TELKOM
NIKA
, Vol.11, No
.11, Novemb
er 201
3, pp. 6290
~6
295
e-ISSN: 2087
-278X
6290
Re
cei
v
ed Ap
ril 10, 2013; Revi
sed
Jun
e
22, 2013; Accepted July 5,
2013
Visible and Infrared Image Fusion using the Lifting
Wavelet
Yuelin Zou
1
*, Xiaoqiang Liang
1
, Tong
ming Wang
2
1
Soft
w
a
r
e
eng
i
neer
ing d
e
p
a
rtment,
Shiji
azh
uan
g Informati
on Eng
i
n
eeri
n
g
Vocation
al C
o
l
l
eg
e, Shiji
azh
u
ang
050
03
5, PR Chin
a
2
T
he Center of Educati
on T
e
chno
log
y
, H
eng
s
hui U
n
ivers
i
t
y
,
Hengs
hui, 0
5
3
000, PR Ch
in
a
*
Corres
p
o
ndi
n
g
author, e-ma
i
l
:
5382
395
3@
q
q
.com
A
b
st
r
a
ct
In rece
nt ye
ars i
m
ag
e fus
i
o
n
plays
a v
i
tal r
o
l
e
i
n
the
i
m
a
g
e
process
i
ng
ar
e
a
. F
u
sed
i
m
a
g
e
s w
oul
d
hel
p in d
o
i
ng
ma
ny ap
pl
icati
ons in
i
m
ag
e p
r
ocessi
ng lik
e
seg
m
e
n
tatio
n
, ima
ge e
n
h
anc
e
m
e
n
t an
d man
y
. In
order to i
m
pr
o
v
e the effect of fusion visib
l
e
and in
frar
ed i
m
a
ge i
m
ages
of the same scene, this p
a
p
e
r
prese
n
ts an i
m
a
ge fusi
on
meth
od b
a
se
d
on lifting w
a
velet do
main.
F
i
rstly,
the source i
m
a
ges
are
deco
m
pose
d
u
s
ing
liftin
g
w
a
v
e
let
do
ma
in
tra
n
sform (LW
T
). Secon
d
ly,
a w
e
ighte
d
av
era
g
e
ap
proac
h
bas
ed
on n
o
rmal
i
z
e
d
Shan
no
n entro
py is use
d
to fuse low
freq
ue
ncy lifting w
a
v
e
let co
efficient
s of the visib
l
e
a
n
d
infrare
d
i
m
ag
e
s
. T
he fusion
rule of loc
a
l
energy
maxi
mu
m is
used
to comb
in
e corresp
ond
in
g
hig
h
freque
ncy c
oef
ficients. After f
u
sin
g
l
o
w
an
d
hi
gh
frequ
enc
y coeffici
ents
of the
sourc
e
i
m
a
ges, th
e fi
n
a
l
fused i
m
a
ge is
obtai
ne
d usin
g
the invers
e LW
T
.
T
he
experi
m
ents show
that the
pro
pose
d
meth
od provi
d
es
improve
d
su
bj
ective a
nd
obj
ectives res
u
lts
as co
mp
are
d
to previ
ous
i
m
a
ge fus
i
o
n
meth
ods s
u
ch
as
Lap
laci
an trans
form a
nd traditi
ona
l W
a
velet transfor
m
.
Ke
y
w
ords
:
L
W
T
,
LEM, ima
ge fusio
n
, w
e
ig
hted aver
ag
e
Copy
right
©
2013 Un
ive
r
sita
s Ah
mad
Dah
l
an
. All rig
h
t
s r
ese
rved
.
1. Introduc
tion
Visible ima
g
e
s
are the ima
ges
obtaine
d
in t
he visible
spe
c
tru
m
an
d vary acco
rding to
the illuminati
on conditio
n
s. Visible ima
ge hol
ds t
h
e
details
of the ne
ce
ssary
feature
s
of t
he
image
s, req
u
i
red in the p
r
oces
s of bi
ometri
c auth
enticatio
n [1]. Since visibl
e image
s va
ry
according to
the lumina
nce and
con
d
itions
unde
r
which th
e test
image
s are
taken, they a
r
e
viable to
re
su
lt in e
rro
r
or
ot
herwise wrong re
cog
n
ition.
