Intern
ati
o
n
a
l
Journ
a
l of
Re
con
f
igur
able
and Embe
dded
Sys
t
ems
(I
JRES)
V
o
l.
3, N
o
. 3
,
N
o
v
e
m
b
er
2
014
, pp
. 10
4
~
11
3
I
S
SN
: 208
9-4
8
6
4
1
04
Jo
urn
a
l
h
o
me
pa
ge
: h
ttp
://iaesjo
u
r
na
l.com/
o
n
lin
e/ind
e
x.ph
p
/
IJRES
Bilinear Interpolatio
n Image Scaling Processor for VLSI
Architecure
Pa
wa
r A
s
hwini D
ilip*
,
K
Ra
meshbabu*
*, Kana
se Pra
j
akta Asho
k*
**
, Shita
l
A
r
jun
Shiv
d
a
s
*
***
*
KBPCE
,
Sa
ta
ra,
India
**JCEM, Shivaji University
, Kolhapur, M
a
harash
tra, India.
***RIT, Sakh
ar
ale, Ind
i
a
****ADCET, S
h
ivaji University, Kolhapur
, M
a
h
a
rashtra, Ind
i
a.
Article Info
A
B
STRAC
T
Article histo
r
y:
Received Apr 19, 2014
Rev
i
sed
Ju
l 17
,
20
14
Accepted Aug 12, 2014
We introduce
image scaling pr
ocessor us
ing VLSI techniqu
e. I
t
consist of
Biline
a
r in
terpo
l
ation
,
c
l
am
p fil
t
e
r and
a sharp
e
ning spati
a
l f
ilt
er. B
ilin
ear
interpolation alg
o
rithm is popular due
to its co
mputational
efficien
cy
and
image quality
.
But resultant image cons
ist of blurring edges
and aliasing
artif
acts
aft
e
r s
c
aling
.
To
redu
ce
the blu
rring
and
ali
a
s
i
ng ar
tifa
c
ts
s
h
arpenin
g
s
p
atial
filt
er and
clam
p fil
t
ers
ar
e us
ed as
pre-fi
lt
er. Th
es
e fi
lters
are re
ali
z
e
d
b
y
using T-model and inversed
T-mode
l convolution kernels. To reduce the
memory
buffer
and computing resour
ces for proposed image processor
design two T-model or inversed
T-model
filt
ers
a
r
e com
b
ined int
o
com
b
ined
filte
r which r
e
qu
ires onl
y on
e lin
e buffer m
e
m
o
r
y
.
Also, to
redu
ce ha
rdware
cost Reconfigu
r
able
cal
cula
tio
n unit (RCU) is invented
.
The VLSI
architecture in this work can achiev
e 280 MHz with 6.08-K gate counts, and
its cor
e
area is 3
0
378
μ
m2
sy
nt
he
si
z
e
d
by
a
0.
1
3
-
μ
m CMOS process.
Keyword:
Bilin
ear
Cla
m
p
filter
Reco
nfigu
r
ab
l
e
calcu
latio
n un
it
Sha
r
pe
n
Sp
atial filter
VLS
I
Copyright ©
201
4 Institut
e
o
f
Ad
vanced
Engin
eer
ing and S
c
i
e
nce.
All rights re
se
rve
d
.
Co
rresp
ond
i
ng
Autho
r
:
K R
a
m
e
shbab
u
,
JCEM, Sh
iv
aj
i
Un
i
v
ersity, M
h
.S.
Em
a
il: d
r
.krb.en
g
@g
m
a
i
l
.co
m
1.
INTRODUCTION
In
m
a
ny
appl
i
cat
i
ons,
f
r
om
co
nsum
er el
ect
ro
ni
cs
to m
e
dical im
aging im
ag
e
scaling algori
thm
can be
use
d
t
o
im
pro
v
e
t
h
e rest
ruct
ur
ed im
age qual
i
t
y
and pr
ocessi
ng
per
f
o
r
m
a
nce of ha
rd
wa
re i
m
pl
em
ent
a
t
i
on. For
ex
am
p
l
e i
m
ag
e scalin
g
is can
b
e
u
s
ed to
scal
e do
wn
the h
i
gh
-qu
a
lity p
i
ctu
r
es
o
r
v
i
d
e
o
fra
m
es to
fit th
e
min
i
si
ze l
i
qui
d cry
s
t
a
l
di
spl
a
y
panel
of t
h
e m
obi
l
e
pho
ne or t
a
bl
et
PC
or a vi
deo so
u
r
ce wi
t
h
a 64
0×4
8
0
vi
de
o
g
r
aph
i
cs array (VG
A
)
reso
l
u
tio
n
m
a
y n
eed
to
fit th
e
1
920×1
080
reso
l
u
tio
n
of a
h
i
gh
-defin
itio
n
m
u
ltimed
ia
in
ter
f
ace (H
DMI
)
.
I
m
ag
e scalin
g
algor
ith
m
h
a
s t
w
o m
a
in
t
y
p
e
po
lyno
m
i
a
l
b
a
sed
an
d non
p
o
l
yno
m
i
al b
a
sed
.
I
n
th
e m
o
st b
a
sic case, t
h
e fractio
n
a
l
p
a
r
t
of
an
y sub
p
i
x
e
l add
r
ess is trun
cated
or
r
ound
ed, so
each
p
i
x
e
l sim
p
ly
ta
k
e
s th
e
v
a
lu
e of th
e n
e
arest “real” p
i
x
e
l
in the source im
age. This
is called “nearest nei
ghbor”
app
r
oxi
m
a
t
i
on. Thi
s
m
e
t
hod i
s
t
h
e sim
p
l
e
st
, and i
n
v
o
l
v
es
n
o
cal
cul
a
t
i
o
n
.
I
t
al
so onl
y
re
q
u
i
r
es
one
pi
xel
from
t
h
e so
urce i
m
age f
o
r eac
h s
u
b pi
xel
w
h
i
c
h
i
s
bei
ng cal
c
u
lated; hence it
can operate
at the full s
p
eed
of the
su
rroun
d
i
ng
circu
it. Howev
e
r, th
e
resu
ltan
t
ap
pro
x
i
m
a
tio
n
is no
t
o
p
tim
al
with
th
is techniq
u
e
.
Bilin
ear in
terpo
l
atio
n m
e
th
o
d
h
a
s h
i
g
h
qu
ali
t
y an
d low com
p
lex
i
t
y
. By usin
g b
ilin
ear i
n
terp
o
l
ation
alg
o
rith
m
th
e targ
et p
i
x
e
l can b
e
ob
tain
ed
by u
s
in
g
th
e linear in
terp
o
l
atio
n
m
o
d
e
l in
bo
th
of th
e
h
o
rizo
n
t
al
and ve
rtical directions.
Bicu
b
i
c in
terpo
l
atio
n
is
often
ch
o
s
en
ov
er b
ilin
ear in
terpo
l
atio
n or
nearest n
e
i
g
hbor in
im
ag
e
resam
p
ling, when s
p
ee
d is not an issue
.
