TELKOM
NIKA
, Vol.14, No
.1, March 2
0
1
6
, pp. 195~2
0
2
ISSN: 1693-6
930,
accredited
A
by DIKTI, De
cree No: 58/DIK
T
I/Kep/2013
DOI
:
10.12928/TELKOMNIKA.v14i1.1831
195
Re
cei
v
ed Ap
ril 9, 2015; Re
vised Decem
ber
4, 201
5; Acce
pted De
cem
ber 2
2
, 2015
R-L-MS-L Filter Function for CT Image Reconstruction
Huiling Hou*
, Mingquan Wang, Xiaop
e
ng Wa
ng
1
Nation
al Ke
y
Lab
orator
y for Electron
ic Mea
s
ureme
n
t T
e
ch
nol
og
y,
Ke
y
Lab
orator
y of Instrumentation
Scie
nce a
nd D
y
n
a
mic M
easur
ement
of Ministr
y
of
Edu
c
ation,
North Un
iversit
y
of Ch
ina, T
a
iyu
an 0
3
0
051,
Shan
xi, Chin
a
*Corres
p
o
ndi
n
g
author, em
ail
:
hou_h
ui
lin
g@
126.com
A
b
st
r
a
ct
In X-ray c
o
mp
uter to
mo
gra
p
h
y (CT
)
, co
nvo
l
utio
n b
a
ck
pro
j
ectio
n
is
a
fun
d
a
m
e
n
tal
al
gor
ithm for
CT
i
m
a
ge r
e
co
nstruction.
As f
ilterin
g
pl
ays a
n
i
m
portant
pa
rt in c
onv
oluti
o
n b
a
ck
proj
ecti
on, th
e ch
oic
e
of
filter has a dire
ct impact up
on
t
he quality of reconstructe
d i
m
a
ges. Ai
m at impr
ovin
g rec
onstructed i
m
a
g
e
qua
lity, a new
mix
ed filter
ba
sed o
n
the i
d
e
a
of “fir
st w
e
ighted av
erag
e then l
i
n
ear
mix
i
ng
” is d
e
si
gne
d i
n
this artic
l
e, d
e
noted
by
R-L-
MS-L. Her
e
,
R-L filt
er
is
re
lied
o
n
to
g
u
a
r
antee
the
sp
atial
reso
luti
on
of
reconstructe
d i
m
a
ge
an
d S-L
filter is
proce
ssed vi
a
3-
poi
nt w
e
ighte
d
av
erag
ing
to i
m
p
r
ove the
de
nsi
t
y
resol
u
tion, thu
s
enha
nci
ng the over
all r
e
c
onstructio
n
qu
ality. Gaussia
n
noise
of different coefficie
n
t
s
is
add
ed to the
pr
ojecti
on d
a
ta to contrast the
nois
e
perfo
r
m
a
n
ce of the n
e
w
and trad
it
io
nal
mixe
d filters.
T
he
simulati
on
an
d
exper
iment res
u
lts show
that t
he n
e
w
f
ilter is
better in
anti-
n
o
ise
perfor
m
a
n
c
e an
d pr
oduc
es
reconstructe
d i
m
a
ges w
i
th not
ably i
m
prove
d
qua
lity.
Ke
y
w
ords
:
CT
ima
ge rec
onst
r
uction; conv
ol
uti
on b
a
ck pro
j
ection; filter fun
c
tion
Copy
right
©
2016 Un
ive
r
sita
s Ah
mad
Dah
l
an
. All rig
h
t
s r
ese
rved
.
1. Introduc
tion
Today, the
most
pop
ular CT
techniq
ue i
s
X-ray
CT. Analytical reconst
r
u
c
tion an
d
iterative re
co
nstru
c
tion
are
two b
a
si
c te
chni
que
s
of CT
im
age re
constructio
n
. Convol
ution back
proje
c
tion i
s
the key alg
o
rit
h
m for analyti
c
al re
co
nst
r
u
c
tion. It has widely appli
e
d in comm
ercial
CT, as it
s go
od computati
onal efficie
n
cy, good re
co
nstru
c
ted i
m
age qu
ality, and well pa
rallel
pro
c
e
ssi
ng a
c
hieva
b
le by mean
s of hardwa
r
e [1, 2].