Infra
r
ed
i
m
age
s
are
capture
d
usin
g a
n
infrared
sen
s
or
came
ra i
n
the far inf
r
ared regi
on. Inf
r
ared Ima
ge
gives the
me
asu
r
e
of ene
rgy
radiatio
ns fro
m
the obje
c
t and it is le
ss
sen
s
it
ive to illumination
ch
ange
s an
d is
even ope
rabl
e in
darkne
s
s [2]. The en
ergy radiate
d
from
the obj
e
c
t, may cha
nge
according to
the su
rroun
di
ng
s
and du
e to th
e physi
cal ex
ertion
s. The f
eature
s
of
th
e obje
c
t, the prima
r
y req
u
i
s
ite for a
c
q
u
iring
the correlatio
n with
the
da
tabase im
age
s a
r
e
indi
stin
guishable
in
ca
se
of Infra
r
ed im
age
[3]. In
addition, infra
r
ed ima
ge a
s
a stand
alon
e
doe
s not pr
ovide high
-resol
ution data [4]
Hen
c
e, fu
sio
n
of visible
and
infra
r
ed
ima
ges provide
s
better
sol
u
tion to
achiev
e the
be
st fe
ature
of both
the
image
s for ta
rget re
co
gniti
on syste
m
[5]. Hence
this
pape
r propo
ses a novel
scheme of fusi
on
visible an
d infrared imag
es.
Becau
s
e of the multi-re
solution and
good p
r
op
ert
i
es of time–freque
ncy a
nalysi
s
,
wavelet tran
sform [6] reve
als its go
od
perfo
rman
ce
in the image
fusion field.Si
ngh propo
se
d a
fusion
schem
e based on
pixel, employs gen
etic alg
o
rithm
s
to op
erate the tra
d
itional wave
let
coeffici
ents t
o
deci
de h
o
w
to co
mbin
e IR with visible inform
ation [7]. Howe
ver, the gen
etic
algorith
m
s
require a la
rge
amout
of the time to work. Ha
rih
a
ran p
r
opo
se
d a
new image fusion tec
hni
que, utilizing Empiri
cal
Mode
Decompositio
n (EM
D
),
for
improve
d
face
recognitio
n
. EMD i
s
a n
o
n
-pa
r
am
etri
c
data-d
r
iven
a
nalysi
s
tool
that de
com
p
o
s
e
s
non-li
nea
r no
n-statio
na
ry sign
als into I
n
trinsi
c M
ode
Functio
n
s
(I
MFs) [8]. We
can
kn
ow t
he
EMD h
a
s Hig
h
time
com
p
lexity. A fusion alg
o
rithm i
s
pro
p
o
s
ed
to
combi
ne
pairs of m
u
ltisp
e
c
tra
l
magneti
c
resonan
ce im
agi
ng such a
s
T
1
, T2 and
proton de
nsity
brain i
m
ag
es.
The p
r
op
ose
d
algorith
m
utilize
s
different feature
s
of re
dund
ant
discrete wavelet transfo
rm, mutual informatio
n
based n
o
n
-
lin
ear
re
gistration a
nd e
n
tro
p
y inform
atio
n to imp
r
ove
perfo
rman
ce
[7]. However,
it
can o
n
ly ca
pture limited
orientation i
n
formatio
n, inclu
d
ing ho
ri
zontal, vertical and diag
o
nal
orientatio
ns i
n
ea
ch
de
co
mpositio
n
sta
ge [9]. In
ord
e
r to
overco
me the
probl
em, lifting
wa
velet
transfo
rm i
s
prop
osed to
use to fu
se i
m
age
s. The
norm
a
lized S
hann
on ent
ro
py is ado
pte
d
as
the fusion
rul
e
in low frequ
ency
coefficie
n
t. The lo
cal energy feature is u
s
ed to
select the fu
si
on
coeffici
ent in
high
-Frequ
e
n
cy coefficie
n
t. The
expe
riment
s de
m
onstrate the
fusion
metho
d
is
effective to fuse the visibl
e
and infra
r
ed
image fusi
on
and outpe
rfo
r
m the DWT
based meth
o
d
and La
pla
c
ia
n based meth
od.
