T
E
L
KO
M
N
I
KA
T
e
lec
om
m
u
n
icat
ion
,
Com
p
u
t
i
n
g,
E
lec
t
r
on
ics
an
d
Cont
r
ol
Vol.
18
,
No.
1
,
F
e
br
ua
r
y
2020
,
pp.
394
~
406
I
S
S
N:
1693
-
6930,
a
c
c
r
e
dit
e
d
F
ir
s
t
G
r
a
de
by
Ke
me
nr
is
tekdikti
,
De
c
r
e
e
No:
21/E
/KP
T
/2018
DO
I
:
10.
12928/
T
E
L
KO
M
NI
KA
.
v18i1.
12837
394
Jou
r
n
al
h
omepage
:
ht
tp:
//
jour
nal.
uad
.
ac
.
id/
index
.
php/T
E
L
K
OM
N
I
K
A
Hal
f
t
on
in
g
-
b
ase
d
B
T
C
i
m
age
r
e
c
o
n
st
r
u
c
t
io
n
u
si
n
g
p
at
c
h
p
r
oc
e
ssi
n
g w
it
h
b
o
r
d
e
r
c
on
s
t
r
ai
n
t
Her
i
P
r
as
e
t
yo
1
,
Chi
h
-
Hs
ien
Hs
ia
2
,
B
e
r
t
on
Ar
ie
P
u
t
r
a
Akardi
h
as
3
1,
3
D
e
p
art
me
n
t
o
f
I
n
fo
rma
t
i
c
s
,
U
n
i
v
er
s
i
t
as
Seb
e
l
a
s
Maret
(U
N
S),
In
d
o
n
es
i
a
2
D
ep
ar
t
men
t
o
f
Co
m
p
u
t
er
Sci
en
ce
a
n
d
In
fo
rma
t
i
o
n
E
n
g
i
n
eeri
n
g
,
N
a
t
i
o
n
a
l
Il
an
U
n
i
v
er
s
i
t
y
,
T
a
i
w
a
n
Ar
t
icle
I
n
f
o
AB
S
T
RA
CT
A
r
ti
c
le
h
is
tor
y
:
R
e
c
e
ived
Apr
4
,
2019
R
e
vis
e
d
Nov
19
,
20
19
Ac
c
e
pted
Nov
30
,
20
19
T
h
i
s
p
a
p
er
p
re
s
en
t
s
a
n
ew
h
al
f
t
o
n
i
n
g
-
b
as
e
d
b
l
o
c
k
t
r
u
n
c
at
i
o
n
c
o
d
i
n
g
(H
B
T
C)
i
mag
e
reco
n
s
t
ru
ct
i
o
n
u
s
i
n
g
s
p
ars
e
re
p
res
e
n
t
a
t
i
o
n
frame
w
o
r
k
.
T
h
e
H
BT
C
i
s
a
s
i
mp
l
e
y
et
p
o
w
erf
u
l
i
ma
g
e
co
mp
re
s
s
i
o
n
t
ech
n
i
q
u
e,
w
h
i
c
h
can
effect
i
v
e
l
y
remo
v
e
t
h
e
t
y
p
i
ca
l
b
l
o
c
k
i
n
g
effec
t
an
d
fal
s
e
co
n
t
o
u
r.
T
w
o
t
y
p
e
s
o
f
H
BT
C
met
h
o
d
s
are
d
i
s
cu
s
s
e
d
i
n
t
h
i
s
p
ap
er,
i
.
e.
,
or
d
ered
d
i
t
h
er
b
l
o
ck
t
ru
n
ca
t
i
o
n
co
d
i
n
g
(O
D
B
T
C)
an
d
erro
r
d
i
ff
u
s
i
o
n
b
l
o
c
k
t
r
u
n
ca
t
i
o
n
co
d
i
n
g
(E
D
BT
C).
T
h
e
p
ro
p
o
s
ed
s
p
ar
s
i
t
y
-
b
as
e
d
me
t
h
o
d
s
u
p
p
re
s
s
e
s
t
h
e
i
mp
u
l
s
i
v
e
n
o
i
s
e
o
n
O
D
B
T
C
an
d
E
D
B
T
C
d
eco
d
e
d
i
mag
e
w
i
t
h
a
co
u
p
l
e
d
d
i
c
t
i
o
n
ar
y
co
n
t
a
i
n
i
n
g
t
h
e
H
BT
C
i
ma
g
e
co
mp
o
n
e
n
t
an
d
t
h
e
cl
ean
i
mag
e
co
mp
o
n
en
t
d
i
ct
i
o
n
ari
e
s
.
H
erei
n
,
a
s
p
ars
e
co
effi
c
i
en
t
i
s
es
t
i
ma
t
ed
fro
m
t
h
e
H
BT
C
d
eco
d
e
d
i
mag
e
b
y
mean
s
o
f
t
h
e
H
BT
C
i
mag
e
d
i
c
t
i
o
n
ar
y
.
T
h
e
rec
o
n
s
t
ru
ct
e
d
i
ma
g
e
i
s
s
u
b
s
e
q
u
e
n
t
l
y
b
u
i
l
t
an
d
al
i
g
n
e
d
fro
m
t
h
e
cl
ea
n
,
i
.
e.
n
o
n
-
co
m
p
res
s
ed
i
ma
g
e
d
i
c
t
i
o
n
ar
y
an
d
p
red
i
ct
e
d
s
p
ars
e
c
o
effi
c
i
en
t
.
T
o
f
u
rt
h
er
red
u
ce
t
h
e
b
l
o
c
k
i
n
g
effect
,
t
h
e
i
ma
g
e
p
at
c
h
i
s
fi
rs
t
l
y
i
d
e
n
t
i
fi
e
d
as
“b
o
rd
er”
a
n
d
“n
o
n
-
b
o
r
d
er”
t
y
p
e
b
efo
re
a
p
p
l
y
i
n
g
t
h
e
s
p
ars
e
re
p
res
e
n
t
a
t
i
o
n
fr
ame
w
o
r
k
.
A
d
d
i
n
g
t
h
e
L
ap
l
aci
a
n
p
ri
o
r
k
n
o
w
l
ed
g
e
o
n
H
B
T
C
d
ec
o
d
e
d
i
ma
g
e,
i
t
y
i
e
l
d
s
b
e
t
t
e
r
reco
n
s
t
r
u
ct
e
d
i
ma
g
e
q
u
a
l
i
t
y
.
T
h
e
e
x
p
er
i
men
t
al
re
s
u
l
t
s
d
em
o
n
s
t
ra
t
e
t
h
e
effect
i
v
en
e
s
s
o
f
t
h
e
p
ro
p
o
s
ed
H
B
T
C
i
ma
g
e
rec
o
n
s
t
r
u
ct
i
o
n
.
T
h
e
p
r
o
p
o
s
e
d
met
h
o
d
a
l
s
o
o
u
t
p
erf
o
rms
t
h
e
f
o
rmer
s
c
h
emes
i
n
t
erm
s
o
f
rec
o
n
s
t
r
u
ct
e
d
i
mag
e
q
u
a
l
i
t
y
.
K
e
y
w
o
r
d
s
:
E
r
r
o
r
dif
f
us
ion
H
a
lf
toni
ng
-
B
T
C
O
r
de
r
dit
he
r
ing
S
pa
r
s
e
r
e
pr
e
s
e
ntation
Th
i
s
i
s
a
n
o
p
en
a
c
ces
s
a
r
t
i
c
l
e
u
n
d
e
r
t
h
e
CC
B
Y
-
SA
l
i
ce
n
s
e
.
C
or
r
e
s
pon
din
g
A
u
th
or
:
He
r
i
P
r
a
s
e
tyo,
D
e
pa
r
t
ment
of
I
nf
or
mat
ics
,
Unive
r
s
it
a
s
S
e
be
las
M
a
r
e
t
(
UN
S
)
,
S
ur
a
ka
r
ta,
I
ndone
s
ia
.
E
mail:
he
r
i
.
pr
a
s
e
tyo@s
taf
f
.
uns
.
a
c
.
id
1.
I
NT
RODU
C
T
I
ON
B
lock
tr
unc
a
ti
on
c
oding
(
B
T
C
)
a
nd
it
s
va
r
iants
ha
ve
be
e
n
playing
a
n
im
p
or
tant
r
ole
on
i
mage
pr
oc
e
s
s
ing
a
nd
c
omput
e
r
vis
ion
a
ppli
c
a
ti
ons
,
s
uc
h
a
s
im
a
ge
/vi
de
o
c
ompr
e
s
s
ion
[1
-
3]
,
im
a
ge
wa
ter
mar
king
[
4
,
5]
,
da
ta
hidi
ng
[
3,
6]
,
im
a
ge
r
e
t
r
ie
va
l
a
nd
c
las
s
if
ica
ti
on
[7
-
10]
,
im
a
ge
r
e
s
tor
a
ti
on
,
[
11
-
13]
e
tc.
M
a
ny
e
f
f
or
ts
ha
ve
be
e
n
f
oc
us
e
d
on
f
ur
ther
im
p
r
oving
the
pe
r
f
o
r
manc
e
of
B
T
C
a
nd
it
s
va
r
iants
,
i
nc
ludi
ng
the
c
omput
a
ti
ona
l
c
ompl
e
xit
y
r
e
duc
ti
on
,
de
c
ode
d
i
mage
qua
li
ty
i
mpr
ove
ment,
a
nd
it
s
a
ppli
c
a
ti
ons
,
a
s
r
e
por
te
d
in
[
1
,
2,
7
-
9,
12,
13]
.