Acco
rdi
ng to
CT
recon
s
truction
p
r
in
ciple,
ba
ck
projectio
n
i
s
in
natu
r
e
pain
t
ing th
e
proje
c
tion ta
ken from a fin
i
te object sp
ace eve
n
ly back (p
roje
ctin
g back) to e
a
ch p
o
int go
ne
throug
h
by
X-ray i
n
a
n
infinite spa
c
e
,
inclu
d
ing
p
o
ints
wh
ose
origin
al pixel
value
is ze
ro.
Therefore, i
m
age
s recon
s
tructe
d by
ba
ck proje
c
ti
on
sh
ow
obvio
us
aste
roid
traces. A filte
r
is
need
ed at th
e output term
inal in orde
r to elimi
nate th
e aste
roid tra
c
e
s
and
pro
d
u
ce hi
gh
-qu
a
lity
recon
s
tru
c
ted
images [3].
Comm
on filte
r
s in
clu
de
R-L filter and
S-L
filter, u
s
ing re
ctan
gul
ar wi
ndo
w a
nd si
n
c
w
i
n
d
o
w
r
e
s
pec
tive
ly. R
-
L filte
r
is
s
i
mp
le in
fo
r
m
,
e
a
s
y
to
us
e
,
pr
o
v
ide
s
c
l
ea
r
r
e
c
o
n
s
tr
uc
te
d imag
e
conto
u
rs, an
d
pro
d
u
c
e
s
a
high
spatial
resol
u
tion; ho
wever, it i
s
af
fected by
se
ri
ous
Gibb
s
effect
[4]. S-L filter
has
a lo
w o
scillation re
sp
o
n
se
and
displ
a
ys some
su
ppre
s
sion
effect on
noi
se
s but
its re
con
s
tru
c
tion quality is not as go
od a
s
R-L filter wh
en deali
ng wit
h
low-f
r
eq
uen
cy band [5].
In an
attemp
t to improve
image
re
co
n
s
tru
c
tion qual
ity,
schol
ars prop
osed so
me
ne
w
filters a
nd
so
me mixed filt
ers such a
s
R-L
an
d
S-L
mixed filter (denote
d
by
R-L
-
S-L),
R-L
and
NEW mixed filter
,
(denoted by
R-L-NE
W)[], etc.
Mixed filters give cons
i
d
eration to both spatial
resolution
an
d den
sity re
solution of reconstructe
d im
age
s an
d ha
ve a goo
d su
ppre
s
sion
effect
upon n
o
ises [
6
, 7].
On the
ba
sis of the oth
e
rs, a ne
w mixe
d filter
d
enot
ed by
R-L
-
M
S
-L i
s
p
r
opo
sed u
s
ing
the idea of
weighte
d
a
v
erage foll
o
w
ed by
line
a
r mixing.
After simulat
i
on analy
s
is and
experim
entati
on, the
ne
w filter i
s
p
r
o
v
ed to b
e
strongly
re
straining
o
scill
a
t
ion in i
m
ag
e
recon
s
tru
c
tio
n
. Beside
s, its ant-noi
se p
e
rform
a
n
c
e i
s
found imp
r
ov
ed. Than
ks to
the ne
w filter, a
good b
a
lan
c
e
is struck b
e
twee
n spatial
resolution
an
d den
sity re
solution for th
e image
with
a
better re
co
nst
r
ucte
d quality
.
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 : 195 – 2
0
2
196
2. Design Principle of Filter
Assu
ming the
proje
c
tion da
ta after filtering is as follo
ws:
,,
rr
r
px
p
x
h
x
(1)
Whe
r
e,
,
r
px
is the colle
cted p
r
ojectio
n
data,
r
hx
is the filtering function.
The id
eal filter fun
c
tion i
s
obtaine
d by
inve
rse Fo
uri
e
r tra
n
sfo
r
m,
and its expression i
s
[8]:
ex
p
2
rr
hx
i
x
d
(2)
Whe
r
e,
is th
e spatial f
r
eq
uen
cy. The
o
retically
spe
a
king, the f
r
eq
u
ency
re
spo
n
se of the
filter
shall
be
such
that
()
H
.The ide
a
l filter i
s
infi
nite-fre
que
ncy and
diverg
ent on
the i
n
finite
integral inte
rv
al.