It is anot
her
popular
m
e
thod use
s
an extende
d
cubic m
odel to acquire
t
h
e t
a
r
g
et
pi
xel
by
a
2
-
D
re
gul
ar
gri
d
.
In
ou
r
prev
i
o
us wo
rk
, an
ad
ap
tiv
e real-ti
m
e
,
low-
co
st, an
d h
i
gh
-qu
a
lity i
m
ag
e scalar
was propo
sed.
It successfully i
m
proves the image qua
lity by adding sha
r
pening spatial a
nd clam
p filters as pre filters
with
Evaluation Warning : The document was created with Spire.PDF for Python.
I
S
SN
:
2
089
-48
64
I
J
RES Vo
l. 3
,
N
o
. 3
,
No
v
e
m
b
er
201
4
:
1
04
–
11
3
10
5
an
ad
ap
tiv
e tech
n
i
q
u
e
b
a
sed
on
th
e b
ilin
ear i
n
terp
o
l
ation
alg
o
rith
m
.
Alth
ou
gh
th
e h
a
rdware co
st and
me
m
o
ry
requ
irem
en
t h
a
d
b
een efficiently red
u
c
ed
, t
h
e d
e
m
a
n
d
o
f
m
e
m
o
ry still co
sts fou
r
lin
e
b
u
ffers. Hen
ce, a
lo
w-
co
st an
d low me
m
o
ry-requ
iremen
t i
m
ag
e scalar d
e
si
g
n
is
propo
sed in
t
h
is
b
r
ief.
Table 1 com
p
ares the com
puting
resources a
nd m
e
m
o
ry access per pixel of four well-known
in
terpo
l
atio
n
al
g
o
rith
m
s
. Th
e b
ilin
ear in
terpo
l
atio
n
algo
rit
h
m
d
e
m
a
n
d
s
l
o
w co
m
p
u
tin
g
resource an
d
me
m
o
ry
access per
pixel. Howeve
r, i
t
causes the edge
s of the
sc
aled im
ages to becom
e
blurred and aliased after
in
terpo
l
atio
n. Kim
et al.
pres
ented a
n
a
r
ea-pixel m
odel ca
lled the “
W
i
n
s
cale” algorithm
.
It uses a maxim
u
m
of f
o
ur
pi
xel
s
of t
h
e
ori
g
i
n
al
im
age t
o
cal
cul
a
t
e
one
pi
xel
of a scal
ed i
m
age. In a
ddi
t
i
on,
An
d
r
eadi
s
et.al.
Propose
d a modi
fied
W
i
n sc
ale algorithm
that uses a m
a
s
k
of no m
o
re than
four
pixel
s
and calculate
s the
final lum
i
nosity of eac
h pi
xel to scale grey
-s
cale and c
o
lor im
ages.
This m
e
thod offe
rs
better quality than t
h
e
W
i
n
scale algorithm
.
Howeve
r, it re
quires m
o
re
com
put
i
n
g
reso
u
r
ces t
h
a
n
doe
s t
h
e
W
i
n s
cal
e al
go
ri
t
h
m
.
Tabl
e
1. C
o
m
p
ari
s
o
n
of
C
o
m
put
i
n
g R
e
s
o
urc
e
& M
e
m
o
ry
A
ccess/
Pi
xel
of
Fo
ur
I
n
t
e
r
pol
at
i
on
Al
g
o
r
i
t
h
m
s
Bilinear
Winscale
M-
Winscale
Bicubic
Co
m
puting
Resour
ces
7 Mult
7 add
10 M
u
lt
11 add
13 M
u
lt
20 add
32 M
u
lt
53 add
Me
m
o
r
y
access
4 read
1 write
4 read
1 write
4 read
1 write
4 read
16 wr
ite
Th
is p
a
p
e
r is o
r
g
a
n
i
zed
as fo
llo
ws. In
Sectio
n
II, th
e
b
ilin
ear in
terpo
l
atio
n
,
clam
p
filter, and
sh
arpen
i
ng
sp
atial filter are briefly in
tro
d
u
ced.
Fi
gu
re
1.
B
l
oc
k
di
ag
ram
of i
m
age scal
i
n
g
a
l
go
ri
t
h
m
1
.
1
.
Bilinea
r Interpo
l
atio
n
In
b
ilin
ear in
t
e
rpo
l
atio
n, th
e v
a
lu
e of a sub
-
p
i
x
e
l
is i
n
terp
o
l
ated fro
m
its fou
r
n
e
arest
n
e
ighb
ours
lin
early. Th
e
h
o
rizon
t
al fractio
n
a
l co
m
p
on
en
t
(of th
e
su
b-p
i
x
e
l co
ord
i
n
a
te) is
u
s
ed
to
co
m
p
u
t
e t
w
o
interpolated
points that still lie on
t
h
e horizontal gri
d
, t
h
en the
vertic
al
fractional c
o
m
pone
nt is used to
interpolate
between these
two
points.
Figure 2 s
h
ows
how this
is
done for
a
n
im
age. In each case, the line
a
r
in
terpo
l
atio
n between
t
w
o
v
a
lu
es a and
b
by fraction
k
is
co
m
p
u
t
ed
as (b
- a)
X
k
+ a. Bilin
ear in
terp
o
l
ation
o
f
fers sign
ifican
tly en
h
a
n
c
ed
i
m
ag
e qu
ality ov
er n
e
arest
n
e
ig
hbo
r app
r
ox
imatio
n
.
Bilin
ear
interpo
l
atio
n
is
an
im
ag
e-restorin
g
al
g
o
rith
m
,
wh
ich
lin
early
in
terpo
l
ates fo
ur
n
earest
-
n
e
i
g
hbo
r
pi
xel
s
o
f
a
n
u
n
rest
ore
d
i
m
age t
o
o
b
t
a
i
n
t
h
e
pi
xel
o
f
a
re
s
t
ore
d
i
m
age as a f
o
rwa
r
d
f
u
n
c
t
i
on.
The
pri
n
ci
pl
e
b
e
h
i
n
d
th
e
b
ilin
ear in
terpo
l
atio
n
algo
rit
h
m is ex
ecu
tin
g
a lin
ear in
terpo
l
atio
n
i
n
o
n
e
d
i
rection
,
and
th
en
rep
eating
th
e
sam
e
fu
n
c
tion in
th
e o
t
h
e
r
d
i
rection
.
As
sh
own
in
Figure 2
,
P(i, j
)
, P(i+1
,
j
)
, P(i,j
+
1), and
P(i
+
1
,
j+
1) a
r
e
t
h
e fo
u
r
nea
r
es
t
nei
g
h
b
o
r
pi
xe
l
s
of t
h
e o
r
i
g
i
n
al
im
age wi
t
h
i = [0,
1, 2
,
. . .
M
]
and j = [
0
,
1, 2
,
.
. .
N]
.