Evaluation Warning : The document was created with Spire.PDF for Python.
TELKOM
NIKA
e-ISSN:
2087
-278X
Visible an
d Infrare
d
Im
age Fusio
n
usi
ng
the Lifting Wa
velet (Yu
e
lin
Zou)
6291
2. The Propo
sed Me
thod
2.1. Lifting Wav
e
let Transform
Lif
t
in
g sc
he
me, pu
t f
o
rwar
d by
Swel
den i
n
t
h
e 19
9
0
s
,
i
s
a k
i
nd of
wav
e
le
t
cons
tru
c
t
i
on
metho
d
d
o
e
s
not
re
ly on
the Fou
r
i
e
r
tran
sform
.
.Compa
red
with
tradition
al
WT
(wavel
et tran
sform,
WT
), LWT
(Lifting
Wa
velet T
r
ansfo
rm) po
sse
s
ses
seve
ral advanta
g
e
s
,
inclu
d
ing p
o
ssibility of ada
ptive and no
nlinea
r de
si
g
n
, in place calcul
ation
s
, irregul
ar
sam
p
les
and inte
gral
transfo
rm. It
can
be
se
en
as
an alte
rn
ate implem
e
n
tation of tra
d
itional
wave
le
t
transfo
rm. Th
e main featu
r
e of the lifting
wavelet tr
a
n
s
form i
s
that i
t
provide
s
an
entirely spati
a
l
domain
interpretation
of the tra
n
sfo
r
m,
in c
ontra
st t
o
the tra
d
itio
nal fre
quen
cy domain
ba
sed
con
s
tru
c
tion
s [10] lifting
wavelet
algo
rithm
reali
z
at
ion i
s
divid
e
d into th
re
e
step
s:
divisi
on,
predi
ction
an
d upd
ate. P
l
a
nd U
l
de
note
the predi
ction
and
upd
ate
operator
of th
e lifting wavel
e
t
at level
l
, respec
tively. a is
l
+1 level
d
e
com
p
o
s
ition
by LWT of t
he inp
u
t si
gn
al. d
l
+1
and
a
l
+1
respe
c
tively are th
e detail
and a
p
p
r
oxi
m
ate c
oeffici
ents afte
r L
W
T de
comp
osit
ion of the
a
l
. The
predi
ction
co
efficient and
update
coef
ficients at le
vel l in the lifting wavelet
transfo
rm a
r
e
expre
s
sed a
s
follows:
11
1
01
2
1
21
2
1
,
0
,
..0
,
,
...,
,
0
,
...0
,
ll
l
mm
p
pp
p
p
(1)
11
1
01
2
1
21
21
,
0
,
.
.0
,
,
..,
,
0
,
...0
,
ll
l
mm
uu
u
u
u
(2)
The d
e
comp
osition
re
sult
s of a
n
ap
proximat
ion
sig
nal at level
via lifting sta
t
ionary
wavelet are e
x
presse
d by followin
g
equ
a
t
ions.
11
11
1
,
ll
ll
l
l
l
l
da
P
a
a
a
U
d
(3)
Whe
r
e d
l
+1
an
d a
l
+1
are deta
il signal a
nd a
pproxim
ation
sign
al of a
l
at level
l
+
1
.
The re
con
s
truction proced
ure of
lifting
stati
onary wav
e
let
tran
sform
is dire
ctly achi
eved
from its forwa
r
d tran
sfo
r
m, whi
c
h is exp
r
essed a
s
bel
ow.
11
1
11
1
1
1
1
()
2
ll
l
ll
l
l
l
l
aa
U
d
d
P
a
U
d
(4)
The fo
rwa
r
d
and inve
rse t
r
an
sform
of li
fting st
ationa
ry wavel
e
t transfo
rm i
s
shown in
Figure 2. Co
mpared with t
he DWT, lifting wa
velet tra
n
sform do no
t down
s
ampl
e and up
sam
p
le
the high
pa
ss and th
e lo
wpass
coeffici
ents d
u
ri
ng
t
he de
co
mpo
s
ition and
re
constructio
n
of
the
image.