2
2
Hd
d
(3)
In accordan
ce with Pal
e
y-Wie
n
e
r
theore
m
, su
ch
an ideal filter is un
ach
i
evable.
Ho
wever,
in
pra
c
tical
ima
g
ing
ope
ratio
n
, if the
se
nsor
ha
s a
sam
p
ling i
n
terval
small
eno
ugh
, the
high-f
r
eq
uen
cy comp
one
nt of sa
mpled
proje
c
t d
a
ta
woul
d be
min
i
mal, if not cl
ose
to zero. So,
the ideal filter may be pro
c
essed by wi
n
dowi
ng, as fol
l
ow:
()
(
)
HW
(4)
Whe
r
e,
()
W
being a wind
ow f
unctio
n
. Real
ization of
filtering is
essentially choosing
the wind
ow fu
nction
()
W
[9].
In ord
e
r to
ge
t the better re
con
s
tru
c
tion
i
m
age
re
soluti
on, the
choi
ce of
windo
w f
unctio
n
sho
u
ld abid
e
by certain p
r
i
n
cipl
es:
1) The
width
of the main lobe sh
ould b
e
narro
w, so a
s
to obtain a steep tran
sition
zone;
2) Relative to the main lob
e
, the maximum sid
e
lobe
sho
u
ld be a
s
small a
s
po
ssible, in
orde
r to improve the usu
a
l
smooth de
gree
and in
crea
se the sto
pba
nd attenuatio
n.
But gene
rally
spe
a
ki
ng, th
e win
d
o
w
fun
c
tion
whi
c
h h
a
s hi
gh a
nd
narro
w main
lobe, its
side lo
be i
s
a
l
so p
r
oje
c
ting
. Therefo
r
e, i
n
the actu
al, we
can n
o
t bl
indly req
u
ire
s
high resolution,
otherwise
it will cau
s
e
se
riou
s Gibb
s p
henom
eno
n.
In additio
n
, t
he
choi
ce
of
wind
ow fun
c
tion
depe
nd
s on
the interna
l
stru
cture
of wo
rk
pie
c
e and
a
c
tu
al internal
comp
one
nts and
recon
s
tru
c
tio
n
requi
rem
e
n
t
s. The most t
y
pical f
iltering
function
s are
R-L filter an
d
S-L filter.
3. Traditiona
l Mixed Filters
Referen
c
e [6
] prop
osed t
he R-L
-
S-L
filter
that
com
b
ine
s
the fe
a
t
ure of
R-L
and S-L
filters,
two commonly used
filt
ers. It improves spatial resolut
i
on
while
mi
nimizin
g
im
a
ge
oscillation. The impulse res
ponse (sam
pl
e
sequence) is:
12
2
12
22
2
22
2
12
22
2
2
2
2
(
)
()
()
8
0
4
2
(4
1
)
2
(4
1
)
RL
S
L
RL
S
L
hn
d
k
h
n
d
k
h
n
d
kk
n
d
k
n
dn
kk
n
nd
d
n
,
,
i
s
ev
en
,
is
o
d
d
(5)
Evaluation Warning : The document was created with Spire.PDF for Python.
TELKOM
NIKA
ISSN:
1693-6
930
R-L
-
MS-L Filter Fun
c
tion fo
r CT Im
age Reco
nstructio
n
(Huilin
g Ho
u)
197
Whe
r
e,
()
RL
hn
d
and
()
SL
hn
d
are
the
spati
a
l dom
ain ex
pre
ssi
on
of R-L a
nd S-L filters
r
e
spec
tively,
1
k
and
2
k
are weig
hted co
efficie
n
ts, with
12
1
kk
.
Referen
c
e [7]
prop
osed R-L-NE
W filter
by mi
xing R-L and
NEW fi
lter. It is expressed a
s
belo
w
:
12
12
22
2
22
2
12
22
2
2
2
2
()
(
)
()
,0
46
,
2
,
2
RLN
E
W
R
L
N
E
W
hn
d
k
h
n
d
k
h
n
d
kk
n
dd
k
n
dn
kk
n
dn
dn
i
s
ev
en
i
s
odd
(6)
Whe
r
e,
()
RL
hn
d
and
()
NE
W
hn
d
are the sp
atial domain
expre
ssi
on o
f
R-L and NEW
filters respect
i
vely,
1
k
and
2
k
are weig
hted coefficient
s, wi
th
12
1
kk
.