T
he
B
T
C
-
ba
s
e
d
im
a
ge
c
om
pr
e
s
s
ion
f
inds
a
ne
w
r
e
pr
e
s
e
ntation
of
a
n
im
a
ge
t
o
f
u
r
ther
r
e
duc
e
the
s
tor
a
ge
r
e
quir
e
ment,
a
nd
a
c
hieve
a
s
a
ti
s
f
a
c
tor
y
c
oding
ga
in.
I
t
is
c
las
s
if
ied
a
s
a
los
s
y
im
a
ge
c
ompr
e
s
s
ion,
in
whic
h
a
given
im
a
ge
block
is
p
r
oc
e
s
s
e
d
to
yield
a
ne
w
r
e
pr
e
s
e
ntation
c
ons
is
ti
ng
of
t
wo
c
olor
qua
nti
z
e
r
s
a
nd
the
c
or
r
e
s
ponding
bit
map
im
a
ge
.
T
he
two
c
olor
qua
nti
z
e
r
s
a
nd
bit
map
im
a
ge
pr
o
duc
e
d
a
t
the
e
nc
oding
s
tage
a
r
e
then
tr
a
ns
mi
tt
e
d
to
t
he
de
c
ode
r
.
T
he
typi
c
a
l
B
T
C
tec
hniques
de
ter
mi
ne
the
t
wo
c
olor
qua
nti
z
e
r
s
,
na
mely
low
a
nd
high
mea
ns
,
by
mainta
ini
ng
the
int
r
ins
ic
s
tatis
ti
c
a
l
p
r
ope
r
ti
e
s
o
f
a
n
i
mage
s
uc
h
a
s
Evaluation Warning : The document was created with Spire.PDF for Python.
T
E
L
KO
M
NI
KA
T
e
lec
omm
un
C
omput
E
l
C
ontr
o
l
Halft
oning
-
bas
e
d
B
T
C
i
mage
r
e
c
ons
tr
uc
ti
on
us
ing
patch
pr
oc
e
s
s
ing
w
it
h
bor
de
r
c
ons
tr
aint
(
He
r
i
P
r
as
e
tyo)
395
f
ir
s
t
mo
ment,
s
e
c
ond
mom
e
nt,
e
tc
.
T
he
c
or
r
e
s
ponding
bit
map
im
a
ge
is
s
im
ply
obtaine
d
by
a
pplyi
ng
the
thr
e
s
holdi
ng
ope
r
a
ti
on
on
e
a
c
h
im
a
ge
block
wit
h
the
mea
n
va
lue
of
th
is
pr
oc
e
s
s
e
d
block.
T
his
bit
m
a
p
im
a
ge
c
ons
is
ts
of
two
binar
y
va
lues
(
0
a
nd
1)
,
in
whic
h
the
va
l
ue
0
is
r
e
plac
e
d
with
the
low
mea
n
va
lue,
whe
r
e
a
s
the
va
lue
1
is
s
ubs
ti
tu
ted
with
high
mea
n
va
lue,
in
the
de
c
oding
pr
oc
e
s
s
.
T
he
B
T
C
-
ba
s
e
d
im
a
ge
c
ompr
e
s
s
ion
c
a
n
pr
ovide
low
c
omput
a
ti
ona
l
c
ompl
e
xit
y,
howe
ve
r
,
it
of
ten
s
uf
f
e
r
s
f
r
o
m
the
blocking
e
f
f
e
c
t
a
nd
f
a
ls
e
c
ontour
i
s
s
ue
s
[
1,
2,
7
,
12
]
.
T
he
s
e
pr
oblems
make
it
les
s
s
a
ti
s
f
a
c
tor
y
f
or
human
pe
r
c
e
pti
on.
A
ne
w
type
o
f
tec
hnique,
na
mely
ha
lf
toni
ng
-
ba
s
e
d
block
tr
unc
a
ti
on
c
oding
(
HB
T
C
)
,
ha
s
be
e
n
pr
opos
e
d
to
ove
r
c
ome
thes
e
p
r
oblems
.
F
i
gur
e
1
de
picts
the
s
c
he
matic
diagr
a
m
of
HB
T
C
te
c
hnique.
T
he
H
B
T
C
s
ubs
ti
tut
e
s
the
B
T
C
b
it
map
i
mage
with
the
ha
lf
tone
im
a
ge
p
r
oduc
e
d
f
r
om
s
pe
c
if
ic
im
a
ge
h
a
lf
toni
ng
metho
ds
s
uc
h
a
s
void
-
a
nd
-
c
lus
ter
ha
lf
toni
ng
[
1
]
,
dit
he
r
ing
a
ppr
oa
c
h
[7
-
9]
,
e
r
r
or
dif
f
us
ion
tec
hnique
[
2
,
10,
11]
,
dot
dif
f
us
e
d
ha
lf
toni
ng
[
12]
,
e
tc.
T
his
tec
hnique
c
ompens
a
tes
the
f
a
ls
e
c
ontour
a
nd
blocking
e
f
f
e
c
t
pr
oblems
by
e
njoyi
ng
pe
c
uli
a
r
dit
he
r
e
f
f
e
c
ts
f
r
om
the
ha
lf
ton
e
im
a
ge
.
I
n
a
ddit
ion,
the
HB
T
C
o
f
f
e
r
s
a
lowe
r
c
omp
utational
c
ompl
e
xi
ty
dur
ing
the
p
r
oc
e
s
s
of
the
two
c
olor
qua
nti
z
e
r
s
de
ter
mi
na
ti
on
.
He
r
e
in,
the
c
olor
qua
nt
ize
r
s
a
r
e
s
im
ply
r
e
plac
e
d
with
the
mi
nim
u
m
a
nd
maximum
pixel
va
lues
f
ound
in
a
n
i
mage
block.
T
wo
popula
r
HB
T
C
methods
,
na
mely
the
o
r
de
r
e
d
dit
he
r
e
d
block
t
r
u
nc
a
ti
on
c
odi
ng
(
OD
B
T
C
)
[
1]
a
nd
e
r
r
or
d
if
f
us
io
n
block
tr
unc
a
ti
on
c
oding
(
E
DB
T
C
)
[
2
]
,
ha
ve
be
e
n
de
ve
lo
pe
d
a
nd
r
e
por
ted
in
the
li
ter
a
tur
e
.
T
he
OD
B
T
C
a
nd
E
DB
T
C
c
ha
nge
the
B
T
C
bit
map
im
a
ge
with
the
ha
lf
tone
im
a
ge
pr
oduc
e
d
f
r
om
the
or
de
r
e
d
d
it
he
r
ing
a
nd
e
r
r
or
dif
f
us
e
d
ha
lf
toni
ng
methods
,
r
e
s
pe
c
ti
ve
ly.
B
oth
o
f
the
O
DB
T
C
a
nd
E
DB
T
C
s
c
he
mes
yield
be
tt
e
r
im
a
ge
qua
li
ty
c
ompar
e
d
to
that
of
the
c
las
s
ica
l
B
T
C
method
a
s
r
e
por
ted
in
[
1
,
2
]
.
T
he
two
methods
c
a
n
be
a
ppli
e
d
to
other
im
a
ge
p
r
oc
e
s
s
ing
a
nd
c
omput
e
r
vis
ion
a
ppli
c
a
ti
ons
,
including
low
c
omput
a
ti
ona
l
im
a
ge
c
ompr
e
s
s
ion
[
1,
2
,
11
]
,
c
ontent
-
ba
s
e
d
im
a
ge
r
e
tr
i
e
va
l
[7
-
10]
,
r
e
c
ognit
ion
of
c
olor
buil
ding
[
14,
15
]
,
blood
im
a
ge
a
na
lys
is
[
16]
,
ob
jec
t
de
tec
ti
on
a
nd
t
r
a
c
king
[
17]
,
e
tc.
I
n
p
u
t
I
m
a
g
e
B
i
t
m
a
p
I
m
a
g
e
G
e
n
e
r
a
t
i
o
n
Q
u
a
n
t
i
z
e
r
D
e
t
e
r
m
i
n
a
t
i
o
n
H
a
l
f
t
o
n
i
n
g
-
B
T
C
D
e
c
o
d
i
n
g
M
i
n
a
n
d
M
a
x
Q
u
a
n
t
i
z
e
r
B
i
t
m
a
p
I
m
a
g
e
D
e
c
o
d
e
d
I
m
a
g
e
T
r
a
n
s
m
i
s
s
i
o
n
C
h
a
n
n
e
l
F
igur
e
1.
S
c
he
matic
diagr
a
m
of
the
ha
lf
to
ning
-
ba
s
e
d
B
T
C
Although
the
OD
B
T
C
a
nd
E
DB
T
C
s
igni
f
i
c
a
ntl
y
r
e
duc
e
the
blocking
e
f
f
e
c
t
a
nd
f
a
ls
e
c
ontour
is
s
ue
s
oc
c
ur
r
e
d
in
c
las
s
ica
l
B
T
C
tec
hnique,
the
im
pu
ls
ive
nois
e
is
a
lwa
ys
pr
e
s
e
nt
a
t
c
ons
ider
a
bly
high
leve
l.
T
o
r
e
duc
e
the
im
puls
ive
nois
e
a
nd
mi
ti
ga
te
the
bounda
r
y
e
f
f
e
c
t,
a
n
a
ddi
ti
ona
l
s
tep
c
a
n
be
a
ppl
ied
f
or
the
HB
T
C
de
c
ode
d
im
a
ge
s
.
T
he
nois
e
f
il
ter
ing
is
a
s
im
ple
a
nd
na
ïve
a
ppr
oa
c
h
to
s
uppr
e
s
s
the
a
ppe
a
r
e
d
im
puls
ive
nois
e
,
in
whic
h
a
s
pe
c
if
ic
window
s
ize
a
nd
ke
r
ne
l
va
lue
a
r
e
a
ppli
e
d.