Here,
M
i
s
t
h
e
num
ber
of
pi
xel
s
ha
vi
n
g
t
h
e
wi
dt
h
of
t
h
e o
r
i
g
i
n
al
i
m
age a
n
d
N
i
s
t
h
e
num
ber
of
pi
xel
s
co
rresp
ond
ing
to
th
e leng
th
of th
e
ori
g
inal i
m
age.The tem
p
orary
pixels
P
(
x
’
,j
) a
nd
P
(
x
’
,j
+1) are
creat
ed by
lin
ear in
terpo
l
atio
n
in horizontal d
i
rection and ca
n
be calc
u
lated as
P
(
x
’
, j
)= (
1
−
xf
) ×
P
(
i, j
)+
xf
×
P
(
i
+1
, j
)
(
1
)
P
(
x
’
, j
+1) =
(1
−
xf
) ×
P
(
i, j
+1) +
xf
×
P
(
i
+1
, j
+
1
)
(
2
)
whe
r
e
xf
is th
e scale p
a
ram
e
ter in
ho
rizo
n
t
al d
i
rectio
n. Afte
r in
terpo
l
atin
g
i
n
ho
rizo
n
t
al d
i
rectio
n, th
e v
a
lu
es
of t
e
m
pora
r
y
p
i
xel
s
P
(
x’
,
j
) a
n
d
P
(
x’,j
+1
) are
g
e
n
e
rated. Th
e resu
lting
o
u
t
pu
t p
i
x
e
l
P
(
x’
,
y
’
)
can be obtained
by
Evaluation Warning : The document was created with Spire.PDF for Python.
I
J
RES
I
S
SN
:
208
8-8
7
0
8
Bilin
ea
r In
terpo
l
a
tion
Imag
e S
c
a
ling
Pro
c
esso
r
f
o
r
VLSI
Arch
itecu
re
(K Ra
m
e
shb
abu
)
10
6
o
n
e
m
o
re lin
ear in
terp
o
l
atio
n
in
t
h
e o
t
h
e
r
d
i
rection
.
Altern
ativ
ely, th
e ou
tpu
t
can
b
e
p
r
o
duced
by
im
pl
em
ent
i
ng l
i
n
ear i
n
t
e
r
pol
at
i
on i
n
t
h
e
v
e
r
tical direction a
n
d can be
calc
u
lated as
P(
x’
,
y
’)
= [(
1
−
x
f
)×P(i
,
j)+xf×P(i+1,j)]×(1
−
y
f)+
[(
1
−
xf
)
×
P(
i,j
+
1)
+x
f
×
P(
i+1,j
+
1)
]×yf
(
3
)
whe
r
e
yf
is th
e scale p
a
ram
e
te
r in
v
e
rtical d
i
rectio
n
.
Bilin
ear in
terp
o
l
ation
is p
o
p
u
l
ar in
the i
m
p
l
e
m
en
tat
i
o
n
of
VLSI ch
ip
s
du
e to
its lo
w co
m
p
lex
ity
a
n
d
sim
p
le ar
chitecture. Howeve
r,
its
h
i
gh
-f
r
e
qu
en
cy r
e
sp
on
se
beha
vi
o
r
i
s
po
or as a resul
t
o
f
l
i
n
ear cha
nge
s t
o
t
h
e out
p
u
t
pi
xel
val
u
e acc
or
di
n
g
t
o
sam
p
l
i
ng p
o
si
t
i
on.
R
e
sul
t
s
sho
w
t
h
at
t
h
e
edge
s
becom
e
bl
ur
ry
an
d t
h
e al
i
a
si
ng
ef
fe
ct
i
s
vi
si
bl
e
af
t
e
r bei
n
g
pr
oc
essed
usi
n
g
bi
l
i
n
ear
in
terpo
l
atio
n.
Fig
u
re
2
.
Bilin
ear in
terpo
l
ation
g
r
i
d
Gi
ve
n a p
o
i
n
t
(x
,y
),
We i
n
t
e
r
pol
at
e bet
w
ee
n
fl
oo
r (
x
) an
d
cei
l
(
x)
on eac
h
of t
w
o
ro
ws
f
l
oo
r(y
) a
n
d
ceil(y)u
s
ing
fractio
n
a
l p
a
rt
of x
to
ob
tain
t
w
o
in
term
ed
i
a
te poi
nt
s s
h
o
w
n
.t
hen w
e
use f
r
act
i
onal
pa
rt
of y
t
o
in
terpo
l
ate b
e
t
w
een th
ese two
p
o
i
n
t
s to ob
t
a
in
fi
n
a
l in
terpo
l
ated
v
a
lu
e
By
(1),
we can
easily fin
d
th
at th
e co
m
p
utin
g
reso
urces
o
f
th
e
b
ilin
ear co
st eig
h
t
m
u
ltip
ly, fo
ur
su
b
t
ract, and
t
h
ree add
itio
n
op
eration
s
.
It costs a co
nsid
erab
le ch
ip area t
o
im
p
l
e
m
en
t a
b
ilin
ear in
terpo
l
ato
r
with
ei
g
h
t
m
u
ltip
lies an
d seven
ad
ders. Thu
s
, an alg
e
b
r
ai
c m
a
n
i
p
u
l
atio
n sk
ill h
a
s
b
een u
s
ed
t
o
redu
ce th
e
co
m
p
u
tin
g
reso
urces o
f
t
h
e bilin
ear
in
terp
o
l
atio
n
.
Th
e o
r
i
g
in
al equ
a
tio
n
of b
ilin
ear in
terp
o
l
ation
is
p
r
esen
ted
in
(1), and
si
m
p
l
i
fied
p
r
o
c
ed
ure of b
ilin
ear in
terpo
l
ation
can
b
e
d
e
scrib
e
d
fro
m
(2
)-(3
), on
e o
f
t
h
e two
cal
cul
a
t
i
ons
f
o
r t
h
i
s
al
geb
r
ai
c
f
unct
i
o
n ca
n
b
e
re
duce
d
P
(
x
’
,y
’)
=[(
1
−
yf
)×
P
(
i
+1
,j
)+
yf
×
P
(
i
+1
,j
+1
)
]
xf
+
[(
1
−
yf
)×
P
(
i,
j
)+
yf
×
P
(
i,j
+1)
](1
−
xf
)
(
4
)
=[P(i+
1,j)+yf P(i+1,
j+
1)
-P
(i+1,
j
)]
x
f
+[
P
(
i,j
)+
yf
×
P
(
i,j
+1)-
P
(
i,j
)]
(1
−
xf
)
(
5
)
Low-C
o
m
p
lex
ity Sh
arp
e
n
i
ng
Sp
atial and
Clam
p
Filters
Th
e sh
arp
e
n
i
ng
sp
atial filter, a k
i
nd
of h
i
g
h
-p
ass
filter,
is u
s
ed
to
redu
ce b
l
u
r
ring
artifacts an
d
d
e
fi
n
e
d
b
y
a
ker
n
el
t
o
i
n
c
r
e
a
se t
h
e i
n
t
e
nsi
t
y
of
a ce
nt
er
pi
xel
rel
a
t
i
v
e t
o
i
t
s nei
g
h
b
o
r
i
n
g
pi
xel
s
.