With re
spe
c
t to image re
co
nstru
c
tion, we usually u
s
e
two indice
s
,
spatial resolu
tion and
den
sity resol
u
tion
,
to eval
uate the reco
nstru
c
ted
ima
ge qu
ality. Here
,
spatial
resol
u
tion i
s
the
ability to identify
the small
e
st obje
c
t on
CT image
s.
Den
s
ity resol
u
tion refe
rs t
o
discrimin
a
ting
detecte
d obj
ects by mea
n
s of image
gray scale.
Gene
rally sp
eaki
ng, wh
en
the main lobe is
highe
r
and
n
a
rrower,
a b
e
tter spatial
re
solutio
n
can
be g
a
in. T
h
e
smalle
r th
e
si
de lo
be
and
t
h
e
quicke
r
the
converg
e
n
c
e, t
he hig
h
e
r
the
den
sity
re
sol
u
tion of the i
m
age
wo
uld
be. At a give
n set
of proj
ectio
n
data, the t
w
o in
dices
a
r
e
confli
cting
.
Table 1
gi
ves the
widt
h and
amplit
ude
(d=1mm
) of the main lo
be
and the first side lo
be ex
pre
s
sed by spatial dom
ain
curve
s
of th
e
above filterin
g function
s
after Fou
r
ier tr
ansfo
rmati
on. Both mixed filters ta
ke
1
0.5
k
and
2
0.5
k
.
Table 1. Amp
litude and
wid
t
h of the main lobe and the
first sid
e
lobe
Filter Main
lobe
height
Main lobe
wi
d
t
h
First side lobe
height
First side lobe
wi
d
t
h
R-L
0.25
1.42
-0.101
0.389
S-L 0.202
1.498
-0.067
0.232
R-L-S
-
L
0.226
1.532
-0.084
0.227
R-L-
NEW 0.187
1.456
-0.051
0.201
It is found from Ta
ble 1
that when t
he
mixed filters
give a b
e
tter and b
a
l
anced
con
s
id
eratio
n
to spatial
re
solutio
n
and
den
sity resol
u
tion of re
co
nstru
c
ted i
m
age
s compa
r
ed
with tradition
al R-L
and
S-L filters. T
he
R-L
-
NE
W mixed filter is th
e b
e
st
in table
1
. When
comp
ari
ng filt
ers in th
e
rem
a
inde
r of thi
s
article,
we
wil
l
com
p
a
r
e
a n
e
w mixe
d filte
r
with
the
R-L
-
Ne
w mixed filter.
3.1. A Ne
w
M
i
xed Filter
Whe
n
de
sig
n
i
ng a n
e
w fil
t
er, we
wo
ul
d attempt to
give better
con
s
id
eratio
n
to both
spatial a
nd d
ensity re
soluti
ons
so a
s
to
improve the o
v
erall re
co
nst
r
uctio
n
qualit
y.
Firstly, start
out with S-L f
ilter functio
n
to
desi
gn a n
e
w filter fun
c
tion usi
ng the
idea of
weig
hted av
erag
e. Its ba
sic
ratio
nale
is p
e
rf
o
r
min
g
wei
ghted
averag
e of t
he poi
nts in
the
neigh
borhoo
d
starting from null field filteri
ng functio
n
and
using
sign
al and sy
stem
viewpoi
nts
,
so as to sh
ort
en the main l
obe an
d dimi
nish the
side
lobe an
d imp
r
ove the den
sity
resolution
at
the expe
nse
of the
spatial
re
soluti
o
n
.
There i
s
a
certain
re
stri
ction relation
ship
betwe
en
spat
ial re
sol
u
tion
and d
e
n
s
ity resol
u
tion. in
view of the
specifi
c
situ
ation, choo
se t
h
e
right
prio
ritie
s
. In vie
w
of the
sp
ecifi
c
situat
io
n,
ch
oose the
ri
g
h
t pri
o
ritie
s
.
The
app
rop
r
i
a
te
empha
si
s will
be determi
ne
d In view of the sp
ecifi
c
circum
stan
ce
s [10, 11].
