al imag
e an
Figur
e
The de
c
o
l
atio
n the le
v
emove
d
. W
e
3
. After this
p
identifie
d b
y
In this
e
x
e
captu
r
e
a
n
P
GA, we
ch
o
o
ve
pr
oce
s
s
i
o
ut it sim
u
lta
w
ing:
A Wavelet
-
on (12)
is
inverse pro
c
each col
u
m
n
e
the d
e
-
no
i
s
of the alg
o
ri
t
k
di
a
g
ram of
d Experi
m
e
a
nd 3
sho
w
t
d Figu
re
3 s
h
e
4. Orig
inal
o
mp
osition
v
el of wavel
e
e
us
e
s
o
ft-t
h
p
ro
ce
ss mo
s
y
sy
st
em.
x
pe
ri
ment,
w
n
d L
CD fo
r
d
o
ose liftin
g
-
b
i
n
g
s
p
ee
d.
W
neo
usly to
g
t
I
S
-
b
a
s
ed Algo
r
thresh
old,
a
c
ess
o
f
de
c
o
n
of the re
co
n
s
ed i
m
age
t
hm ba
se
d o
imp
r
oved al
Figure 3.
T
e
nt
t
he si
mulati
o
h
o
w
s the de
-
image
level [9] is
e
t is 5, and
a
l
h
re
sh
old de
-
s
t of the n
o
is
e
w
e ch
oose
F
d
isplay. We
r
b
as
ed
2D
D
W
W
e use
a s
e
g
et the vehi
c
y
Image ca
p
Wav
e
le
t
S
SN: 2302-4
0
r
ithm for Ve
h
a
nd is
w
a
o
mposition t
o
n
st
ru
ct
ed m
a
could b
e
o
n co
ntra
st a
n
gorithm is
s
h
T
he
improve
d
o
n of wavele
t
-
noised im
a
g
an im
porta
l
l the compo
-
noi
sing m
e
t
h
e
s a
r
e r
e
m
o
v
F
PG
A
-base
d
r
epla
c
e sno
w
W
T alg
o
rith
m
e
t of equip
m
c
le flow info
ˆ
y
ture
t
de
co
mposi
n
De-noi
s
Wav
e
l
e
0
46
h
icle Fl
ow In
f
a
velet c
oeffi
c
o
re
con
s
tru
c
a
trix.
o
btaine
d, a
n
d
n
d lumina
nc
e
h
own in
Figu
d
algorithm
t
image
pro
c
g
e.
Figure
nt fa
cto
r
in
nents in lev
e
h
od an
d th
e
v
ed, and th
e
d
eq
uip
m
ent
w
by pape
r
s
m
[11] with
p
m
ent with w
a
r
mation in 1
n
g
ing
e
t rec
o
ns
tru
c
The algo
rit
on co
ntr
luminan
c
e
f
ormation E
x
c
ient, s
o
c
t t
he matrix
d
wav
e
l
e
t p
r
e
distortion.
re 3.
e
ssi
ng. F
i
gu
5. De-noise
pr
ac
tic
a
l a
e
l
1, level 2,
e
threshold
v
e
passing
ca
r
[10], using
s
cra
p
s.
A
s
f
o
p
arallel archi
t
a
velet proce
s
hour. And
t
c
tion
hm
base
d
a
s
t and
di
stortion
x
tra
c
tion (Zhi
. Aft
e
. Th
en we
d
r
ocess i
s
fin
re 2
sho
w
s
t
d imag
e
ppli
cation.
I
le
vel 3 and l
v
al
ue of lev
e
r
i
s
still so o
b
CC
D ca
m
e
o
r wav
e
let p
r
t
ecture whi
c
s
sing and a
n
t
he
r
e
su
lts
a
[]
j
dk
Qiao
)
415
e
r de-
d
o th
e
is
h
e
d
.
t
he
n this
evel 4
e
l 5 is
b
vi
ous
e
ra fo
r
r
ocess
c
h can
n
ot
her
a
re
as
Evaluation Warning : The document was created with Spire.PDF for Python.
ISSN: 23
02-4
046
TELKOM
NIKA
Vol. 11, No
. 1, Janua
ry 2013 : 411– 4
1
6
416
Table 1. Experime
n
t results
No. Weather
Vehicle
flow
Results of
original algorithm/
Error/Ac
c
u
rac
y
Results of improved
algorithm/
Error/ Ac
c
u
rac
y
1 Sun
1587
1635/48/96.9
8
%
1544
/
43/97.2
9
%
2 Sun
1462
1517/55/96.2
4
%
1412/50/96.5
8
%
3 Sno
w
1256
1328/72/94.2
7
%
1300/44/96.5
0
%
4 Sno
w
1518
1621/103/93.
21
%
1568/50/96.8
1
%
5 Sandstorm
1188
1260/72/93.9
4
%
1232/44/96.3
0
%
6 Sandstorm
1203
1274/71/94.1
0
%
1246/43/96.4
2
%
From the ex
perim
ent re
sults it is sho
w
n that
the accuracy of improve
d
alg
o
rithm is
highe
r than
o
r
iginal
algo
rit
h
m. This is
e
s
pe
cia
lly o
b
vious
und
er
ba
d we
ather su
ch a
s
sno
w
a
n
d
san
d
sto
r
m. T
he average
a
c
cura
cy of o
r
i
g
inal alg
o
rith
m in bad
we
a
t
her (No.3
-
6
)
is 93.8
8
%, bu
t
it
is 96.51% of improve
d
algo
rithm.