It
us
ual
l
y
cont
ai
ns
a si
ngl
e
p
o
s
itiv
e
v
a
lu
e
at its cen
ter and
co
m
p
letely s
u
rroun
d
e
d
b
y
n
e
g
a
tiv
e
v
a
lu
es. Th
e
fo
llowing
array is an
ex
am
p
l
e
o
f
a
3
×
3
k
e
rnel for a
sh
arp
e
n
i
ng
sp
atial filter
Whe
r
e
S
is a sharp param
e
ter that can be se
t according to
the cha
r
acteristics of the
im
ages. The clam
p filter
[19
]
, [20
]
, is a k
i
n
d
o
f
low-pass filter, is u
s
ed
to
sm
o
o
t
h
un
wan
t
ed
d
i
scon
tin
uou
s edg
e
s o
f
bo
und
ary reg
i
on
s
and re
duce aliasing effects.
It
can
be
rep
r
ese
n
t
e
d by
a
c
o
nv
ol
ut
i
o
n ker
n
el
.
Evaluation Warning : The document was created with Spire.PDF for Python.
I
S
SN
:
2
089
-48
64
I
J
RES Vo
l. 3
,
N
o
. 3
,
No
v
e
m
b
er
201
4
:
1
04
–
11
3
10
7
Here
,
C
is a
clam
p param
e
ter that can be
set a
cco
rd
ing
t
o
th
e ch
aracteristics of th
e im
ag
es.
(a)
(b
)
(c)
Fi
gu
re 3.
W
e
i
g
ht
s of
t
h
e
c
o
n
v
o
l
u
t
i
o
n ker
n
el
s
.
(a)
3 ×
3
co
n
v
o
l
ut
i
on
ke
rnel
.
(
b
)
C
r
oss-m
ode
l
con
v
o
l
u
t
i
o
n
k
e
rnel
.
(c)
T-m
odel
a
n
d i
n
ve
rsed
T-m
odel
c
o
nv
ol
ut
i
o
n
ke
r
n
el
s.
Th
is k
e
rn
el is co
m
b
in
ed
with
m
a
trix
co
efficien
ts th
at sh
ow th
e d
e
p
e
n
d
e
n
c
e o
f
a filtered
p
i
x
e
l on
its
n
e
igh
bors.
A l
a
rg
er size
o
f
co
nvo
lu
tion
k
e
rn
el will pro
d
u
c
e h
i
gh
er
q
u
a
lity o
f
im
ag
es. Howev
e
r, a larg
er size
o
f
conv
o
l
u
tio
n
filter will
also
d
e
m
a
n
d
m
o
re me
m
o
ry
an
d hardware
co
st. To
red
u
c
e
th
e co
m
p
lex
ity
o
f
th
e 3
×
3 c
o
nv
ol
ut
i
o
n
ker
n
el
, a
cr
oss
-
m
odel
fo
rm
ed i
s
use
d
t
o
repl
a
ce t
h
e
3
×
3
co
nv
ol
ut
i
o
n
ke
rn
el
, as s
h
ow
n
i
n
Fi
g
.
2(b). It succes
sfully cuts down
on
fo
ur
of
nine param
e
ters in the 3
×
3 con
v
o
l
u
t
i
o
n ke
rnel
. F
u
rt
herm
ore
,
t
o
decrease m
o
re
com
p
l
e
xi
t
y
and m
e
m
o
ry
requi
rem
e
nt
of t
h
e cr
oss-m
ode
l
con
vol
ut
i
on
ker
n
el
, T-m
o
d
e
l
and
in
v
e
rsed
T-m
o
d
e
l conv
o
l
u
tion
k
e
rn
els are
p
r
op
o
s
ed
fo
r realizin
g
th
e sh
arp
e
n
i
ng
sp
atial filter an
d
cla
m
p
filters.
Co
m
b
in
ed
Filter
In
p
r
op
osed
scalin
g
algo
rithm to
redu
ce mo
re co
m
p
u
tin
g resource and
me
m
o
ry requ
ire
m
en
t sp
atial
an
d clam
p
filter wh
ich fo
rmed
b
y
T-m
o
d
e
l and
In
v
e
rs
ed
T-m
o
d
e
l
sh
ou
l
d
b
e
com
b
in
ed
to
g
e
t
h
er in
t
o
co
m
b
in
ed
filter. Th
e inp
u
t
imag
e is filtered
b
y
sh
arp
e
n
i
ng
sp
atial filter and
th
en
filtered
b
y
cla
m
p
filter ag
ain
.
Both filters re
qui
re two line
buffe
rs to store input data
or interm
ediate value for
each
T
m
odel or inversed T
m
odel
.
Du
e to
t
h
is th
ese two
filters co
m
b
in
ed
tog
e
t
h
er in
to co
m
b
in
ed filter as
Evaluation Warning : The document was created with Spire.PDF for Python.
I
J
RES
I
S
SN
:
208
8-8
7
0
8
Bilin
ea
r In
terpo
l
a
tion
Imag
e S
c
a
ling
Pro
c
esso
r
f
o
r
VLSI
Arch
itecu
re
(K Ra
m
e
shb
abu
)
10
8
Wh
ere S an
d
C
are th
e sh
arp
an
d
clam
p
p
a
rameters an
d
P’(i
,j) is th
e filtered
resu
lt o
f
th
e t
a
rg
et p
i
x
e
l P(i,j
)
b
y
the c
o
m
b
ined
filter. A T-m
o
del sha
r
pe
ning
spatial filter a
n
d a
T-m
odel cl
a
m
p filter
ha
ve bee
n
re
placed
by
a
co
m
b
in
ed
T-mo
d
e
l
filter as sh
own
in Figu
re 3
To
re
d
u
c
e
th
e on
e-lin
e-b
u
ffer m
e
m
o
ry, th
e
o
n
l
y
p
a
ram
e
ter in
th
e th
ird
lin
e, p
a
ram
e
ter
−
1 of P(i
,
j-
2
)
, i
s
rem
oved an
d
t
h
e wei
ght
o
f
param
e
t
e
r
−
1
is ad
d
e
d
in
to
the
p
a
ram
e
ter S-C o
f
P(i,j-1
)
b
y
S-C-1
as sho
w
n
in
(2). Th
e com
b
in
ed
in
v
e
rsed
T-m
o
d
e
l filter can
b
e
p
r
od
uced
in
the sam
e
way.
The dem
a
nd
of
m
e
m
o
ry can be efficiently
de
creased
fr
om
two t
o
o
n
e l
i
n
e
bu
ffe
r by
usi
n
g t
h
i
s
filter-co
m
b
i
n
a
tio
n
techn
i
qu
e. In
th
is two
T-m
o
d
e
l or
i
n
v
e
rsed
T-m
o
d
e
l filters are co
m
b
in
ed in
t
o
o
n
e
com
b
ined T-m
odel
or inverse
d
T-m
odel filter. It gr
eatly reduce
s
m
e
m
o
r
y
access require
m
ents for software
syste
m
s or
hardwa
re m
e
m
o
ry co
st
s f
o
r VL
SI
im
pl
em
ent
a
t
i
on.