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 : 195 – 2
0
2
198
Cla
ssi
cal S-L filter has the f
o
llowin
g
sa
m
p
ling seque
n
c
e:
22
2
2
()
,
0
,
1
,
2
,
(4
1
)
SL
hn
d
n
dn
(7)
Theo
retically, a transl
a
tio
n
in the sp
atial domain,
when
rep
r
e
s
ente
d
in freque
ncy
domain, is m
u
ltiplication
b
y
an fluctuation facto
r
2/
(
2
1
)
ip
k
N
e
; and whe
n
there
are 2n
-1 poi
nts in
the spatial n
e
i
ghbo
rho
od fo
r weig
hted av
eragi
ng, it
is expre
s
sed in
freque
ncy do
main as b
e
lo
w:
2
(
)
(
1
)
(
)
e
xp(
)
21
n
k
kS
L
kn
ik
Hw
W
N
(8)
Whe
r
e,
N
i
s
the samplin
g
freq
uen
cy in
the n
u
ll fiel
d;
1
n
k
kn
W
is the
no
rmali
z
ed
expre
ssi
on.
k
W
may be modif
i
ed to suit sp
ecific
situatio
n.
By the rese
arch, averagin
g
has a
smo
o
thing e
ffec
t
on rec
o
ns
truc
t
ed images
, but more
than
3
-
poi
nt weig
hted ave
r
agin
g
woul
d
disto
r
t
an
d b
l
ur the
ed
ge
of re
con
s
tructed imag
es.
We
pro
c
e
s
s S-L
filter via 3-p
o
int wei
ghte
d
aver
aging,
denote
d
by
MS-L filter,
who
s
e
sam
p
ling
seq
uen
ce i
s
:
()
0
.
2
(
)
0
.
6
()
0
.
2
(
)
MS
L
S
L
S
L
S
L
hn
d
h
n
d
d
h
n
d
h
n
d
d
(9)
The co
rrespo
nding time do
main is repre
s
ente
d
as:
22
2
2
2
2
22
2
0.
4
1
.
2
0.
4
(
)
0,
1
,
2,
(4
(
1
)
1
)
(
4
1
)
(
4
(
1
)
1
)
MS
L
hn
d
n
dn
d
n
d
n
,
(10)
Figure 1
and
2 are the
di
screte di
stri
buti
on of
the
mai
n
lobe
an
d th
e far
sid
e
lo
b
e
of
R-L
and MS-L filtering fun
c
tion
s re
spe
c
tively.
Figure 1. Main lobe of R-L and MS-L in
spatial
domain
Figure 2. Far
side lo
be of R-L and MS
-L i
n
spatial domai
n
Figure 1 sh
o
w
s that the
main lobe of
R-L f
ilter i
s
hi
gher a
nd na
rrower, indi
cat
i
ng goo
d
spatial
re
solu
tion, but its
side l
obe i
s
l
a
rge
r
in
amp
litude an
d wi
dth, signifyin
g seve
re Gi
b
b
s
effect; while
MS-L filter fu
nction
produ
ce
s a
sh
ort
and
wide
ma
in lobe. Fi
gu
re 2
sh
ows t
hat
conve
r
ge
nce
effect of its f
a
r
side
lob
e
i
s
n
o
t very
well, indi
cating
poo
r
sp
atial
resolution
;
while
MS-L filter ha
s highe
r
rate
of
convergen
ce, whi
c
h hel
ps supp
re
ss
Gibb
s effect and noi
se
s. It ca
n
be obtaine
d by Figure 1 and Figu
re 2
that when
desig
ning a mi
xed filter, we may rely on
R-L
filter to guara
n
tee the spat
ial re
solution
of reco
n
s
tru
c
ted image a
n
d
use
MS-L f
ilter to impro
v
e
Evaluation Warning : The document was created with Spire.PDF for Python.
TELKOM
NIKA
ISSN:
1693-6
930
R-L
-
MS-L Filter Fun
c
tion fo
r CT Im
age Reco
nstructio
n
(Huilin
g Ho
u)
199
the den
sity resolutio
n
. The new mixed fil
t
er is den
ote
d
by R-L
-
MS-L. Its respo
n
se function is
as
follow:
12
()
()
(
)
RL
M
S
L
R
L
M
S
L
ht
k
h
t
k
h
t
(11)
Whe
r
e,
12
0,
0
kk
,
12
1
kk
.