T
he
Ga
us
s
ian
f
il
ter
is
a
n
e
xa
mpl
e
o
f
no
is
e
f
il
t
e
r
ing.
I
t
pe
r
f
or
ms
gl
oba
l
f
il
ter
ing,
i
.
e
.
a
ll
pixels
a
r
e
p
r
oc
e
s
s
e
d
in
the
s
a
me
manne
r
r
e
ga
r
dles
s
their
s
tatis
ti
c
a
l
int
r
ins
ic
pr
ope
r
ti
e
s
.
How
e
ve
r
,
it
ha
s
a
li
mi
ted
e
f
f
e
c
t
in
r
e
d
uc
ing
the
nois
e
leve
ls
.
An
e
xtende
d
Ga
u
s
s
ian
f
il
ter
ing
ha
s
be
e
n
pr
opos
e
d
in
[
13
]
,
na
mely
va
r
ianc
e
c
las
s
if
ied
f
il
ter
ing.
T
his
a
ppr
oa
c
h
a
ppli
e
s
va
r
ious
ke
r
ne
l
f
unc
t
ions
f
or
va
r
ious
pixels
,
a
nd
the
c
hoice
of
ke
r
ne
l
f
unc
ti
on
is
de
ter
mi
ne
d
by
the
va
r
ianc
e
withi
n
a
n
im
a
ge
block.
I
n
th
is
pa
r
ti
c
ular
method,
a
s
e
t
of
ke
r
ne
l
f
unc
ti
ons
c
a
n
be
it
e
r
a
ti
ve
ly
of
f
li
ne
-
tr
a
ined
us
ing
the
lea
s
t
-
mea
n
-
s
qua
r
e
d
(
L
M
S
)
ove
r
va
r
ious
im
a
ge
s
t
r
a
ini
ng
s
e
t.
T
he
s
e
s
e
t
of
ke
r
ne
l
f
unc
ti
ons
a
r
e
r
e
c
or
de
d
a
s
a
look
-
up
-
table
(
L
UT
)
f
or
f
u
r
t
he
r
us
a
ge
.
As
r
e
por
ted
in
[
13]
,
the
va
r
ianc
e
c
las
s
if
ied
f
il
ter
yields
a
s
igni
f
ica
ntl
y
be
tt
e
r
im
a
g
e
qua
li
ty
c
ompar
e
d
to
that
of
the
global
Ga
us
s
ian
low
pa
s
s
f
il
ter
ing
with
the
tr
a
de
-
of
f
of
h
igher
s
tor
a
ge
r
e
quir
e
ment.
T
he
s
pa
r
s
e
r
e
pr
e
s
e
n
tation
lea
r
ns
a
n
ove
r
-
c
ompl
e
te
dictionar
y
f
r
o
m
a
s
e
t
of
im
a
ge
pa
tche
s
a
s
tr
a
ini
ng
da
ta
[
18]
.
T
he
K
-
S
ingul
a
r
va
lue
de
c
ompos
it
ion
(
K
S
VD
)
s
pa
r
s
e
r
e
pr
e
s
e
ntation
tec
hnique
[
19,
20
]
o
f
f
e
r
s
s
table
r
e
s
ult
s
ove
r
the
e
xis
ti
ng
c
onve
x
r
e
laxa
ti
on
a
ppr
oa
c
he
s
f
or
the
s
pa
r
s
e
c
o
ding
lea
r
ning
a
nd
a
ppr
ox
im
a
ti
on.
As
r
e
por
ted
in
[
19
]
,
the
KSVD
a
ppr
oa
c
h
outpe
r
f
or
ms
t
he
matc
hing
pur
s
uit
[
21]
,
or
thogonal
matc
hing
pu
r
s
uit
[
22]
,
ba
s
is
pur
s
uit
[
23]
,
a
nd
maximum
a
pr
ior
i
(
M
AP)
a
ppr
oa
c
h
[
24
]
.
T
he
s
pa
r
s
e
r
e
pr
e
s
e
ntation
ha
s
be
e
n
de
mons
tr
a
ted
to
yield
a
pr
omi
s
ing
r
e
s
ult
in
s
e
ve
r
a
l
im
a
ge
pr
oc
e
s
s
ing
a
nd
c
omput
e
r
vis
ion
a
ppli
c
a
ti
ons
s
uc
h
a
s
im
a
ge
de
nois
ing
[
19,
25]
,
im
a
ge
r
e
s
tor
a
ti
on
[
26
,
2
7]
,
e
tc
.
Evaluation Warning : The document was created with Spire.PDF for Python.
I
S
S
N
:
1693
-
6930
T
E
L
KO
M
NI
KA
T
e
lec
omm
un
C
omput
E
l
C
ontr
o
l
,
Vol.
18
,
No
.
1
,
F
e
br
ua
r
y
2020
:
394
-
406
396
I
n
thi
s
pa
pe
r
,
a
ne
w
method
on
HB
T
C
im
a
ge
r
e
c
ons
tr
uc
ti
on
is
de
ve
loped
us
ing
the
s
pa
r
s
it
y
-
ba
s
e
d
a
ppr
oa
c
h.
He
r
e
in,
the
i
mpul
s
ive
nois
e
leve
ls
a
r
e
r
e
duc
e
d
by
mea
ns
of
c
oupled
dictionar
ies
,
in
w
hich
one
dictionar
y
is
c
r
e
a
ted
f
r
om
the
HB
T
C
im
a
ge
s
,
w
hil
e
the
other
is
lea
r
ne
d
f
r
om
c
lea
n
im
a
ge
s
(
unc
or
r
upted
im
a
ge
s
)
.
I
n
the
s
pa
r
s
e
c
oding
s
tage
,
the
s
pa
r
s
e
c
oe
f
f
icie
nts
a
r
e
f
i
r
s
tl
y
e
s
ti
mate
d
f
r
om
the
de
c
ode
d
im
a
ge
c
ontaining
high
im
puls
ive
nois
e
leve
ls
.
T
he
r
e
c
ons
tr
uc
ted
im
a
ge
is
then
pr
e
dicte
d
us
ing
the
c
le
a
n
im
a
ge
dictionar
y
with
the
pr
e
dicte
d
s
pa
r
s
e
c
oe
f
f
icie
nts
.
T
he
r
e
s
t
of
thi
s
pa
pe
r
is
or
ga
nize
d
a
s
f
oll
ows
.
S
e
c
ti
on
2
de
li
ve
r
s
the
VQ
-
ba
s
e
d
H
B
T
C
im
a
g
e
r
e
c
ons
tr
uc
ti
on.
S
e
c
ti
on
3
p
r
e
s
e
nts
the
s
pa
r
s
it
y
-
ba
s
e
d
HB
T
C
im
a
ge
r
e
c
ons
tr
uc
ti
on.
E
xtens
ive
e
xpe
r
im
e
ntal
r
e
s
ult
s
a
r
e
r
e
por
ted
a
t
s
e
c
ti
on
4.
F
inally,
the
c
onc
lu
s
ions
a
r
e
dr
a
wn
a
t
the
e
nd
of
thi
s
pa
pe
r
.
2.
VE
CT
OR
QUAN
T
I
Z
AT
I
ON
-
B
ASE
D
I
M
AG
E
RE
CONST
RU
CT
I
ON
T
his
s
e
c
ti
on
e
labor
a
tes
the
pa
tch
-
ba
s
e
d
pr
oc
e
s
s
i
ng
f
or
HB
T
C
im
a
ge
r
e
c
ons
tr
uc
ti
on
us
ing
ve
c
tor
qua
nt
iza
ti
on
(
VQ
)
a
ppr
oa
c
h.
He
r
e
in
,
the
im
a
ge
pa
tch
r
e
f
e
r
s
to
the
pr
oc
e
s
s
e
d
im
a
ge
block.
I
n
thi
s
m
e
thod,
a
s
ingl
e
im
a
ge
pa
tch
(
HB
T
C
de
c
ode
d
im
a
ge
)
is
r
e
plac
e
d
with
a
n
im
a
ge
pa
tch
obtaine
d
f
r
om
VQ
c
ode
wor
ds
ba
s
e
d
on
the
c
los
e
s
t
matc
hing
r
ule
[
11]
.
I
n
our
p
r
opos
e
d
method,
the
c
los
e
s
t
matc
hing
is
c
onduc
t
e
d
unde
r
the
E
uc
li
de
a
n
dis
tanc
e
s
im
il
a
r
it
y
c
r
it
e
r
ion.
T
he
s
m
a
ll
e
r
s
c
or
e
o
f
E
uc
li
de
a
n
dis
tanc
e
indi
c
a
tes
the
mor
e
s
im
il
a
r
.
T
he
s
e
lec
ted
im
a
ge
pa
tch
e
f
f
e
c
ti
ve
ly
im
pr
ove
s
the
qua
li
ty
of
HB
T
C
de
c
ode
d
i
mage
.
Dif
f
e
r
e
nt
im
a
ge
pa
tche
s
r
e
quir
e
di
f
f
e
r
e
nt
pr
oc
e
s
s
e
s
ba
s
e
d
on
their
pr
ope
r
ti
e
s
,
i.
e
.
“
bor
de
r
”
or
“
non
-
bor
de
r
”
type.
F
igu
r
e
2
s
h
ows
thr
e
e
pos
s
ibl
e
bor
de
r
pr
oc
e
s
s
ing
s
c
e
na
r
io
s
/cons
tr
a
int
s
,
i.
e
.
hor
izonta
l
,
ve
r
ti
c
a
l,
a
nd
c
or
ne
r
bor
de
r
s
.