1.
2. VL
SI
Arc
h
i
t
ecture
Fo
r
VLSI i
m
p
l
e
m
en
tatio
n
,
the b
ilin
ear in
terp
o
l
ator
can
d
i
rectly o
b
t
ain
two
in
pu
t p
i
x
e
ls
P’(i,j), and
P’(i,j
+1
) fro
m
two
co
m
b
in
ed
p
r
efilters wi
t
h
ou
t an
y add
itio
nal lin
e-bu
ffer
me
m
o
ry.
Fi
gu
re
4.
B
l
oc
k
di
ag
ram
of t
h
e VL
SI a
r
c
h
i
t
ect
ur
e for proposed real-tim
e
image scaling proces
sor
Fi
gu
re 4 s
h
ow
s t
h
e bl
oc
k di
a
g
ram
of t
h
e V
L
SI arc
h
i
t
ect
ur
e fo
r t
h
e p
r
o
p
o
se
d desi
gn
. It
consi
s
t
s
o
f
fou
r
m
a
in
b
l
o
c
k
s
: a reg
i
ster
ban
k
, a co
m
b
in
ed
filter,
a b
ilin
ear in
terpo
l
ato
r
, and
a con
t
ro
ller. Th
e d
e
tails o
f
each part will
be descri
bed
in
the following sections.
Th
e co
m
b
in
ed filter is filterin
g
t
o
p
r
o
d
u
c
e th
e targ
et
p
i
xels
o
f
P’(i,j), an
d
P’(i,j+1) b
y
u
s
i
n
g
ten
sou
r
ce pi
xel
s
. The re
gi
st
er ba
nk i
s
desi
gne
d wi
t
h
a one
-l
i
n
e
m
e
m
o
ry
buffe
r, w
h
i
c
h i
s
use
d
t
o
pr
o
v
i
d
e t
h
e t
e
n
v
a
lu
es fo
r th
e i
mmed
i
ate u
s
age of th
e co
m
b
in
ed filter.
Evaluation Warning : The document was created with Spire.PDF for Python.
I
S
SN
:
2
089
-48
64
I
J
RES Vo
l. 3
,
N
o
. 3
,
No
v
e
m
b
er
201
4
:
1
04
–
11
3
10
9
Figure. 5.
Arc
h
itecture of t
h
e register
bank
Figure
5 shows the arc
h
itecture of the
regist
er ba
nk
with
a stru
cture o
f
ten
sh
ift
reg
i
sters. W
h
en
t
h
e
sh
ifting
co
n
t
rol sig
n
a
l is
p
r
odu
ced fro
m
th
e
co
n
t
ro
ller, a
new
v
a
lu
e
of P(i+3
,
j)
will b
e
read
in
t
o
Reg
4
1
,
and
each value stored in othe
r
registers belonging to row
n
+ 1 will b
e
sh
ifted
rig
h
t
in
to
th
e n
e
x
t
reg
i
ster or lin
e-
bu
ffe
r m
e
m
o
ry
. The
Reg
4
0
r
eads a
ne
w
va
lue o
f
P
(
i+2
,
j) fro
m
th
e lin
e-b
u
ffer m
e
m
o
ry, and eac
h
val
u
e in
o
t
h
e
r
reg
i
sters b
e
lon
g
i
n
g
to
row n
will
b
e
sh
i
f
ted
righ
t
in
t
o
t
h
e n
e
x
t
reg
i
ster.
Fig
u
re
6
.
Co
mp
u
t
ation
a
l sch
e
d
u
ling
of th
e propo
sed
co
m
b
in
ed filter and
si
m
p
lified
b
ilinear in
terpo
l
ato
r
It
sh
ort
e
n
s
t
h
e
del
a
y
pat
h
t
o
im
pro
v
e t
h
e p
e
rf
orm
a
nce by
pi
pel
i
n
e t
ech
n
o
l
o
gy
. I
n
fi
g
u
r
e 6 st
ages
1
an
d
2
sho
w
the co
m
p
u
t
atio
nal sch
e
du
ling
o
f
a
T-m
o
d
e
l
co
m
b
in
ed
an
d an
inv
e
rsed T-m
o
d
e
l filter.
Th
e T-
m
odel
or i
nve
r
s
ed T
-
m
odel
fi
l
t
e
r co
nsi
s
t
s
o
f
t
h
ree
reco
nfigu
r
ab
le calcu
lat
i
o
n
un
its (RC
U
s),
o
n
e
m
u
lti
p
lier–
adde
r (M
A
)
, th
ree adde
rs (+
),
three su
btracte
r
s (
−
), an
d thre
e shifters (S
).
The
ha
rdware architecture
of the
T-
m
o
d
e
l co
m
b
in
ed
filter can
b
e
d
i
rectly
m
a
p
p
e
d
with
th
e
con
v
o
l
u
tion
eq
u
a
tio
n
shown
i
n
(2
). Th
e
v
a
lu
es
o
f
the
t
e
n s
o
u
r
ce
pi
xe
l
s
can
be
o
b
t
a
i
n
ed
f
r
om
t
h
e re
gi
st
er
ban
k
.
The sy
m
m
et
ri
cal
ci
rcui
t
,
as
s
h
o
w
n i
n
st
ages
1 a
n
d
2
of
Fi
g
u
re
6,
i
s
t
h
e
i
n
verse
d
T-m
ode
l
com
b
i
n
ed
filter d
e
si
g
n
e
d fo
r
produ
cing th
e
filtered
resu
lt of
P’(i,j
+1
) Th
e arch
itectu
r
e
o
f
th
is sy
mme
trical circu
it is a
si
m
ilar sy
mme
trical stru
cture
o
f
t
h
e T-m
o
d
e
l
co
m
b
in
ed
filte
r, as sho
w
n
in
stag
es 1 and
2
o
f
Figu
re
6
.
B
o
th
o
f
th
e co
m
b
in
ed filter an
d
symmetrical circu
it co
n
s
ist
o
f
on
e
MA and
t
h
ree
RCUs. Th
e M
A
can
b
e
im
p
l
e
m
en
ted
b
y
a
m
u
ltip
lier
an
d
an
add
e
r. Th
e RCU is d
e
sig
n
e
d
fo
r
produ
cing
th
e calculatio
n
fun
c
tions o
f
(S-C) and
(S-C
-
1)
t
i
m
e
s of t
h
e
sou
r
ce
pi
xel
va
l
u
e,
whi
c
h m
u
st
be i
m
pl
em
ented
wi
t
h
C
a
n
d
S pa
ram
e
t
e
rs.
Evaluation Warning : The document was created with Spire.PDF for Python.