The value of
1
k
and
2
k
is adju
s
t
able. Wh
en
1
0
k
, it
become
s
an
R-L filter; wh
en
2
0
k
it become
s
an MS-L filter. Its sampli
ng se
que
nce is:
12
22
2
1
2
22
2
14
0
41
5
()
()
()
RL
M
S
L
M
S
L
MS
L
kk
n
dd
hn
d
h
n
d
n
k
kh
n
d
n
nd
,
,
is
e
v
e
n
,
i
s
odd
(12)
4. Experimental Simulation and Re
su
lt Analy
s
is
Parallel be
a
m
recon
s
tru
c
tion is pe
rf
orme
d on 2
D
Sheep
-Lo
gan mod
e
l [5] using
respe
c
tively R-L filter, R-L-New mixe
d
filter, and t
he filter desig
ned in this a
r
ticle. In ord
e
r
to
evaluate re
constructe
d image qu
ality in the pr
esence of noises, Gau
s
sia
n
noise with
5%
intensity are
adde
d into the proje
c
tion d
a
ta, the mean
and varian
ce
being 0 an
d 1 respe
c
tively.
Figure 3 sh
ows the orig
inal Shepp
-L
ogan mo
del
and its gra
y
curve of line 128
.
Figure 4-6
sh
ow reconst
r
u
c
ted results o
f
R-L, R-
L-Ne
w, and R-L
-
M
S
-L filter and
their re
sp
ecti
ve
gray cu
rve of line 128.
(a) Mo
del
(b) G
r
ay curv
e of line 128
Figure 3. Orig
inal model
(a) Re
con
s
tru
c
ted
re
sult
(b) G
r
ay curv
e of line 128
Figure 4. Re
constructe
d
re
sult of R-L filter
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 : 195 – 2
0
2
200
(a) Re
con
s
tru
c
ted
re
sult
(b) G
r
ay curv
e of line 128
Figure 5. Re
constructe
d re
sult of R-L-New filter
(a) Re
con
s
tru
c
ted
re
sult
(b) G
r
ay curv
e of line 128
Figure 6. Re
constructe
d re
su
lt of R-L-M
S
-L filter
By compa
r
in
g Figure 3 to
Figure
6, we
find t
hat the new filter p
e
rf
orm
s
better t
han R-L
filter and R-L
-
Ne
w filter in
that
its reco
nstru
c
ted im
a
ge is not onl
y smoothe
r b
u
t also clo
s
e
r
to
the origin
al image.
For the sake
of evaluating recon
s
tru
c
t
ed im
age q
u
a
lity of differ
ent filters in a more
obje
c
tively manne
r, we u
s
e two evaluat
ion func
tio
n
s,
namely normalize
d
mea
n
squ
a
re di
st
ance
d and no
rmali
z
ed me
an ab
solute di
stan
ce
r [12]. They are expresse
d by:
1
2
2
,,
11
2
,
11
()
()
NN
uv
uv
uv
NN
uv
uv
tr
d
tt
(13)
,,
11
,
11
NN
uv
uv
uv
NN
uv
uv
tr
r
t
(14)
Whe
r
e,
v
u
t
,
and
v
u
r
,
are
re
spe
c
t
i
vely the test model a
nd t
he pixel d
e
n
s
ity of variou
s
r
e
co
ns
tr
uc
te
d p
o
i
n
t
s
.
t
is the avera
ge value of test
model d
e
n
s
ity.
d
is sen
s
iti
v
e to the gre
a
t
error
of a fe
w points,
as a
great
error
of
individual
poi
nts may le
ad
to a g
r
eat
d
value. Whe
r
ea
s
r
reveal
s se
nsi
t
ively the small errors of more p
o
in
ts.
Images
re
con
s
tru
c
ted by di
fferent filters
are
asse
ssed u
s
i
ng the
above
two di
stan
ce
indices.
The
asse
ssment
results
are
gi
ven in T
able
2.
Evaluation Warning : The document was created with Spire.PDF for Python.