Whe
n
LWT i
s
int
r
o
duced to im
a
ge fusi
on, m
o
re i
n
form
ation for fu
sio
n
can
be o
b
tai
ned
and the impa
cts of mis-reg
i
stration o
n
the fus
ed
results ca
n also be
redu
ced effe
ctively.
Figure 1. De
compo
s
ition a
nd Re
co
nstru
c
tion Di
agra
m
of LWT
2.2. Fusion Rule
Con
s
id
erin
g the ch
ara
c
te
ri
stics of deco
m
pos
itio
n su
bban
ds, we adopt differe
nt fusion
rule
s to low-freque
ncy coef
ficient and hi
gh-frequ
en
cy coeffici
ent.
2.2.1. Fusion
Rule of Lo
w
-
Freq
uency
Coe
fficien
t
An app
ro
ach
[11] is
a
weig
hted ave
r
agi
n
g
pr
opo
se
d, i
n
which the
weig
hts
are
o
b
tained
usin
g a regi
o
n
-ba
s
e
d
activ
i
ty measurem
ent of
the low frequen
cy lifting wavel
e
t coefficient
s:
22
2
1
()
l
o
g
(
)
LL
SS
S
nR
W
R
xn
xn
R
(5)
Whe
r
e
R is
the regio
n
with size |R|,
and x
S
are t
he input lo
w-Freque
ncy
wavelet
coeffici
ents, and L
repre
s
ent
s
th
e
coarse
st resol
u
tion level.
Hen
c
e, th
e
comp
osite
lo
w-
Freq
uen
cy co
efficient is ge
nerate
d
usi
n
g
:
Evaluation Warning : The document was created with Spire.PDF for Python.
e-ISSN: 2
087-278X
TELKOM
NIKA
Vol. 11, No
. 11, Novemb
er 201
3: 629
0 – 6295
6292
(,
)
,
(
,
)
,
,
(,
)
(
,
)
LL
AA
BB
L
F
AB
w
i
jx
i
j
w
i
jx
i
j
xi
j
wi
j
w
i
j
(6)
Whe
r
e
,
L
F
x
ij
,
L
A
x
ij
and
,
L
B
x
ij
are the fuse
d and i
nput low-F
r
eque
ncy
c
oeffic
i
ent wavelet c
oeffic
i
ents
,
(,
)
A
wi
j
an
d
(,
)
B
wi
j
are
obtaine
d u
s
i
ng (6), an
d L
rep
r
e
s
ent
s
t
he coa
r
s
e
st
r
e
sol
u
t
i
on lev
e
l.
2.2.2. Fusion
Rule of Hig
h
-Fre
quen
c
y
coefficient
We u
s
e
d
the
fusion
rule
o
f
local e
nergy maximum to
com
b
ine
co
rrespon
ding
subba
nd
coeffici
ents.
We first cal
c
ulate the lo
cal ene
rgy fe
ature
s
,
A
li
E
and
,
B
li
E
of the high
-freque
ncy
coeffici
ent of the visible an
d infrared ima
ges. Th
e loca
l energy featu
r
e is defin
ed
as:
''
2
''
'
'
,l
,
i
x
,
y
W
x,
y
D
x
+
x,
y
li
xp
y
q
Ey
(7)
Whe
r
e D
l,i
(x
+x
’
,y+
y
’
) den
otes the
de
comp
ositio
n
coeffici
ent at
the
i
-th di
rection
a
l
sub
ban
d of t
he level
l.
W is
the ke
rnel
operator,
the
size
of
the
lo
cal
wind
ow is p×q, th
e
cen
t
er
kernel
ope
rat
o
r
coeffici
ent
W (0,0) i
s
e
q
ual to 1/
2, a
n
d
the othe
r
kernel
ope
rato
r coefficie
n
ts
are
equal to 1/[2
×(p
×q
−
1
)
]. The si
ze of the
local e
nergy featur
e
wind
o
w
that we
ch
oose is 3
×
3,
so
the W is
1/
1
6
1/
1
6
1
/
1
6
1/
1
6
1
/
2
1
/
1
6
1/
1
6
1/
1
6
1
/
1
6
W
.