I
J
RES
I
S
SN
:
208
8-8
7
0
8
Bilin
ea
r In
terpo
l
a
tion
Imag
e S
c
a
ling
Pro
c
esso
r
f
o
r
VLSI
Arch
itecu
re
(K Ra
m
e
shb
abu
)
11
0
Figure 7.
Archi
t
ecture of
t
h
e RCU
Th
e arch
itecture of th
e
p
r
opo
sed
low-co
st
co
m
b
in
ed
filter can
filter th
e who
l
e im
ag
e with
on
ly a
one
-line-buffer m
e
m
o
ry, whic
h s
u
ccess
fully decrease
s
the
me
m
o
ry require
m
ent from
four to
one line
buffer
o
f
t
h
e co
m
b
in
ed
filter i
n
o
u
r
p
r
ev
iou
s
wo
rk
[1
]. Tab
l
e
2
lists th
e p
a
ram
e
t
e
rs an
d co
m
p
utin
g
resou
r
ce fo
r t
h
e
R
C
U
.
W
i
t
h
t
h
e
sel
ect
ed C
and S val
u
es l
i
s
t
e
d i
n
Tabl
e 2, t
h
e gai
n
of t
h
e
cl
am
p or shar
p
con
vol
ut
i
on f
unct
i
o
n
i
s
{8,
1
6
,
3
2
}
o
r
{
4
,
8,
1
6
},
w
h
i
c
h ca
n
be
el
i
m
i
n
at
ed by
a
s
h
i
f
t
e
r
rat
h
e
r
t
h
an a
di
vi
der
.
Tabl
e
2.
Param
e
t
e
rs an
d c
o
m
put
i
n
g
res
o
urce
fo
r R
C
U
Para
m
e
ter
Values
Co
m
puting Resour
ces
s
5,
13,
29
Add and Shift
c
7,
11,
19
Add and Shift
s-c
2,
-
6
,
-
22,
6, -
2
, -
1
8,
14,
6,
-
10
Add,
Shift and sign
s-c-1
1,
-
7
,
-
23,
5, -
3
, -
1
9,
13,
5,
-
11
Add,
Shift and sign
Fig
u
re
7
sho
w
s th
e arch
itectu
r
e
of th
e RCU.
It con
s
ists
o
f
fo
ur
sh
ifters,
three
m
u
ltip
lex
e
rs (MUX),
th
ree
ad
d
e
rs, and
one sig
n
circu
it.
By th
is RCU d
e
sig
n
, th
e h
a
rd
ware co
st of the co
m
b
in
ed
filters can
b
e
efficien
tly
red
u
ce
d.
Th
e b
ilin
ear interpo
l
atio
n
is si
m
p
lified
as sho
w
n
in
(3). The stag
es 3
,
4
,
5
,
and
6
in
Figu
re
6
show
th
e fo
ur-stag
e
p
i
p
e
lin
ed
arch
i
t
ectu
r
e, an
d
t
w
o
-
stag
e p
i
p
e
lined
m
u
ltip
liers are u
s
ed
to
shorten
th
e
d
e
lay p
a
th
o
f
th
e b
ilin
ear in
t
e
rpo
l
ato
r
. Th
e
in
pu
t v
a
l
u
es
o
f
P’(i,j) a
n
d
P’(i,j
+1) are ob
tai
n
ed fro
m
th
e co
m
b
in
ed
filter
an
d
sym
m
et
ri
cal
circui
t
.
B
y
t
h
e hard
ware s
h
ari
n
g t
echni
que
, as sho
w
n i
n
(
3
) The co
nt
r
o
l
l
e
r i
s
im
pl
em
ent
e
d by
a
fin
ite-state m
a
ch
in
e circu
it. It
p
r
o
d
u
ces con
t
ro
l sign
als
to
co
n
t
ro
l th
e tim
i
n
g
and
p
i
p
e
line stag
es of th
e
reg
i
ster
b
a
nk
, co
m
b
in
ed
filter, and
b
i
l
i
n
ear i
n
terp
o
l
at
o
r
.
2.
SIMULATION RESULTS AND
CHIP IMPLEME
N
T
A
TION
To
b
e
ab
le to
an
alyze th
e q
u
a
l
ities o
f
th
e scal
ed
i
m
ag
es b
y
v
a
riou
s scalin
g
alg
o
rith
m
s
, a p
eak
sig
n
a
l
-
t
o
-
noi
se
rat
i
o
(
PSNR
)
i
s
use
d
t
o
q
u
ant
i
f
y
a
noi
sy
ap
p
r
o
x
i
m
at
i
on of t
h
e r
e
fi
ne
d an
d t
h
e
ori
g
i
n
al
i
m
ages. Si
nce
the m
a
xim
u
m
value
of eac
h
pixel is 255, t
h
e
PSNR e
x
pres
s
e
d in dB
can be calculated as
whe
r
e M
an
d
N are t
h
e wi
dt
h a
nd
hei
g
h
t
of t
h
e o
r
i
g
i
n
al im
age. Four
well-known scaling algorith
m
s
,
W
i
nscale (Wi
n
), m
odified W
i
nscale (M
−
W
i
n
)
, bi
c
ubi
c
(B
C
)
, an
d bi
l
i
n
ear (B
L)
, and eac
h st
ep
of t
h
e
propose
d
a
d
a
p
tive scaling algorithm
are used to analyze t
h
e
qualities of
each al
gorithm
by using t
h
e eight
testin
g
im
ag
es.
Evaluation Warning : The document was created with Spire.PDF for Python.
I
S
SN
:
2
089
-48
64
I
J
RES Vo
l. 3
,
N
o
. 3
,
No
v
e
m
b
er
201
4
:
1
04
–
11
3
11
1
Fig
u
re
7
.
Eigh
t sam
p
le i
m
ag
e
s
in
t
h
e test set. (a)
Len
a
. (b) Pep
p
e
rs.
(c) Airp
la
n
e
. (d
) Mand
rill. (e)
Girl.
(f
)
Sailb
oat. (g
)
S
p
lash
. (h
) H
ous
e.
In
th
e
q
u
a
lity ev
alu
a
tion
proced
ure, each
test i
m
ag
e sh
ou
ld
b
e
filtered b
y
a fix
e
d
low p
a
ss
filter
(av
e
rag
i
n
g
filter) and
th
en
scaled
up
/do
w
n to
d
i
fferen
t
sizes su
ch
as
25
6
×
256
(hal
f size), 352
×
28
8
com
m
on interm
ediate form
at (CIF),
6
40
×
4
80 vi
de
o gra
p
hi
cs
array
(V
GA
), 72
0
×
48
0 (
D
1),
1
0
2
4
×
1024
(d
o
ubl
e si
ze)
, a
nd
1
9
8
0
×
1080 hi
gh-definiti
on m
u
lti
m
e
di
a
interface
(HDMI) as listed
in Table 3. The average
PSNR
o
f
th
e
bilin
ear in
terpo
l
atio
n
or th
is wo
rk
is 28
.1
5
o
r
2
8
.