TELKOM
NIKA
ISSN:
1693-6
930
R-L
-
MS-L Filter Fun
c
tion fo
r CT Im
age Reco
nstructio
n
(Huilin
g Ho
u)
201
The re
sult
s in
dicate that th
e new filter
p
e
rform
s
g
ene
rally better a
nd re
co
nstructs highe
r qu
al
ity
image un
de
r the sam
e
noi
se con
d
ition.
Table 2. No
rmalize
d
dista
n
ce
s of three
filters und
er t
he sam
e
noi
se intensity
F
ilter
5% intensit
y
10% intensit
y
d r d r
R-L
filter
0.4818
0.5992
0.8331
1.1182
R-L-
Ne
w
Filter
0.4191
0.4949
0.6496
0.8279
R-L-MS
-L
Filter
0.3738
0.4094
0.5125
0.6258
What
nee
ds i
llustratin
g
i
s
t
he
choi
ce
of
1
k
and
2
k
in Equation 11. As
12
1
kk
, we
study the inf
l
uen
ce of
1
k
to R-L-MS
-L
filter functi
on
by sele
cting differe
nt
1
k
values to
recons
truc
t. The rela
tion
sh
ip
be
tw
ee
n
1
k
and
no
rmali
z
ed
me
an
sq
uare
di
stan
ce is
sho
w
n
i
n
Figure 7. It c
an be se
en that the best value of
1
k
is
0.7 or so to be better inhib
i
ting oscill
atio
n
and noi
se.
Figure 7. The
relation
ship
betwe
en
1
k
and normali
ze
d mean squa
re
distan
ce
FDK re
co
nstruction i
s
pe
rforme
d on 36
0 pictu
r
e
s
of 1024*1
024
rocket moto
r mode
l
proje
c
tion d
a
t
a acqui
red b
y
the lab. Figure 8
sho
w
s the sin
ogra
m
of the
recons
truc
tion data.
Figure 9(a) a
nd 9(b) a
r
e t
he 150th
sli
c
e image
re
co
nstru
c
ted
by R-L
-
Ne
w filter and th
e ne
w
filter. It is rea
d
ily se
en th
at the ima
ge
re
con
s
tru
c
ted
b
y
R-L
-
MS-L fi
lter ha
s
cle
a
rer d
e
tails an
d
better imag
e quality.
Figure 8. The
sinog
ram of reco
nstructio
n
data
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 : 195 – 2
0
2
202
(a) R-L
-
Ne
w
(b) R-L
-
MS-
L
Figure 9. Co
mpari
s
o
n
of FDK usin
g different filters
5. Conclusio
n
In this article
,
we desig
n a new filter
named
R-L
-
MS-L filter, base
d
on the
idea of
weig
hted ave
r
agin
g
and li
n
ear mixing. E
x
perime
n
tati
o
n
and
simulat
i
on sh
ow that
R-L
-
MS-L filter
doe
s better i
n
image re
co
nstru
c
tion tha
n
traditional
filters, be
cau
s
e it not only strongly re
strai
n
s
the oscillation in image
recons
truction but al
so has better noise
immunity. The new fil
t
er
improve
s
the
image
den
si
ty resol
u
tion
signifi
cantly
while en
suri
n
g
the
spatial resolution of the
recon
s
tru
c
ted
image. G
o
o
d
re
con
s
truct
i
on re
sult
s a
r
e achieved
b
o
th in 2
D
CT
re
con
s
tru
c
ti
on
and 3
D
CT re
con
s
tru
c
tion.
Ackn
o
w
l
e
dg
ements
This
wo
rk i
s
supp
orted
by the Na
tional Natu
ral Scien
c
e
Found
ation
of chin
a
(No.6
117
117
7).
Referen
ces
[1]
Shi Ho
ngl
i, Lu
o Shuq
ia
n. A novel sch
em
e to
desig
n the filter for CT
reconstruction usi
ng F
B
P
algorithm.
Bio
M
edic
a
l Eng
i
n
eeri
ng Onli
ne
,
201
3; 12(1): 50
-65.
[2]
Yang
Me
ng, Y
an P
e
imi
n
, H
u
ang
Hu
i, L
i
Jif
e
i.