Then, the hig
h
-fre
que
ncy coefficient de
ci
sion ma
p is:
AB
,,
,
1
i
f
E
x,
y
E
x,
y
x,
y
0
li
l
i
li
D
ot
he
r
w
i
s
e
(8)
In ord
e
r to
ke
ep the
con
s
i
s
tency in th
e h
i
gh-frequ
en
cy detailed
co
m
pone
nts of th
e fuse
d
image,
we a
dopt the
“maj
ority” p
r
in
cipl
e to do
co
nsi
s
ten
c
y dete
c
tion an
d mo
d
u
lation fo
r fu
sed
deci
s
io
n ma
p
,
x,
y
li
D
. We ca
n get the high
frequen
cy coefficient
s u
s
e th
e
,
x,
y
li
D
modified by the majo
rity consi
s
ten
c
y de
tection.
2.3. Fusion Appro
ach
We p
r
e
s
ent t
he propo
se
d
visible an
d in
frar
e
d
imag
e
fusion
algo
rithm. Figure 1
sho
w
s
the block dia
g
ram of the p
r
opo
se
d meth
od, whi
c
h
con
s
ist
s
of a nu
mber of e
s
se
ntial stage
s:
Step 1: Th
e
sou
r
ce im
age
s a
r
e
de
com
posed int
o
di
fferent di
re
ctions an
d scal
es
usi
n
g
the LWT
Step 2: L
o
w frequ
en
cy lifting
wavelet
coeffici
ent
s of
the final fu
se
d imag
e
are
obtaine
d
via weighte
d
averagi
ng, in whi
c
h the wei
ghts are obtai
ned u
s
ing (5)
Step 3: Hig
h
frequ
en
cy lifting wavelet coeffi
cient
s of
the so
urce i
m
age
s a
r
e in
tegrated
using local energy maximum rule.
“Maj
o
r
ity” prin
ciple
is ado
pted to do co
nsi
s
ten
cy detection a
n
d
modulatio
n for fused
high
-frequ
en
cy coe
fficients
Step 4: T
he i
n
verse L
W
T
of the n
e
w lo
w a
nd
high
freque
ncy liftin
g
wavelet
co
efficients
gene
rate
s the
final fused i
m
age.
3. Results a
nd Discu
ssi
on
Two
sets of
i
m
age
s with p
e
rfect
re
gist
ration
a
nd
one
set
of im
age
s
with mi
s-re
gistratio
n
are
used to
evaluate th
e
prop
osed fu
si
on al
gor
ith
m
. The
propo
sed ima
ge fu
sion meth
od
wa
s
tested
agai
nst several
state-of-th
e
-a
rt i
m
age fu
sio
n
method
s in
cl
uding
the
sim
p
le ave
r
agi
ng
, the
Lapla
c
ian
Transfo
rm (LT),
the discrete
wavelet tra
n
sform with the
same fu
sion
rule
s wei
ght
ed
averag
e (WA
)
and lo
cal
energy maximum (LEM
)
algorith
m
. Fu
rtheremo
r
e, prop
osed fu
sion
method i
s
co
mpared
with region
al ene
rg
y contra
st
pyramid alg
o
rith
m with re
sp
e
c
t to the hum
an
visual characteristics [
12] calle
d as Co
ntrast Pyram
i
d in this pa
per.
In the LT method five
decompo
sitio
n
levels i
s
u
s
ed fo
r ima
g
e
de
comp
osi
t
ion. For the
DWT ba
se
d
method
s, th
e
Evaluation Warning : The document was created with Spire.PDF for Python.
TELKOM
NIKA
e-ISSN:
2087
-278X
Visible an
d Infrare
d
Im
age Fusio
n
usi
ng
the Lifting Wa
velet (Yu
e
lin
Zou)
6293
available
“d
b2”
wavelet
is u
s
ed a
nd five decompo
sition l
e
vels a
r
e u
s
ed fo
r ima
g
e
decompo
sitio
n
. In the Con
t
rast Pyramid
method,
all the pap
ram
e
ter u
s
ed in th
e pape
r [12]
is
adopte
d
.