54
,
wh
ich
mean
s th
at th
e co
m
b
in
ed
T-m
o
d
e
l
an
d inv
e
rsed
T-m
o
d
e
l filters i
m
p
r
o
v
e
t
h
e imag
e
q
u
a
lity
b
y
0
.
3
9
d
B
. As listed
in Tab
l
e
4, th
e m
u
ltip
licatio
n
o
p
e
ration
s
o
f
are 32
wh
ich
is
eig
h
t
tim
es th
e
q
u
a
n
tity o
f
t
h
is wo
rk
, and
the m
e
m
o
ry req
u
irem
en
t o
f
or
is six
or
f
o
u
r
l
i
nes
w
h
i
c
h
i
s
si
x
or
f
o
u
r
t
i
m
es t
h
e a
m
ount
o
f
t
h
e
o
n
e-l
i
n
e
b
u
ffe
r
m
e
m
o
ry
i
n
t
h
i
s
w
o
r
k
.
Ta
bl
e
4 l
i
s
t
s
the com
puting res
o
urce a
n
d
me
m
o
ry re
qu
ire
m
en
t o
f
t
h
e
fiv
e
prev
i
o
u
s
l
o
w-co
m
p
lex
ity
scalin
g
al
g
o
ri
th
ms
with
th
is wo
rk
. Accord
i
n
g
to
Tab
l
e 4
,
t
h
is wo
rk
n
e
ed
s
o
n
l
y
fou
r
m
u
ltip
lic
atio
n
op
erations, wh
ich
is m
u
ch
less
th
an
7
,
10
, 32
,
o
r
13
i
n
b
ilin
ear ,
W
i
n
, BC ,
or Ed
g
e
-Orien
ted
, resp
ectiv
ely
.
The
VLS
I
a
r
c
h
i
t
ect
ure
o
f
t
h
i
s
w
o
r
k
was
i
m
pl
em
ent
e
d by
usi
n
g t
h
e
har
d
ware
desc
ri
pt
i
o
n
l
a
n
gua
ge
Veri
l
o
g.
The e
l
ect
roni
c
desi
g
n
a
u
t
o
m
a
t
i
on t
ool
Desi
g
n
Vi
s
i
on
has
bee
n
u
s
ed t
o
sy
nt
he
si
ze t
h
e V
L
SI
C
i
rcui
t
base
d
on
Tai
w
an Sem
i
cond
uc
t
o
r M
a
nu
fact
u
r
i
ng C
o
m
p
any
0.
18
-
μ
m
an
d
0
.
13
-
μ
m
proc
ess st
an
dar
d
ce
l
l
s
.
Table
3. C
o
m
p
arisons
of ave
r
age PS
NR
fo
r
Vari
o
u
s
scaling
algorith
m
s
BL
Win
BC
E
dge
Ada
Th
is Wo
rk
A
B
C
Half
27.
04
27.
30
28.
5
27.
42
28.
49
27.
68
27.
77
28.
27
CIF
27.
8
27.
75
28.
7
27.
82
29.
45
27.
68
27.
83
28.
27
VGA
28.
5
28.
55
28.
7
28.
58
29.
44
28.
39
28.
42
28.
22
DI
28.
5
28.
51
28.
92
28.
55
29.
44
28.
54
28.
45
28.
61
DOU
28.
5
28.
52
28.
92
28.
58
29.
37
28.
56
28.
8
28.
78
HDM
I
28.
56
26.
94
28.
97
26.
96
29.
38
27.
87
28.
66
28.
76
AVE
28.
15
27.
93
28.
96
27.
99
29.
2
28.
29
28.
32
28.
54
A: Using Sharp fil
t
er and bilinear
interpolation, B: Using cla
m
p filter an
d bilin
ear interpol
ation, C: Using com
b
ined filter an
d bilinear
interpolation
Tabl
e
4. C
o
m
p
ari
s
o
n
of
com
put
i
n
g
res
o
urce
and
m
e
m
o
ry
requi
rem
e
nt
Multiplication
Addition
M
e
m
o
ry
Buffe
r
BL
7
7
4 lines
Win
10
11
4 lines
Pr
eviouswor
k
3
50
4 lines
Th
is Wo
rk
4
36
1 line
Evaluation Warning : The document was created with Spire.PDF for Python.
I
J
RES
I
S
SN
:
208
8-8
7
0
8
Bilin
ea
r In
terpo
l
a
tion
Imag
e S
c
a
ling
Pro
c
esso
r
f
o
r
VLSI
Arch
itecu
re
(K Ra
m
e
shb
abu
)
11
2
Th
e layou
t for th
e p
r
op
osed
d
e
sign
was g
e
n
e
rated
with
IC Co
m
p
iler. Th
e ch
i
p
pho
tomicro
g
r
aph
is
illu
strated
in
Fig
u
re 7. Fu
rth
e
rm
o
r
e, th
e p
r
o
p
o
s
ed
d
e
sig
n
was ev
alu
a
ted
and
verified
b
y
an field
pr
o
g
ram
m
abl
e
gat
e
ar
ray
(F
PG
A) em
ul
at
ion
b
o
ar
d
w
ith an
Altera
FPGA
EP2
C
70
F8
96
C
6
core. Th
is work
cont
ai
n
s
o
n
l
y
6.
08
-K
gat
e
c
o
unt
s
,
an
d t
h
e c
h
i
p
a
r
ea i
s
30
37
8
μ
m
2
syn
t
hesized
b
y
a 0.13
-
μ
m
CMOS proces
s.
More
ove
r, t
h
is work can
proces
s the
whole im
ag
e with
on
ly a on
e-lin
e-b
u
ffer me
m
o
ry. Th
e
po
wer
con
s
um
pt
i
on
o
f
t
h
e p
r
op
ose
d
desi
g
n
wa
s m
e
asure
d
by
usi
n
g S
Y
N
O
PS
YS
Pri
m
e Power.
It
co
nsum
es 6.
9 m
W
at
a 280-M
H
z
operat
i
o
n f
r
eq
uency
wi
t
h
a 1.
1-
V su
ppl
y
vol
t
a
ge
. Fu
rt
h
e
rm
ore, t
h
e t
h
ro
u
g
h
p
u
t
of t
h
i
s
wo
r
k
achi
e
ves 2
8
0
m
e
gapi
xel
s
pe
r
secon
d
. It
i
s
fast
enou
g
h
t
o
achi
e
ve t
h
e de
m
a
nd of real
-t
i
m
e graphi
c an
d vi
d
e
o
ap
p
lication
s
with
a HDM
I o
f
WQSXGA
(320
0
×
204
8)
r
e
so
lu
tion
at
30
fra
m
e
s p
e
r
second
.
Fi
gu
re 7.
C
h
i
p
ph
ot
om
i
c
rog
r
a
p
h
3.
CO
NCL
USI
O
N
In th
is
p
a
p
e
r
efficien
t and
l
o
w m
e
m
o
ry VLSI arch
itectu
r
e
o
f
b
ilin
ear
in
terpo
l
ator and
co
m
b
in
ed
filter was presen
ted
fo
r im
ag
e scalin
g
app
licatio
n
.