F
ilter
ed
bac
k pro
j
ectio
n
r
e
constructio
n
r
e
search
bas
ed
on Gaussi
an
in PET
ima
g
e
s
. 2012 Inter
natio
nal C
onf
erenc
e on Au
dio, La
ng
uag
e
and Image
Processi
ng, Procee
din
g
s. Sh
ang
hai. 2
012;
1: 360-3
64.
[3]
Bilg
ot
An
ne, D
e
sbat
La
urent, Perrier
Val
é
rie.
F
BP
and
the
i
n
terior
pro
b
l
e
m
in
2D
to
mo
grap
hy.
20
11
IEEE Nuclear
Scienc
e S
y
mposiu
m
and M
e
dical Imaging
Conf
er
ence(
NSS/MIC). Valencia.
20
12; 1:
408
0-40
85.
[4]
Ramac
han
dra
na GN, Laks
h
minara
y
a
n
an
AV, Kolaska
r
a
AS.
T
heor
y o
f
the non-
pl
an
ar pe
ptide
un
it.
Biochimica et
Biophysica Ac
t
a
(BBA) - Protein Structure
, 19
73, 303(
2): 385
-388.
[5]
Shee
p
L A,
Lo
gan
B F
.
T
he
F
ourier
reco
ns
truction
of a
h
ead
secti
on.
I
EEE trans.
Nucl. Sci
. 19
74
;
21(1): 21-
43
[6]
Z
hai J
i
ng,
Pan
Jin
x
i
ao,
Li
u B
i
n. Ap
plic
ati
on o
f
Mixe
d
Fil
t
er Fu
n
c
ti
on
in
FD
K Al
go
ri
th
m.
Journ
a
l of
Nanc
han
g Ha
n
g
kon
g
Univ
ersi
ty (Nature Scie
nce)
. 200
7, 21(
1): 263-2
66
[7]
Z
hang Bi
n. Pan Jin
x
ia
o. A ne
w
t
y
p
e
of
mixe
d filters for CT
image reconstructio
n
.
Mi
cro
c
om
pu
ter
Information
. 2
0
09; 25(3): 2
98-
308
[8]
Wei
Yu
Chuan
,
W
ang Ge,
H
s
ieh J
i
a
n
. An I
n
tuitive
Disc
u
s
s
ion
on
the I
d
e
a
l R
a
mp F
i
lter
in C
o
mp
uted
T
o
mograph
y.
Co
mp
uters an
d Mathe
m
atics
w
i
th Applicati
o
ns
. 2005; 4
9
(5-
6
)
:
73
1-74
0
.
[9]
Ren Z
h
o
ng,
Li
u Guodo
ng,
H
uan
g Z
hen.
Improve
m
ent of w
a
velet thresh
old filter
ed b
a
ck-projecti
o
n
imag
e reco
ns
truction al
gor
i
t
hm.
Proce
edi
ngs of SPIE-T
he Internationa
l Soci
et
y for Optical
Engi
neer
in
g. Beiji
ng. 20
14; 93
01: 1-10
[10]
Xi
e Hui, Du
an
W
anchun,
Su
n Yong
he, Du
Yuan
w
e
i. D
y
n
a
mic DEMAT
E
L grou
p decis
i
on ap
proac
h
base
d
o
n
i
n
tui
t
ionistic fuzz
y
numb
e
r.
Te
l
k
om
ni
ka
(Tel
e
c
omm
u
n
i
ca
tio
n
C
o
m
p
u
t
i
n
g El
ectro
n
i
cs and
Contro
l)
. 2014;
12(4): 106
4-1
072
[11]
Shi Be
n
y
i, W
a
ng C
h
e
ng, C
h
en Si
ha
i, Bi K
un. A N
o
vel
M
e
thod
of
CT
Reconstructi
on
F
ilter F
unctio
n
Desig
n
.
CT
T
heory an
d App
l
i
c
ations
. 20
10;
19(4): 35-
43
[12]
Ji Don
g
j
i
an
g, He W
enz
han
g.
T
he correcti
o
n SART
al
gorit
hm
bas
ed o
n
circular sy
mmetrical
obj
ect
.
200
9 Internati
o
nal C
onfere
n
ce
on Comp
utatio
nal In
tel
lig
enc
e
and Secur
i
t
y
.
Beiji
ng. 20
09;
1: 151-1
54.
Evaluation Warning : The document was created with Spire.PDF for Python.