As
s
hown
in
thi
s
f
igur
e
,
s
uppos
e
that
ther
e
a
r
e
f
our
di
f
f
e
r
e
nt
im
a
g
e
blocks
de
noted
a
s
two
gr
a
y
im
a
ge
blocks
a
nd
two
white
im
a
ge
blocks
.
T
he
s
e
im
a
ge
blocks
a
r
e
pr
oduc
e
d
f
r
om
the
HB
T
C
pr
oc
e
s
s
ind
e
pe
nde
ntl
y.
E
a
c
h
im
a
ge
block
is
unc
or
r
e
late
d
with
the
other
.
T
he
c
ha
r
a
c
ter
is
ti
c
s
of
one
im
a
ge
block
is
dis
s
im
il
a
r
wi
th
the
other
bl
oc
k
e
ve
n
though
they
a
r
e
a
djac
e
nt
ne
ighbor
s
.
He
r
e
in
,
a
n
i
mage
pa
tch
with
bor
de
r
c
ons
tr
a
int
r
e
f
e
r
s
to
a
s
e
t
of
pixels
whic
h
la
y
on
two
or
s
e
ve
r
a
l
dif
f
e
r
e
nt
im
a
ge
blocks
.
T
he
r
e
d
pa
r
t
o
f
F
igu
r
e
2
(
a
)
de
picts
a
s
e
t
of
pixe
ls
laying
on
s
e
ve
r
a
l
im
a
ge
bor
de
r
s
.
S
ince
thes
e
im
a
ge
pixel
s
a
r
e
in
hor
izonta
l
pos
it
ion,
we
r
e
ga
r
d
it
a
s
a
n
im
a
ge
pa
t
c
h
with
a
hor
izonta
l
bor
de
r
.
F
igur
e
s
2
(
b)
a
n
d
(
c
)
a
r
e
e
xa
mpl
e
s
of
im
a
ge
pa
tch
in
ve
r
ti
c
a
l
dir
e
c
ti
on
a
nd
c
or
ne
r
pa
r
t,
r
e
s
pe
c
ti
ve
ly.
I
mage
pa
tche
s
e
xc
luded
in
the
s
e
thr
e
e
bor
de
r
c
ons
tr
a
int
s
a
r
e
c
ons
ider
e
d
a
s
no
n
-
bor
de
r
im
a
ge
pa
tche
s
.
(
a
)
(
b)
(
c
)
F
igur
e
2.
T
ype
s
o
f
im
a
ge
pa
tch
pr
oc
e
s
s
ing
:
(
a
)
ho
r
izonta
l,
(
b
)
ve
r
t
ica
l,
a
nd
(
c
)
c
or
ne
r
bo
r
de
r
T
he
VQ
-
ba
s
e
d
method
r
e
moves
the
im
puls
ive
nois
e
on
the
bor
de
r
a
nd
non
-
bor
de
r
c
a
s
e
s
by
mea
ns
of
a
tr
a
ined
vis
ua
l
c
ode
book,
a
s
e
xplaine
d
be
low
.
T
his
method
r
e
plac
e
s
the
HB
T
C
de
c
ode
d
im
a
ge
pa
tch
with
the
vis
ua
l
c
ode
book
ge
ne
r
a
ted
f
r
o
m
c
lea
n
im
a
ge
s
.
T
he
c
lea
n
im
a
ge
s
c
a
n
be
s
im
ply
obtaine
d
f
r
o
m
na
tur
a
l
im
a
ge
s
,
i.
e
.
s
ome
im
a
ge
s
without
HB
T
C
pr
oc
e
s
s
i
ng.
S
uppos
e
=
{
(
,
)
}
a
r
e
the
c
lea
n
im
a
ge
s
whic
h
a
r
e
us
e
d
a
s
the
tr
a
ini
ng
s
e
t,
whe
r
e
=
1
,
2
,
…
,
.
T
he
s
ymbol
de
n
otes
the
number
o
f
t
r
a
ini
ng
im
a
ge
pa
tche
s
.
T
he
VQ
c
lus
ter
ing
it
e
r
a
ti
ve
ly
p
r
oc
e
s
s
e
s
thi
s
c
lea
n
im
a
ge
s
e
t
to
p
r
oduc
e
the
r
e
pr
e
s
e
ntative
vis
ua
l
c
ode
book
=
{
1
,
2
,
…
,
}
,
c
ontaining
c
ode
wor
ds
.
L
e
t
(
,
)
be
a
ha
lf
toned
-
ba
s
e
d
B
T
C
de
c
ode
d
im
a
ge
pa
tc
h.
T
he
VQ
-
ba
s
e
d
a
ppr
oa
c
h
f
ir
s
tl
y
identif
ies
whe
ther
the
im
a
ge
pa
tche
s
a
r
e
of
bor
de
r
o
r
non
-
bor
de
r
type.
I
f
a
n
im
a
ge
pa
tch
is
c
las
s
if
ied
a
s
non
-
bor
de
r
,
then
it
is
pr
oc
e
s
s
e
d
with
the
non
-
bor
de
r
vis
ua
l
c
ode
book
.
C
onve
r
s
e
ly,
whe
n
the
im
a
ge
pa
tch
is
c
las
s
if
ied
a
s
a
bor
de
r
pa
tch,
then
it
is
pr
oc
e
s
s
e
d
with
e
it
he
r
hor
izonta
l,
ve
r
ti
c
a
l,
or
c
or
ne
r
bor
de
r
vis
ua
l
c
ode
book.
F
igu
r
e
3
il
lus
tr
a
tes
the
VQ
-
ba
s
e
d
a
ppr
oa
c
h
f
or
HB
T
C
de
c
ode
d
im
a
ge
r
e
c
ons
tr
uc
ti
on.
F
igur
e
4
d
is
plays
s
ome
e
xa
mpl
e
s
of
the
vis
ua
l
c
ode
book
ove
r
h
or
izonta
l
,
ve
r
ti
c
a
l
,
a
nd
c
or
ne
r
bor
de
r
s
.
Af
te
r
bor
de
r
a
nd
non
-
bor
de
r
im
a
ge
pa
tch
de
ter
mi
na
ti
on,
th
is
im
a
ge
pa
tch
is
then
pr
oc
e
s
s
e
d
us
ing
VQ
-
c
los
e
s
t
matc
hing
a
s
de
f
ined
be
low:
Evaluation Warning : The document was created with Spire.PDF for Python.
T
E
L
KO
M
NI
KA
T
e
lec
omm
un
C
omput
E
l
C
ontr
o
l
Halft
oning
-
bas
e
d
B
T
C
i
mage
r
e
c
ons
tr
uc
ti
on
us
ing
patch
pr
oc
e
s
s
ing
w
it
h
bor
de
r
c
ons
tr
aint
(
He
r
i
P
r
as
e
tyo)
397
∗
=
a
r
g
m
in
=
1
,
2
,
…
,
‖
(
,
)
−
‖
2
2
(
1)
whe
r
e
de
notes
the
s
e
lec
ted
v
is
ua
l
c
ode
book,
a
nd
indi
c
a
tes
whe
ther
the
c
ur
r
e
nt
pr
oc
e
s
s
e
d
im
a
ge
pa
tc
h
is
of
“
bor
de
r
”
o
r
“
non
-
bor
de
r
”
type.
T
he
s
ymbol
∗
r
e
pr
e
s
e
nts
the
vis
ua
l
c
ode
wor
d
index
with
the
lowe
s
t
dis
tor
ti
on
(
the
mos
t
s
im
il
a
r
)
to
the
im
a
ge
pa
tch
(
,
)
.
F
igur
e
3.
S
c
he
matic
diagr
a
m
o
f
VQ
-
ba
s
e
d
im
a
ge
r
e
c
ons
tr
uc
ti
on
(
a
)
(
b)
(
c
)
F
igur
e
4.
Vis
ua
l
c
ode
book
ge
ne
r
a
ted
f
r
om
:
(
a
)
hor
i
z
ontal,
(
b
)
v
e
r
ti
c
a
l
,
a
nd
(
c
)
c
o
r
ne
r
bo
r
de
r
s
T
he
s
e
lec
ted
vis
ua
l
c
ode
wor
d
∗
r
e
plac
e
s
the
im
a
ge
pa
tch
(
,
)
by
c
ons
ider
ing
bor
de
r
or
non
-
bor
de
r
inf
or
mation
.
T
he
im
a
ge
pa
tch
r
e
plac
e
ment
is
de
noted
a
s
:
̃
(
,
)
=
∗
(
2)
whe
r
e
̃
(
,
)
i
s
the
r
e
s
tor
e
d
im
a
ge
pa
tch
on
pos
it
ion
(
,
)
.
Nota
bly
,
the
VQ
-
r
e
c
ons
tr
uc
ti
on
is
s
ti
ll
im
pleme
nted
a
s
a
pixel
-
by
-
pixel
pr
oc
e
s
s
,
not
a
s
block
-
wis
e
pr
oc
e
s
s
.
C
ons
e
que
ntl
y,
the
r
e
plac
e
ment
p
r
oc
e
s
s
is
Evaluation Warning : The document was created with Spire.PDF for Python.
I
S
S
N
:
1693
-
6930
T
E
L
KO
M
NI
KA
T
e
lec
omm
un
C
omput
E
l
C
ontr
o
l
,
Vol.
18
,
No
.
1
,
F
e
br
ua
r
y
2020
:
394
-
406
398
a
n
ove
r
de
ter
mi
ne
d
pr
oblem
.