In the fused i
m
age o
b
taine
d
by the prop
ose
d
metho
d
, whi
c
h is
also
implemente
d
in the
lifting wavelet trans
form domain, s
o
me details
s
u
ch
as the cont
ours of
trees and the bri
ght
points a
r
e tra
n
sferre
d into the fused im
a
ge. In Fi
gure
3, the road
s’
details from the visible im
a
g
e
and th
e
person’s detail
s
from the
infrared im
age,
an
d
al
so
in
Figu
re
4 the
car’
s detail
s
from
the
visible image
and the hou
se’
s
details from the in
frared image, are better tran
sferred into the
fused ima
ge in the propo
sed method. Gene
rally, it
can be
see
n
in Figure 3
-
4 that the perso
n
is
brighte
r
i
n
th
e p
r
op
osed
method,
and
the contrast
o
f
the fu
sed
im
age
s
i
s
fa
r
be
tter compa
r
e
d
to
other meth
od
s.
Figure 2. Sch
e
matic Di
ag
ram of LWT
-
b
a
se
d Fusi
on
Algorithm
(a)
(b)
(c
) L
p
alac
i
a
n
(d) DWT
(e
) Contr
a
s
t
P
yar
m
i
d
(f) P
r
op
ose
d
m
e
hod
Figure 3. The
Fusion
Re
sul
t
s of ‘Fore
s
t’ Image by Different Meth
od
Re
g
i
st
er
ed
in
f
r
a
r
ed
i
m
ag
e
F
o
rw
ard L
W
T
L
W
T C
oef
fic
i
e
n
ts
L
W
T C
oef
fic
i
e
n
ts
Fu
s
i
on
ru
le
Fuse
F
u
se
d L
W
T
C
oef
fi
cie
n
ts
Inv
e
rse L
W
T
F
u
se
d Im
age
F
Fus
i
o
n
ru
le
Re
g
i
st
er
ed
v
i
s
i
b
l
e
i
m
a
ge
F
o
rw
ard L
W
T
Evaluation Warning : The document was created with Spire.PDF for Python.
e-ISSN: 2
087-278X
TELKOM
NIKA
Vol. 11, No
. 11, Novemb
er 201
3: 629
0 – 6295
6294
(a)
(b)
(c
) L
p
alac
i
a
n
(d) D
W
T
(e
) C
o
ntr
a
s
t
P
yar
m
i
d
(f) P
r
op
ose
d
m
e
hod
Figure 4. The
Fusion
Re
sul
t
s of ‘Hou
se’ Image by Different Meth
od
Two m
e
trics
are
con
s
id
ere
d
in this p
ape
r, whi
c
h d
o
n
o
t requi
re g
r
o
und truth im
a
ges fo
r
evaluation. T
he first m
e
tri
c
is
Q
AB/F
[13], which co
nsid
ers the
amount of e
dge info
rmati
on
transfe
rred from the in
put
image
s to th
e fuse
d im
a
g
e
usi
ng a
So
bel ed
ge d
e
tector to calcu
l
ate
the stren
g
th and ori
entatio
n information
at each
pixel in both sou
r
ce and the fused image
s. The
se
con
d
metri
c
is the e
n
tro
p
y index, whi
c
h m
e
a
s
ures the informati
on content i
n
an ima
ge. A
n
image with h
i
gh informati
on co
ntent will have hig
h
entropy. T
he third metric is the mut
ual
informatio
n (MI) metri
c
[1
4] used to
ev
aluate th
e fu
sion p
e
rfo
r
ma
nce
qu
antitatively in this pa
per.
Table 1
sho
w
s th
e average pe
rform
ance re
sult
s from differe
nt image fu
sion metho
d
s and
different d
a
ta
sets. T
he
re
sults p
r
e
s
ente
d
in th
i
s
exa
m
ple
can
de
monst
r
ate th
at our app
ro
ach
can fu
se th
e
visible a
nd in
frare
d
imag
e
s
while retain
ing mu
ch m
o
re info
rmatio
n than that of
the
other two m
e
thod
s.