Th
is
meth
od
con
s
ists o
f
three step
s
su
ch
as filtering
of
i
m
ag
es u
s
i
n
g cla
m
p
and
sp
atial 2D filter,
b
ilin
ear i
n
te
rp
o
l
at
io
n and
VLSI i
m
p
l
e
m
en
tatio
n
.
Th
e co
m
p
u
t
atio
n
a
l
co
m
p
lex
ity o
f
fun
c
tio
n
is
d
ecreased
b
y
co
m
b
in
ed
filter an
d
al
g
e
braic
m
a
n
i
p
u
l
at
io
n
of th
e
b
ilin
ear
in
terpo
l
atio
n.
Co
m
p
arin
g
with
o
t
h
e
r low com
p
lex
i
t
y
ar
chitecture this
work achie
ve
s at least 34.5%
reduction
in
g
a
te cou
n
t
s
an
d requ
ires
on
ly on
e-lin
e me
m
o
ry bu
ffer.
REFERE
NC
ES
[1]
S.L. Chen
, H.
Y. Huang, and
C.H. Luo
,
“
A
low-cost high-
qualit
y
adap
tive
scalar for
rea
l
-tim
e m
u
ltim
edi
a
applications”,
IEEE Trans. Circu
its
Syst. Video
Technol.
, vol. 21
,
no. 11
, pp
. 1600
–1611, Nov. 201
1.
[2]
F. Cardells-
T
or
mo and J. Arnabat-Be
n
e
dicto, “
F
lexible h
a
rdware-friend
l
y
di
gital ar
chitectur
e f
o
r 2-D separable
convolution-b
a
s
e
d scaling”,
IEEE Trans. Circuits
Syst.
II,
Exp
.
B
r
iefs
, vo
l. 53, no. 7,
pp. 522–526,
Jul. 2006.
[3]
S.
Ridella, S.
Rovetta,
and R.
Zu
nino,
“IAVQ-interval-arithmetic
vector
quantization for
image compression”,
IE
EE
T
r
ans
. Cir
c
uits
S
y
s
t
. II
,
A
nalog
Digit. Signa
l
Process.
, vo
l. 47, no.
12,
pp
. 1378–13
90, Dec. 2000.
[4]
H. Kim, Y. Cha, and S. Ki
m, “Curvature in
terp
olati
on method f
o
r image zooming”,
IEEE Trans. Image Process.
,
vol. 20
, no
. 7
,
pp
. 1895–1903
, Jul. 2011
[5]
J.W. Han, J.H.
Ki
m, S.H. Cheo
n, J.O. Kim,
an
d S.J. Ko, “Anovel imag
e interp
ol
ation
method
using the b
ilater
a
l
filte
r”,
IEEE Trans. Consum. Electron.
, vol. 56
, n
o
. 1
,
pp
. 175–18
1, Feb
.
2010
.
[6]
X. Zhang and X.Wu, “Image interpolation b
y
ad
aptiv
e 2-
D autoregressive modeling a
nd soft-decision estimation
”
,
IEEE Trans. Image Process.
, vol. 17, no. 6, pp.
88
7–896, Jun. 200
8.
[7]
K. Jensen and D. Anastassiou, “
S
ubpixel edg
e
lo
cal
ization and
th
e int
e
rpolation o
f
still im
ag
es”,
I
EEE T
r
ans. Image
Proc
e
ss.
, vol. 4,
no. 3
,
pp
. 285–2
95, Mar
.
1995
.
[8]
C.H. Kim, S.M. Se
ong, J.A. Lee, and L.S.
Kim, “Winscale : An imagescaling
al
g
o
rithm using an area p
i
xel model”,
IEEE Trans. Circuits
S
y
st. Vid
e
o
Technol.
, vol. 1
3
, no
. 6
,
pp
. 549
–553, Jun. 2003.
[9]
P
.
Y. Chen
, C
.
Y.
Li
en,
and C
.
P
.
Lu,
“
V
LS
I im
plem
enta
tion of
an edge oriented
image s
caling
processor”,
IEEE
Trans. Very Large Sca
l
e Integr
.
(
V
LSI
)
Syst.
, vol. 17, no. 9, pp. 12
75–1284, Sep
.
2
009.
[10]
C.
C.
Lin,
M
.
H.
Sheu,
H.
K.
Chiang,
W.
K.
Tsai, and Z.
C.
W
u
,
“
R
eal-tim
e F
P
GA archit
ec
ture o
f
extend
ed l
i
ne
a
r
convolution
for
digital
image scaling”, in
Proc. I
EEE Int. Conf.
Field-
Program. Technol.
, 2008, p
p
. 381–384
.
Evaluation Warning : The document was created with Spire.PDF for Python.
I
S
SN
:
2
089
-48
64
I
J
RES Vo
l. 3
,
N
o
. 3
,
No
v
e
m
b
er
201
4
:
1
04
–
11
3
11
3
BIOGRAP
HI
ES OF
AUTH
ORS
m
s
. Pawar Ashwini Dilip, M.
E., (
E
le
ctron
i
cs
).,
KBPCE, Sat
a
ra. v
e
r
y
m
u
ch interest
ed in
res
earch
,
tea
c
hi
ng. favor
ite
s
ubjec
ts
are
im
age
proces
s
i
ng, v
l
s
i
,
m
i
croproces
s
o
r
s
etc
.
s
h
e
can
reach
a
t:
ashupa
war2412@gm
ail.com
Dr. K. Rameshbabu, professor
& Dean (academic
s
)
, E & TCE
D
e
pt, J
C
EM
, k
a
rad M
h
.S
. he
is
holding B.E (ece), M.Tech
,
Ph.D having 17+
years
of experien
ce in e
l
ectron
i
cs
and
Tel
ecom
m
unicat
ion Engine
erin
g area
.he
is
m
e
m
b
er in IS
TE, IE
EE &
J
a
va Cert
ifie
d
programmer (2,0) PGDST holder. h
e
h
a
s lot of
e
xper
i
enc
e
in
a
cadem
ics and
in
dustrial rel
a
ted
real
tim
e proj
ects
.
He is pap
e
r set
t
e
r for m
a
n
y
auto
nom
ous universities and
also visi
ting professor
for image processing, electron
d
e
vices & commu
nications
etc.
Ms. Kanase Prajakta As
hok, B.E (E&TC), RIT, working
as Asst. prof, JCEM
, interested in
res
earcg
, adm
i
ni
s
t
ration
.
Making good f
r
iends, doing
service
to
the n
a
tion is my
habb
ies.
Shital Arjun Shivdas, B.E from
Annasaheb Dan
g
e college of
En
gg. Ashta. Batch
-
2011-12, ME
parsueing in VLSI & Embedded s
y
stem fron ADCET.
Interested in Lab oper
a
tions & research
work.
Evaluation Warning : The document was created with Spire.PDF for Python.