T
he
a
ve
r
a
ging
pr
oc
e
s
s
ove
r
s
e
ve
r
a
l
r
e
s
tor
e
d
i
mage
s
pa
tche
s
yields
a
s
ingl
e
HB
T
C
r
e
c
ons
tr
uc
ted
im
a
ge
.
T
h
is
pr
oc
e
s
s
is
de
noted
a
s
:
̃
(
,
)
=
∑
̃
(
,
)
∑
(
,
)
(
,
)
(
3)
whe
r
e
(
,
)
de
notes
the
ope
r
a
to
r
o
f
im
a
ge
pa
tch
pr
oc
e
s
s
ing
[
18,
19,
26]
.
T
h
is
ope
r
a
tor
de
notes
the
numbe
r
of
c
e
r
tain
pixels
e
mpl
oye
d
on
the
c
los
e
s
t
matc
h
ing
pr
oc
e
s
s
a
nd
im
a
ge
pa
tch
s
ubs
ti
tut
ion.
T
his
ope
r
a
ti
on
indi
c
a
tes
the
number
of
c
ur
r
e
ntl
y
pr
oc
e
s
s
e
d
pixels
us
e
d
in
the
im
a
ge
pa
tch
c
omput
a
ti
on.
T
he
s
ize
of
matr
ix
(
,
)
is
identica
l
to
the
s
ize
of
im
a
ge
pa
tch
or
VQ
c
o
de
wor
d
s
.
T
he
r
e
s
ult
of
̃
(
,
)
in
(
3)
is
int
uit
ively
identica
l
to
that
o
f
the
s
im
ple
a
ve
r
a
ging
c
omp
utation
in
a
r
it
hmetics
.
T
hus
,
the
im
a
ge
pa
tch
a
ve
r
a
ging
ope
r
a
ti
on
is
e
mpl
oye
d
a
t
the
e
nd
o
f
t
he
VQ
-
ba
s
e
d
im
a
ge
r
e
c
ons
tr
uc
ti
on
to
yield
a
s
ingl
e
va
lue
in
the
o
ve
r
de
ter
mi
ne
d
s
ys
tem.
3.
S
P
AR
S
I
T
Y
-
B
ASE
D
I
M
AGE
RE
CONST
R
UC
T
I
ON
T
his
s
e
c
ti
on
p
r
e
s
e
nts
the
p
r
opos
e
d
HB
T
C
im
a
g
e
r
e
c
ons
tr
uc
ti
on
methods
us
ing
a
s
pa
r
s
it
y
-
ba
s
e
d
a
ppr
oa
c
h.
He
r
e
in,
the
im
a
ge
pa
tch
is
f
ir
s
tl
y
e
xtr
a
c
te
d
f
r
om
HB
T
C
de
c
ode
d
im
a
ge
.
T
he
s
pa
r
s
it
y
-
ba
s
e
d
a
ppr
oa
c
h
s
im
ply
r
e
plac
e
s
thi
s
de
c
ode
d
im
a
ge
pa
tch
with
the
c
los
e
s
t
matc
h
of
c
lea
n
im
a
ge
dictionar
y.
T
his
met
hod
a
ls
o
c
ons
ider
s
the
bor
de
r
c
ons
tr
a
int
on
HB
T
C
im
a
ge
r
e
c
ons
tr
uc
ti
on.
T
he
s
pa
r
s
it
y
-
ba
s
e
d
method
uti
li
z
e
s
two
lea
r
ne
d
c
oupled
diction
a
r
ies
.
An
im
a
ge
pa
tch
is
f
ir
s
tl
y
de
ter
mi
ne
d
a
nd
inves
ti
ga
ted
whe
ther
it
f
a
ll
s
int
o
the
bor
de
r
or
non
-
bor
de
r
r
e
gion
a
s
a
lr
e
a
dy
int
r
od
uc
e
d
in
the
VQ
-
ba
s
e
d
pr
oc
e
s
s
ing.
F
igur
e
5
dis
plays
the
H
B
T
C
im
a
ge
r
e
c
ons
tr
uc
t
ion
us
ing
s
pa
r
s
it
y
-
ba
s
e
d
a
ppr
oa
c
h.
F
igur
e
6
gi
ve
s
s
ome
e
xa
mpl
e
of
lea
r
ne
d
dictiona
r
y.
S
im
il
a
r
ly,
to
the
VQ
-
ba
s
e
d
pos
t
pr
oc
e
s
s
ing
a
ppr
oa
c
h,
the
s
pa
r
s
it
y
-
ba
s
e
d
method
uti
li
z
e
s
both
non
-
bor
de
r
dictionar
y
a
nd
bo
r
de
r
d
ictionar
y.
T
he
lea
r
ne
d
dicti
ona
r
y
is
ge
ne
r
a
ted
f
r
om
a
s
e
t
of
im
a
ge
pa
tche
s
by
c
ons
ider
in
g
the
non
-
bor
de
r
a
nd
bor
de
r
c
ons
tr
a
int
s
.
T
h
e
bor
de
r
dictionar
y
c
a
n
be
c
las
s
if
ied
a
s
hor
izonta
l
,
ve
r
ti
c
a
l,
a
nd
c
or
ne
r
bor
de
r
.
E
a
c
h
dictionar
y
c
ontains
HB
T
C
de
c
ode
d
dictionar
y
a
nd
c
lea
n
im
a
ge
d
ictionar
y.
F
ig
ur
e
5.
S
c
he
matic
diagr
a
m
of
s
pa
r
s
it
y
-
ba
s
e
d
im
a
ge
r
e
c
ons
tr
uc
ti
on
Evaluation Warning : The document was created with Spire.PDF for Python.
T
E
L
KO
M
NI
KA
T
e
lec
omm
un
C
omput
E
l
C
ontr
o
l
Halft
oning
-
bas
e
d
B
T
C
i
mage
r
e
c
ons
tr
uc
ti
on
us
ing
patch
pr
oc
e
s
s
ing
w
it
h
bor
de
r
c
ons
tr
aint
(
He
r
i
P
r
as
e
tyo)
399
F
igur
e
6.
L
e
a
r
ne
d
dictionar
ies
ge
ne
r
a
ted
f
r
om
ho
r
i
z
ontal,
ve
r
ti
c
a
l,
a
nd
c
or
ne
r
bor
de
r
a
s
indi
c
a
ted
in
the
f
ir
s
t
to
the
las
t
r
ow.
T
he
lef
t
a
nd
r
ight
c
olum
ns
de
note
the
c
oupled
dictionar
ies
ge
ne
r
a
ted
f
r
om
ha
lf
toned
-
B
T
C
a
nd
c
lea
n
im
a
ge
pa
tch,
r
e
s
pe
c
ti
ve
ly
L
e
t
=
{
(
,
)
,
(
,
)
}
be
a
tr
a
ini
ng
s
e
t.
T
his
s
e
t
c
ontains
c
lea
n
im
a
ge
pa
tche
s
(
,
)
a
nd
t
he
ir
c
or
r
e
s
ponding
HB
T
C
de
c
ode
d
im
a
ge
pa
tche
s
(
,
)
.
T
he
f
oll
owing
opti
mi
z
a
ti
on
p
r
oc
e
dur
e
lea
r
ns
two
dictionar
ies
,
i.
e
.
c
lea
n
im
a
ge
d
ictionar
y
(
)
a
nd
HB
T
C
im
a
ge
dictionar
y
(
ℎ
)
,
f
r
om
tr
a
ini
ng
s
e
t
.
T
o
ob
tain
the
s
pa
r
s
e
c
oe
f
f
icie
nts
,
plea
s
e
r
e
f
e
r
[
19]
f
o
r
f
ur
the
r
de
tailed
e
xplana
ti
on
of
opti
mi
z
a
ti
on
pr
oc
e
s
s
.
T
his
pr
oc
e
s
s
is
de
noted
a
s
:
m
in
,
ℎ
,
{
‖
−
‖
2
2
+
‖
−
ℎ
‖
2
2
}
+
‖
‖
1
(
4)
whe
r
e
de
notes
s
pa
r
s
e
c
oe
f
f
icie
nts
a
nd
is
s
pa
r
s
e
r
e
gular
iza
ti
on
ter
m
.
S
ince
two
dictionar
ies
a
nd
ℎ
s
ha
r
e
the
s
a
me
s
pa
r
e
c
oe
f
f
icie
nt
,
the
opt
im
iza
ti
on
in
(
4)
c
a
n
a
ls
o
be
pe
r
f
or
med
a
s
:
m
in
,
{
‖
−
‖
2
2
}
+
‖
‖
1
(
5)
whe
r
e
=
[
;
ℎ
]
a
nd
[
,
]
de
note
the
c
onc
a
tena
ted
dictiona
r
y
a
nd
c
onc
a
tena
ted
im
a
ge
pa
tch,
r
e
s
pe
c
ti
ve
ly.
T
he
dictionar
y
c
ontains
the
c
lea
n
a
nd
HB
T
C
d
e
c
ode
d
c
omponent
dictionar
ies
.
T
he
ma
tr
ix
c
ons
is
ts
of
the
c
lea
n
a
nd
HB
T
C
de
c
ode
d
im
a
ge
pa
tche
s
.
T
he
KSVD
[
19]
o
r
the
o
ther
dictionar
y
lea
r
ning
a
lgor
it
hms
c
a
n
be
e
xploi
ted
to
e
f
f
e
c
ti
ve
ly
s
olve
th
i
s
opti
mi
z
a
ti
on
pr
oblem
in
(
5)
.
Af
ter
de
c
idi
ng
the
b
or
de
r
o
r
non
-
bor
de
r
r
e
gion,
the
ne
xt
s
tep
is
to
de
ter
mi
ne
th
e
s
pa
r
s
e
c
oe
f
f
icie
nt
of
the
HB
T
C
im
a
ge
pa
tch
(
,
)
with
a
s
pa
r
s
e
c
oding
s
tep.