5. Conclusio
n
In this pape
r we propo
se
d
an improve
d
algorit
hm for vehicle flow
monitori
ng ba
sed o
n
Dau
b
e
c
hie
s
wavelet. The
experiment proved it
s hi
gher a
c
curacy than the algorithm with
out
wavelet p
r
o
c
essing, e
s
p
e
c
ially un
der
bad
weath
e
r.
The vide
o i
n
telligent traf
fic co
ntrol
sy
stem
with the imp
r
ovezzzd algo
rithm would
surely hav
e
stronge
r pe
rformance of
anti
-
interfe
r
en
ce
and
automatically adju
s
t the len
g
th of time more rationally.
Referen
ces
[1]
T
an Jia-jia, Z
hang Jia
n
-qi
u
. Novel visi
on-b
a
s
ed me
thod for rea1-tim
e
on road veh
i
cl
e flow
inf
o
rmatio
n
extracti
on.
Ch
i
nese Jo
urn
a
l o
f
Scientific Instrument
. 200
8; 2
9
(1): 158-
16
6.
[2]
Stepha
ne G.
Mallat. Mu
lti-r
e
sol
u
tion
ap
pr
oxim
ati
ons an
d
w
a
v
e
let orthon
ormal bas
i
s
of
.
T
r
ansactio
n
s o
f
America
n
Mat
h
e
m
atic
al Soci
ety
. 1989; 31
5(
1): 69-75.
[3]
Liu H
u
i-j
i
n, W
u
Yu-ron
g
, Xi
on
g Lin
g
-l
ing. Ap
p
licati
on of
w
a
velet
multir
eso
l
ution decom
po
sition
us
in
g
Dau
bech
i
es i
n
voltag
e flicker
ana
l
y
sis.
RELA
Y
. 2004; 32(
13
): 6-10.
[4]
Albert Bog
ges
s, F
r
ancis J. N
a
rco
w
ic
h. 200
9
.
A
F
i
rst Course in W
a
velets w
i
t
h
F
ourier A
nal
ysis. Guo-
shen
g Rui. Jia
n
Kang:
Pu
blis
hin
g
Hous
e of Electron
ics Ind
u
stry
. 2010.
[5]
Rajl
a
x
mi Ch
ou
han, Ag
ya Mis
h
ra, Pritee Kh
ann
a.
F
i
nger
pr
int Authentic
ati
on b
y
W
a
vel
e
tbase
d
Dig
ita
l
Watermarking.
Internatio
na
l Journ
a
l of Electr
ical a
nd Co
mp
uter Engi
ne
erin
g
. 2012; 2(
4): 519-5
28.
[6]
W
ang Jia
n
-zh
ong, Z
han
g Hui, W
u
Bin,
Chen
g W
en-
hua. Image-
D
eno
isin
g Meth
od Base
d on
Dau
bech
i
es W
a
vel
e
t T
r
ansform and Med
i
a
n
F
ilter.
Journa
l of W
uhan Un
iv
ersity of T
e
chn
o
lo
gy
. 2
0
01;
23(3): 19-
23.
[7]
Liu Qing-
qin
g
, Guo Li-zho
u. Ar
ithmetic of infrared im
a
ge en
hanc
ement ba
sed on the dis
perse of t
w
o-
dime
nsio
nal an
d w
a
ve
let
trans
form.
Comp
ute
r
Applic
ations
and Softw
are
. 200
6; 23(1
1
): 108-1
10.
[8]
Dua
n
Yon
g
-g
ang, Ma Li-
y
uan, Li Yo
ng-
jun,
W
ang T
i
an-h
u
i. Improv
ed soft-thresh
old d
enos
in
g
alg
o
rithm bas
e
d
on
w
a
v
e
l
e
t anal
ysis.
Sci
enc
e T
e
chno
lo
gy and En
gi
neer
in
g
. 2010; 1
0
(23)
: 5755-5
7
5
8
.
[9]
Jianz
hao
Hu
a
ng, Jia
n
Xi
e, Hon
g
cai
Li, G
u
i T
i
an, Xia
o
b
o
Ch
en. Se
lfa
dapt
iv
e Dec
o
mpositi
on L
e
v
e
l
Den
o
isi
ng Met
hod Bas
ed on
W
a
velet
T
r
ansform.
T
E
LKOMNIKA Indone
sian Jour
na
l of Electrica
l
Engi
neer
in
g
. 2012; 10(
5): 101
5-10
20.
[10]
Jian
g Xin
g
-gu
o
,
Mo Shao-n
i
ng
, Chu F
u
-ga
ng.
Me
thod of si
g
nal
w
a
vel
e
t rea
l
-time proc
essi
ng bas
ed
on
FPGA.
Applica
t
ion of Integrat
ed Circ
u
its
. 20
10; 36(1
2
): 45-
50.
[11]
T
i
an Bao-hu
a,
Li Ba
o-fen
g
. Paral
l
el
archi
t
ecture for lifti
ng-b
a
sed
2D
D
W
T
algorithm.
Journ
a
l of
Co
mp
uter Appl
icatio
ns
. 201
1; 31(1
2
): 336
6-3
370.
2
()
LR
Evaluation Warning : The document was created with Spire.PDF for Python.