Table 1. Perf
orma
nce Evaluation of Different Meth
od
Images Metric
Laplacian
DW
T
Contrast
p
y
ramid
LWT
Fores
t
Q
AB
/
F
0.2920
0.3129
0.3671
0.4019
Entrop
y 4.728
5.0257
5.1233
5.3845
MI 1.4603
1.5981
1.7198
1.9818
House
Q
AB
/
F
0.3715
0.4629
0.5211
0.5316
entrop
y
5.212
5.8210
5.9291
6.1428
MI 2.1027
2.3961
2.5521
2.8075
4. Conclusio
n
In this paper,
we have presented a ne
w lifting wavelet based visible
and infra
r
ed i
m
age
s
fusion meth
o
d
. Propo
sing
new fusio
n
rules for m
e
rgin
g high
and low fre
q
uen
cy wavel
e
t
coeffici
ents,
whi
c
h is the
se
con
d
step i
n
the wave
let
-
ba
sed im
age
fusion, is the
main novelty o
f
this p
ape
r. T
he
weig
hted
averag
e m
e
thod
and
lo
cal
ene
rgy m
a
ximum a
r
e
re
spectively u
s
e
d
o
n
the low frequ
ency an
d hig
h
frequ
en
cy coeffici
ents.
The expe
rim
ental re
sult
s
demon
strated
that
the pro
p
o
s
ed
method o
u
tperfo
rms th
e
stand
ard fu
sion metho
d
s
in the fusio
n
of infrare
d
a
nd
visible im
age
s. Th
e p
r
op
o
s
ed
imag
e f
u
sio
n
al
gorit
hm i
s
a
n
eff
e
ctive, efficie
n
t and
fea
s
i
b
le
algorith
m
. Fi
nally, it is i
m
porta
nt to
note t
hat th
e propo
se
d
LWT
-
ba
se
d f
u
sio
n
alg
o
rit
h
m
outperfo
rm
s the DWT-ba
sed fusio
n
alg
o
rithm in som
e
ca
se
s.
Evaluation Warning : The document was created with Spire.PDF for Python.
TELKOM
NIKA
e-ISSN:
2087
-278X
Visible an
d Infrare
d
Im
age Fusio
n
usi
ng
the Lifting Wa
velet (Yu
e
lin
Zou)
6295
Ackn
o
w
l
e
dg
ement
Some of the image
s used i
n
this pap
er a
r
e
availabl
e in http://www.i
m
agefu
s
ion.o
r
g. This
work
wa
s su
pporte
d in pa
rt by Universi
ty Sci
ence
Rese
arch Proj
ect of He
bei
Province und
er
grant
s 201
14
12 and Shijia
zhu
ang Tie
d
ao Unive
r
sity
proje
c
t of Education and
teaching reform
unde
r g
r
a
n
ts 110
404.Th
e
autho
rs al
so than
k th
e
ano
nymou
s
refe
ree
s
fo
r their valua
b
le
s
u
gg
es
tio
n
s
.
Referen
ces
[1]
Vaid
ehi V,
Ra
m
y
a R,
Pras
an
na Dev
i
M,
Naresh
Babu NT
,
Balam
u
rali
P,
Girish Chandr
a M.
F
u
sion of
Multi-Scal
e Vi
sible
an
d T
her
ma
l Imag
es u
s
ing EM
D for
Improv
ed F
a
c
e
Rec
o
g
n
itio
n.
Internationa
l
Confer
ence of
Engi
neers and Comp
ut
er Scie
ntist. 2011; 1: 543-
548.
[2]
F
eng W
e
i,
Bao W
e
n
x
ing. A
ne
w
tech
no
lo
g
y
of r
e
mote
sensi
ng
imag
e
fusio
n
.
Te
lko
m
nika
. 201
2
;
10(3): 55
1-5
5
6
.
[3]
Peng Gen
g
,
Z
heng
yo
u W
a
ng, Z
h
iga
ng Z
han
g, Z
hong
Xi
ao. Image fusio
n
b
y
p
u
ls
e
coupl
e neur
al
net
w
o
rk
w
i
th
shear
let.