T
he
s
pa
r
s
e
c
oe
f
f
icie
nt
(
,
)
c
a
n
be
pr
e
dicte
d
with
a
he
lp
o
f
ℎ
a
s
f
oll
ow
:
Evaluation Warning : The document was created with Spire.PDF for Python.
I
S
S
N
:
1693
-
6930
T
E
L
KO
M
NI
KA
T
e
lec
omm
un
C
omput
E
l
C
ontr
o
l
,
Vol.
18
,
No
.
1
,
F
e
br
ua
r
y
2020
:
394
-
406
400
m
in
∗
{
‖
(
,
)
−
ℎ
‖
2
2
}
+
‖
‖
1
(
6)
whe
r
e
ℎ
de
notes
the
HB
T
C
de
c
ode
d
im
a
ge
unde
r
bo
r
de
r
or
non
-
bor
de
r
r
e
gion
.
T
he
s
ymbol
∗
r
e
p
r
e
s
e
nts
the
pr
e
dicte
d
s
pa
r
s
e
c
oe
f
f
icie
nt.
S
e
ve
r
a
l
s
pe
c
if
ic
a
lgor
it
hms
,
s
uc
h
a
s
matc
hing
pur
s
uit
(
M
P
)
[
21
]
,
or
t
hogona
l
matc
hing
pur
s
uit
(
OM
P
)
[
22
]
,
B
a
s
is
P
ur
s
uit
(
B
P
)
[
23]
,
maximum
a
pos
ter
ior
i
(
M
AP)
[
24]
,
o
r
othe
r
s
,
c
a
n
be
e
xploi
ted
to
pr
e
dict
∗
.
B
y
mea
ns
of
c
lea
n
im
a
ge
dicti
ona
r
y
with
bo
r
de
r
or
non
-
bor
de
r
r
e
gion
,
the
HB
T
C
de
c
ode
d
im
a
ge
c
a
n
be
r
e
plac
e
d
a
nd
a
li
gne
d
a
s
f
oll
ow
:
m
in
̃
(
,
)
{
‖
(
,
)
̃
(
,
)
−
∗
‖
2
2
}
(
7)
whe
r
e
̃
(
,
)
de
notes
th
e
HB
T
C
r
e
c
ons
tr
uc
ted
im
a
ge
.
S
i
mi
lar
ly,
to
the
VQ
-
ba
s
e
d
pos
t
pr
oc
e
s
s
ing
a
ppr
oa
c
h
,
̃
(
,
)
is
a
n
ove
r
de
ter
m
ined
s
ys
tem.
T
hus
,
the
̃
(
,
)
c
a
n
be
s
ol
ve
d
us
ing
the
or
dina
r
y
lea
s
t
s
qua
r
e
d
method
(
a
ve
r
a
ging
pr
oc
e
s
s
)
a
s
f
oll
ow
:
̃
(
,
)
=
{
∑
(
,
)
(
,
)
}
−
1
∑
(
,
)
∗
(
8)
T
he
im
a
ge
pa
tch
is
e
xtr
a
c
ted
us
ing
(
,
)
ope
r
a
tor
.
I
t
c
a
n
be
e
xpe
c
ted
that
the
HB
T
C
im
a
ge
r
e
c
ons
tr
uc
ti
on
will
yield
be
tt
e
r
im
a
ge
qua
li
ty
s
ince
it
uti
li
z
e
s
the
c
lea
n
im
a
ge
dictionar
y.
I
n
a
ddit
ion,
the
c
e
ntr
a
li
z
e
d
s
pa
r
s
e
r
e
pr
e
s
e
ntation
[
26]
c
a
n
a
ls
o
b
e
e
mbedde
d
int
o
the
s
pa
r
s
e
r
e
pr
e
s
e
ntation
modul
e
t
o
f
u
r
ther
im
pr
ove
the
qua
li
ty
of
the
HB
T
C
de
c
ode
d
im
a
ge
.
I
t
is
ba
s
e
d
on
the
obs
e
r
va
ti
on
that
the
or
igi
na
l
im
a
ge
(
w
it
hout
HB
T
C
c
ompr
e
s
s
ion)
a
nd
the
HB
T
C
de
c
ode
d
i
mage
,
−
,
ha
ve
h
igh
p
r
oba
bil
it
y
to
f
it
the
L
a
plac
e
dis
tr
ibut
ion.
T
he
p
r
os
pe
c
ti
ve
r
e
a
de
r
s
a
r
e
s
ugge
s
ted
to
r
e
f
e
r
to
[
26]
f
o
r
the
f
ull
de
s
c
r
ipt
ion
o
f
c
e
ntr
a
li
z
e
d
s
pa
r
s
e
r
e
pr
e
s
e
ntation.
F
igu
r
e
7
de
picts
t
he
dis
tr
ibut
ion
of
s
pa
r
s
e
c
oding
be
twe
e
n
the
or
igi
na
l
im
a
ge
a
nd
HB
T
C
de
c
ode
d
im
a
ge
.
T
hus
,
incor
po
r
a
ti
ng
the
L
a
plac
ian
pr
ior
knowle
dge
on
the
s
pa
r
s
e
r
e
pr
e
s
e
ntation
f
r
a
mew
or
k
may
pr
oduc
e
be
tt
e
r
qua
li
ty
on
the
HB
T
C
r
e
c
ons
tr
uc
ted
im
a
ge
.
T
he
c
e
ntr
a
li
z
e
d
s
pa
r
s
e
r
e
pr
e
s
e
ntation
[
26]
c
a
n
be
de
ployed
int
o
the
pr
opos
e
d
method
by
c
ons
ider
i
ng
a
bor
de
r
or
non
-
bor
de
r
r
e
gion
.
(
a
)
(
b)
F
igur
e
7.
T
he
dis
tr
ibut
ion
of
s
pa
r
s
e
c
oding
f
r
o
m
H
B
T
C
r
e
c
ons
tr
uc
ted
im
a
ge
obtaine
d
f
r
om:
(
a
)
OD
B
T
C
de
c
ode
d
im
a
ge
,
a
nd
(
b
)
E
DB
T
C
de
c
o
de
d
im
a
ge
4.
E
XP
E
RI
M
E
NT
AL
RE
S
U
L
T
S
T
his
s
e
c
ti
on
de
mons
tr
a
tes
the
e
f
f
e
c
ti
ve
ne
s
s
a
nd
us
e
f
ulnes
s
of
the
pr
opos
e
d
method.
F
igur
e
8
s
hows
the
tr
a
ini
ng
a
nd
tes
ti
ng
s
e
ts
uti
li
z
e
d
f
or
the
e
xpe
r
i
ment.
T
he
t
r
a
ini
ng
s
e
t
c
ons
is
ts
of
s
ixt
e
e
n
gr
a
ys
c
a
l
e
im
a
ge
s
,
whe
r
e
a
s
the
tes
ti
ng
s
e
t
c
ontains
twe
nty
g
r
a
y
s
c
a
le
im
a
ge
s
.
T
he
t
r
a
ini
ng
a
nd
tes
ti
ng
im
a
ge
s
a
r
e
with
va
r
ious
im
a
ge
c
ondit
ions
s
uc
h
a
s
high
-
f
r
e
que
nc
y,
low
-
f
r
e
q
ue
nc
y,
da
r
k
a
nd
li
ght
br
ight
ne
s
s
,
e
tc
.
T
he
p
r
opos
e
d
method
a
ls
o
e
mpl
oys
a
s
e
t
of
tr
a
ini
ng
im
a
ge
s
e
ts
a
s
s
im
il
a
r
ly
us
e
d
in
[
13]
to
ge
ne
r
a
te
s
e
ve
r
a
l
vis
ua
l
c
ode
books
a
nd
dictio
na
r
ies
.
T
he
tr
a
ini
ng
im
a
ge
s
e
t
c
ons
is
ts
of
twe
nty
gr
a
ys
c
a
le
im
a
ge
s
ove
r
va
r
ious
c
ondit
ions
s
uc
h
a
s
va
r
ious
il
lum
ination
c
ondit
ions
,
dif
f
e
r
e
nt
li
ghti
ng,
f
r
e
qu
e
nc
y
a
nd
a
c
ti
vit
y,
e
tc.
E
a
c
h
t
r
a
ini
ng
im
a
ge
is
of
s
ize
-
0
.
0
6
-
0
.
0
4
-
0
.
0
2
0
0
.
0
2
0
.
0
4
0
.
0
6
0
20
40
60
80
100
120
140
-
0
.
1
5
-
0
.
1
-
0
.
0
5
0
0
.
0
5
0
.
1
0
.
1
5
0
.
2
0
20
40
60
80
100
120
140
Evaluation Warning : The document was created with Spire.PDF for Python.
T
E
L
KO
M
NI
KA
T
e
lec
omm
un
C
omput
E
l
C
ontr
o
l
Halft
oning
-
bas
e
d
B
T
C
i
mage
r
e
c
ons
tr
uc
ti
on
us
ing
patch
pr
oc
e
s
s
ing
w
it
h
bor
de
r
c
ons
tr
aint
(
He
r
i
P
r
as
e
tyo)
401
512
×
512
.
W
e
e
xtr
a
c
t
im
a
ge
pa
tche
s
f
r
om
a
ll
t
r
a
ini
ng
im
a
ge
s
us
ing
ove
r
lapping
s
tr
a
tegy.
He
r
e
in
,
the
ove
r
lapping
s
tr
a
tegy
r
e
f
e
r
s
to
the
pr
oc
e
s
s
in
whic
h
the
c
ur
r
e
nt
pr
oc
e
s
s
e
d
im
a
ge
pa
tch
is
moved
to
the
ne
w
pos
it
ion
unde
r
one
pixel
dif
f
e
r
e
nt
.
I
n
thi
s
oc
c
a
s
ion,
one
im
a
ge
pixel
may
be
pr
oc
e
s
s
e
d
s
e
ve
r
a
l
ti
m
e
s
s
ince
s
e
ve
r
a
l
im
a
ge
pa
tche
s
e
mpl
oy
thi
s
pixel.
I
n
a
r
e
s
ult
,
we
ob
tain
mo
r
e
than
f
ive
mi
l
li
on
im
a
ge
pa
tche
s
a
s
tr
a
ini
ng
s
e
t
f
o
r
c
ode
wor
ds
ge
ne
r
a
ti
on
a
nd
c
ode
books
lea
r
ning
in
the
VQ
a
nd
s
pa
r
s
it
y
-
ba
s
e
d
a
ppr
oa
c
h,
r
e
s
pe
c
ti
ve
ly.
T
his
a
mount
of
d
a
tas
e
t
i
s
s
uf
f
icie
nt
to
s
a
ti
s
f
y
the
va
r
iabili
ty
a
s
pe
c
t
of
im
a
ge
pa
tch
in
the
t
r
a
ini
ng
pr
oc
e
s
s
.
T
hus
,
the
pr
opos
e
d
method
c
a
n
be
dir
e
c
tl
y
a
ppli
e
d
to
a
ge
ne
r
a
l
c
a
s
e
e
ve
n
thou
gh
the
t
r
a
ini
ng
da
ta
is
not
f
r
om
the
s
a
me
or
s
pe
c
if
i
c
da
tas
e
t.
T
he
pe
a
k
-
s
ignal
-
nois
e
-
r
a
ti
o
(
P
S
NR
)
s
c
or
e
e
va
luate
s
the
pe
r
f
or
manc
e
of
pr
opos
e
d
method
a
nd
f
or
mer
s
c
he
me
s
objec
ti
ve
ly.
T
he
P
S
NR
is
f
or
mul
a
ted
a
s
:
=
10
l
o
g
10
255
2
1
∑
∑
[
(
,
)
−
̃
(
,
)
]
2
=
1
=
1
(
9)
whe
r
e
̃
(
,
)
a
nd
(
,
)
de
note
the
HB
T
C
d
e
c
ode
d
im
a
ge
a
nd
or
igi
na
l
im
a
ge
,
r
e
s
pe
c
ti
ve
ly.
Highe
r
va
lue
o
f
P
S
NR
indi
c
a
tes
higher
s
im
il
a
r
it
y
be
twe
e
n
two
im
a
ge
s
,
making
it
mor
e
pr
e
f
e
r
a
ble
f
o
r
human
vis
ion.
T
hor
ough
e
xpe
r
im
e
nt,
the
im
a
ge
blocks
a
r
e
s
e
t
a
s
8
×
8
a
nd
16
×
16
f
or
bo
th
OD
B
T
C
a
nd
E
DB
T
C
methods
.
T
he
im
a
ge
qua
li
ty
im
p
r
ove
ment
a
f
ter
pos
t
p
r
oc
e
s
s
ing
is
c
ons
ider
e
d
ba
s
e
d
on
the
incr
e
a
s
ing
P
S
NR
va
lue
obtai
ne
d
a
f
ter
a
pplyi
ng
the
pos
t
-
pr
oc
e
s
s
ing
s
tep
to
the
HB
T
C
de
c
ode
d
im
a
ge
a
ga
ins
t
the
c
a
s
e
of
not
a
pplyi
ng
that
s
t
e
p.
(
a
)
(
b)
F
igur
e
8.
A
s
e
t
o
f
im
a
ge
s
us
e
d
in
thi
s
e
xpe
r
im
e
nt
a
s
:
(
a
)
t
r
a
ini
ng
s
e
t,
a
nd
(
b
)
tes
ti
ng
s
e
t
4.
1.
E
f
f
e
c
t
ivenes
s
of
t
h
e
p
r
op
os
e
d
m
e
t
h
o
d
T
his
s
ubs
e
c
ti
on
r
e
por
ts
the
e
xpe
r
im
e
ntal
r
e
s
ult
s
on
the
HB
T
C
de
c
ode
d
im
a
ge
r
e
c
on
s
tr
uc
ti
on.
T
he
pe
r
f
or
manc
e
of
the
pr
opos
e
d
method
is
vis
ua
ll
y
judged
ba
s
e
d
on
the
im
a
ge
qua
li
ty
of
the
pos
t
p
r
oc
e
s
s
in
g
r
e
s
ult
s
.
T
he
pr
opos
e
d
method
yields
be
tt
e
r
r
e
c
ons
tr
uc
ted
im
a
ge
qua
li
ty
c
ompar
e
d
to
the
o
r
igi
na
l
HB
T
C
de
c
ode
d
im
a
ge
s
.
He
r
e
in,
a
s
ingl
e
im
a
ge
is
f
ir
s
t
ly
e
nc
ode
d
us
ing
e
it
he
r
the
OD
B
T
C
or
the
E
DB
T
C
met
hod
to
yield
two
c
olor
qua
nti
z
e
r
s
a
nd
a
b
it
map
im
a
ge
.
T
he
s
e
two
methods
s
im
ply
s
e
t
the
i
mage
bloc
k
s
ize
a
s
8
×
8
.
T
he
n,
the
de
c
oding
pr
oc
e
s
s
is
f
ur
ther
a
ppli
e
d
to
th
e
two
c
olor
qua
nti
z
e
r
s
a
nd
the
b
it
map
im
a
g
e
to
obt
a
in
the
OD
B
T
C
o
r
E
DB
T
C
de
c
ode
d
im
a
ge
.
T
he
pos
t
p
r
oc
e
s
s
ing
is
c
onduc
ted
on
the
OD
B
T
C
o
r
E
DB
T
C
de
c
ode
d
im
a
ge
to
f
ur
ther
e
xa
mi
ne
the
pe
r
f
or
manc
e
o
f
the
pr
opos
e
d
r
e
c
ons
tr
uc
ti
on
method
in
ter
ms
of
s
pe
c
if
ic
im
a
ging
tas
ks
s
uc
h
im
a
ge
c
ompr
e
s
s
ion,
r
e
tr
ieva
l
a
nd
c
las
s
if
ica
ti
on,
e
tc.
F
ir
s
tl
y,
we
c
ompar
e
the
VQ
a
nd
s
pa
r
s
it
y
-
ba
s
e
d
a
ppr
oa
c
he
s
with
other
s
c
he
mes
s
uc
h
a
s
lowpa
s
s
f
il
ter
ing
a
nd
va
r
ianc
e
-
c
las
s
if
ied
f
il
ter
ing.
T
he
low
pa
s
s
f
il
ter
ing
a
ppr
oa
c
h
e
mpl
oys
a
Ga
us
s
ian
ke
r
ne
l
of
s
ize
11
×
11
with
=
0
a
nd
=
1
to
s
uppr
e
s
s
the
ha
l
f
toni
ng
i
mpul
s
ive
nois
e
[
13]
.
On
the
other
ha
nd,
the
va
r
ianc
e
-
c
las
s
if
ied
f
il
ter
ing
a
ppr
oa
c
h
us
e
s
13
op
ti
mi
z
e
d
ke
r
ne
ls
of
s
ize
7
×
7
[
13]
.
T
he
VQ
-
ba
s
e
d
tec
hnique
e
xploi
ts
the
1024
opti
mal
vis
ua
l
c
ode
books
,
while
the
s
pa
r
s
it
y
-
ba
s
e
d
method
uti
li
z
e
s
1024
dictionar
y
a
tom
s
.
T
he
s
pa
r
s
it
y
-
ba
s
e
d
im
a
ge
r
e
c
ons
tr
uc
ti
on
e
mpl
oys
t
wo
lea
r
ne
d
dictio
na
r
ies
.
I
n
thi
s
a
ppr
oa
c
h,
the
f
ir
s
t
d
ictionar
y
is
f
or
s
uppr
e
s
s
ing
the
im
puls
ive
nois
e
while
the
ot
he
r
dictionar
y
f
or
r
e
duc
ing
the
nois
e
oc
c
ur
r
e
d
in
th
e
HB
T
C
Evaluation Warning : The document was created with Spire.PDF for Python.
I
S
S
N
:
1693
-
6930
T
E
L
KO
M
NI
KA
T
e
lec
omm
un
C
omput
E
l
C
ontr
o
l
,
Vol.
18
,
No
.
1
,
F
e
br
ua
r
y
2020
:
394
-
406
402
im
a
ge
bor
de
r
.
T
he
s
e
c
ond
dictionar
y
c
ons
ider
s
the
c
ondit
ion
of
HB
T
C
im
a
ge
bo
r
de
r
s
uc
h
a
s
mi
xing
h
or
izonta
l,
ve
r
ti
c
a
l,
a
nd
c
or
ne
r
bor
de
r
s
.
F
or
e
a
c
h
im
a
ge
b
or
de
r
,
a
c
oupled
dictionar
y
wi
ll
be
pr
oduc
e
d
c
ontaining
the
HB
T
C
de
c
ode
d
a
nd
c
lea
n
im
a
ge
c
omponents
.
T
he
s
pa
r
s
e
c
oe
f
f
icie
nt
is
p
r
e
dicte
d
f
r
om
the
O
DB
T
C
or
E
DB
T
C
de
c
ode
d
im
a
ge
by
mea
ns
of
HB
T
C
dicti
ona
r
y,
while
the
pos
t
p
r
oc
e
s
s
ing
im
a
ge
is
c
ompo
s
e
d
a
nd
a
li
gne
d
us
ing
the
c
lea
n
im
a
ge
dictionar
y
with
pr
e
dicte
d
s
pa
r
s
e
c
oe
f
f
icie
nt.
T
he
im
a
ge
pa
tch
is
ini
ti
a
ll
y
c
las
s
if
ied
a
s
of
“
bor
de
r
”
or
“
non
-
bor
de
r
”
type.
F
igur
e
9
de
picts
the
im
a
ge
qua
li
ty
c
ompar
is
on
a
f
ter
a
pplyi
ng
the
pos
t
pr
oc
e
s
s
ing
methods
on
OD
B
T
C
de
c
ode
d
im
a
ge
.
As
s
hown
in
thi
s
f
igur
e
,
the
VQ
-
ba
s
e
d
tec
hnique
pr
oduc
e
s
be
tt
e
r
im
a
ge
qua
li
ty
in
c
o
mpar
is
on
with
the
lowpa
s
s
f
il
ter
ing
a
ppr
oa
c
h.
S
ome
im
pul
s
ive
nois
e
s
a
r
e
s
uc
c
e
s
s
f
ull
y
r
e
duc
e
d
by
a
pplyi
ng
lowpa
s
s
f
il
ter
ing,
but
the
r
e
s
olut
ion
o
f
the
r
e
c
ons
tr
uc
ted
im
a
ge
is
de
ter
ior
a
ted
due
to
the
blur
r
ing
e
f
f
e
c
t
of
the
lowpa
s
s
f
il
ter
s
.
T
he
s
pa
r
s
it
y
-
ba
s
e
d
method
of
f
e
r
s
the
be
s
t
OD
B
T
C
r
e
c
ons
tr
uc
ted
im
a
ge
c
om
pa
r
e
d
t
o
the
other
s
c
he
mes
a
s
de
mons
tr
a
ted
in
F
igur
e
9.
T
h
e
pos
t
pr
oc
e
s
s
ing
tec
hnique
f
or
E
DB
T
C
de
c
ode
d
i
mage
is
r
e
por
ted
in
F
igur
e
10
.
S
im
i
lar
ly,
to
the
OD
B
T
C
c
a
s
e
,
the
s
pa
r
s
it
y
-
ba
s
e
d
method
f
or
E
DB
T
C
de
c
ode
d
im
a
ge
s
outper
f
or
ms
the
lowpa
s
s
f
il
ter
ing
a
nd
VQ
-
ba
s
e
d
pos
t
pr
oc
e
s
s
ing.
T
hus
,
the
s
pa
r
s
it
y
-
ba
s
e
d
method
c
a
n
be
r
e
ga
r
de
d
a
s
a
s
a
ti
s
f
a
c
tor
y
pos
t
p
r
oc
e
s
s
ing
tec
hnique
f
or
im
pr
oving
the
HB
T
C
de
c
ode
d
im
a
ge
qua
li
t
y.
Us
ing
the
pr
opos
e
d
method,
the
HB
T
C
is
e
xpe
c
ted
to
a
c
h
ieve
low
c
omput
a
ti
ona
l
c
ompl
e
xit
y
in
the
HB
T
C
e
nc
oding
s
tage
a
nd,
a
t
the
s
a
me
ti
me
,
p
r
oduc
e
s
a
ti
s
f
a
c
tor
y
d
e
c
ode
d
im
a
ge
a
t
the
HB
T
C
de
c
oding
s
ide.
(
a
)
(
b)
(
c
)
(
d)
F
igur
e
9.
E
f
f
e
c
ti
ve
ne
s
s
of
the
p
r
opos
e
d
method
in
OB
T
C
im
a
ge
r
e
c
ons
tr
uc
ti
on
us
ing:
(
a
)
lowpa
s
s
f
il
ter
e
d
tec
hnique,
(
b)
VQ
-
ba
s
e
d
im
a
ge
r
e
c
ons
tr
uc
ti
on,
(
c
)
s
pa
r
s
it
y
-
ba
s
e
d
im
a
g
e
r
e
c
ons
tr
uc
ti
on
us
ing
two
lea
r
ne
d
dictionar
ies
,
a
nd
(
d
)
the
or
igi
na
l
OD
B
T
C
de
c
ode
d
im
a
ge
with
im
a
ge
block
s
ize
8
×
8
Evaluation Warning : The document was created with Spire.PDF for Python.
T
E
L
KO
M
NI
KA
T
e
lec
omm
un
C
omput
E
l
C
ontr
o
l
Halft
oning
-
bas
e
d
B
T
C
i
mage
r
e
c
ons
tr
uc
ti
on
us
ing
patch
pr
oc
e
s
s
ing
w
it
h
bor
de
r
c
ons
tr
aint
(
He
r
i
P
r
as
e
tyo)
403
(
a
)
(
b)
(
c
)
(
d)
F
igur
e
10.
E
f
f
e
c
ti
ve
ne
s
s
of
the
p
r
opos
e
d
method
i
n
E
B
T
C
im
a
ge
r
e
c
ons
tr
uc
ti
on
us
ing:
(
a
)
lowpa
s
s
f
il
ter
e
d
tec
h
nique,
(
b)
VQ
-
ba
s
e
d
im
a
ge
r
e
c
ons
tr
uc
ti
on,
(
c
)
s
pa
r
s
it
y
-
ba
s
e
d
im
a
ge
r
e
c
ons
tr
uc
ti
on
us
ing
two
lea
r
ne
d
dictionar
ies
,
a
nd
(
d
)
the
or
igi
na
l
E
DB
T
C
de
c
ode
d
im
a
ge
with
im
a
ge
block
s
ize
8
×
8
4.
2.
P
r
op
os
e
d
m
e
t
h
od
p
e
r
f
or
m
an
c
e
u
s
in
g
t
wo
d
ict
io
n
ar
ies
I
n
thi
s
e
xpe
r
im
e
nt,
t
he
pe
r
f
o
r
manc
e
of
the
pr
opos
e
d
method
a
nd
other
s
c
he
mes
is
objec
ti
ve
ly
a
s
s
e
s
s
e
d
in
ter
ms
of
P
S
NR
s
c
or
e
.
T
he
i
mage
block
s
ize
s
a
r
e
s
e
t
a
s
8
×
8
a
nd
16
×
16
f
or
both
OD
B
T
C
a
nd
E
DB
T
C
tec
hniques
.
T
he
s
pa
r
s
it
y
-
ba
s
e
d
method
s
im
ply
e
xpl
oit
s
two
d
ictionar
ies
,
i.
e
.
the
im
pu
ls
ive
nois
e
dictionar
y
a
nd
the
im
a
ge
bor
de
r
dictionar
y
.
E
a
c
h
dictionar
y
c
ons
is
ts
of
a
c
ouple
dictionar
ies
(
HB
T
C
im
a
ge
diction
a
r
y
a
nd
c
lea
n
im
a
ge
dictionar
y)
.
T
he
e
xpe
r
im
e
ntal
s
e
tt
i
ng
f
or
low
-
pa
s
s
a
nd
va
r
ianc
e
-
c
la
s
s
if
ied
f
il
ter
ing
r
e
mains
identica
l
to
that
o
f
s
e
c
ti
on
4
.
1.
T
a
ble
1
pr
e
s
e
nts
the
pe
r
f
o
r
manc
e
of
the
pr
o
pos
e
d
s
pa
r
s
it
y
-
ba
s
e
d
method
on
HB
T
C
im
a
ge
r
e
c
ons
tr
uc
ti
on
a
ga
ins
t
low
-
pa
s
s
a
nd
va
r
ianc
e
-
c
las
s
if
ied
f
il
te
r
ing,
VQ
-
ba
s
e
d
pos
t
-
pr
oc
e
s
s
ing
a
nd
a
bs
e
nc
e
of
pos
t
-
pr
oc
e
s
s
ing.
I
n
th
is
table
,
the
c
ompar
e
d
va
lue
(
de
noted
a
s
P
S
NR
s
c
or
e
)
is
the
a
v
e
r
a
ge
va
lue
of
P
S
NR
ove
r
a
ll
tes
ti
ng
im
a
ge
s
.
T
he
va
r
ianc
e
c
las
s
if
ied
f
il
ter
ing
tec
hnique
of
f
e
r
s
be
tt
e
r
im
a
ge
r
e
c
ons
tr
uc
ti
on
c
ompar
e
d
to
that
of
the
lowpa
s
s
f
il
te
r
ing.
W
he
r
e
a
s
the
VQ
-
ba
s
e
d
a
ppr
oa
c
h
is
s
upe
r
ior
c
ompar
e
d
to
f
il
ter
ing
methods
(
lowpa
s
s
f
il
ter
ing
a
nd
va
r
ianc
e
-
c
las
s
if
ied
f
il
ter
ing)
.
As
s
hown
f
r
om
thi
s
table
,
the
s
pa
r
s
it
y
-
ba
s
e
d
metho
d
yields
the
be
s
t
pe
r
f
or
manc
e
f
or
OD
B
T
C
a
nd
E
DB
T
C
de
c
ode
d
im
a
ge
ove
r
im
a
ge
block
s
ize
8
×
8
a
nd
16
×
16
.
T
hus
,
the
s
pa
r
s
it
y
-
ba
s
e
d
m
e
thod
with
two
lea
r
ne
d
dicti
ona
r
ies
is
s
uit
a
ble
to
im
pr
ove
the
im
a
ge
qua
li
ty
of
HB
T
C
de
c
ode
d
im
a
ge
.
Evaluation Warning : The document was created with Spire.PDF for Python.