Optical. En
gin
eer
in
g
. (
http://dx
.
doi.org/10.1
117/
1.OE.51.6.067005
Ju
n 06
,
201
2).
[4]
David
L
oon
e
y
,
Dan
i
l
o
P. Ma
n
d
ic. Multisc
a
l
e
Image F
u
s
i
on
Using
C
o
mpl
e
x E
x
te
nsi
ons
o
f
EMD.
IEEE
T
r
ansactio
n
s o
n
Sign
al Proc
e
ssing
. 20
09; 57
(4): 1626-
16
30
.
[5]
Harih
a
ra
n H, Koscha
n
A, Abidi B, Gribok
A, Abidi MA.
Fusio
n
of visibl
e and i
n
frared
ima
ges usi
n
g
empiric
a
l
mod
e
dec
o
m
p
o
siti
on to
improv
e
face rec
ogn
iti
o
n
. Proceedings of IEEE T
r
ans. on Image
Processi
ng ICIP200
6. Atlanta
,
GA. 2006; 20
49-2
052.
[6]
Arif MH, Shah SS.
Block le
vel multi-foc
u
s
ima
ge fu
si
on
using w
a
ve
let
transform.
Pr
ocee
din
g
s of
Sign
al Acq
u
isiti
on an
d Proces
sing. Ku
a
l
a L
u
m
pur, Mala
ys
ia
. 2009; 21
3-21
6.
[7]
Sing
h Sa
ura
b
h
,
Gy
a
our
ova A
g
lika,
Be
bis G
eorg
e
,
Pavli
d
is
Ioan
nis.
Infrar
ed and
visi
bl
e imag
e
fusi
o
n
for face recognition.
Proce
edi
ngs of SPIE. 2004; 54
04: 58
5
-
596.
[8]
Harih
a
ra
n H
a
ri
sh
w
a
r
an,
Kosc
han A
ndr
eas,
Abid
i Besm
a,Gribok A
ndr
ei,
Abidi
Mon
g
i.
F
u
sion of
visi
bl
e
and i
n
frare
d
i
m
ages us
in
g e
m
pirica
l mod
e
d
e
co
mp
ositio
n to improv
e face
recogn
itio
n
. Procee
din
g
s
of
Internatio
na
l C
onfere
n
ce o
n
Image Proc
essi
ng. 200
6: 204
9
-
205
2
[9]
Hua
ng Ji
anzh
a
o
, Xi
e Jia
n
, Li
Hon
g
ca,
T
i
an Gui,
Chen
Xi
a
obo. Se
lf-ada
pt
ive dec
omp
o
sit
i
on l
e
vel
de-
noisi
ng meth
od
based o
n
w
a
v
e
let transform.
Te
lkom
n
i
ka
. 20
12; 10(5): 1
015
-102
0.
[10]
Cla
yp
oo
le Ro
ger L, Davis
G
eoffre
y
M, S
w
e
l
dens W
i
m,
Baraniuk R
i
char
d G. Nonlin
ear
w
a
vel
e
t
transforms, for imag
e co
din
g
v
i
a l
i
fting.
IEEE
Transactions on Image Processing
. 20
03
; 12
(1
2
)
: 14
49
-
145
8.
[11]
W
an T
,
Canag
araj
ah N, Achi
m.
Segmentati
on-driv
en im
ag
e fusion b
a
sed
on alp
ha-stab
l
e
mode
lin
g of
w
a
velet coefficients.
IEEE Tra
n
sactions on M
u
ltim
edia
. 20
09
; 11(4): 624-6
3
3
.
[12]
He Do
ng-
Xu. Contrast p
y
r
a
mid bas
ed
im
age fus
i
on sc
heme for i
n
fra
r
ed ima
ge
and
visibl
e ima
ge.
Internatio
na
l Geosci
ence
and
Re
mote Se
nsi
ng Sy
mpos
iu
m (IGARSS).
2011; 597-
60
0.
[13]
Piell
a
G. A
ge
nera
l
frame
w
o
r
k for mu
ltires
ol
uti
on
ima
ge
fusion: from
pi
xe
ls to
reg
i
o
n
s
.
Information
Fu
sio
n
. 20
03;
4(4): 259
–2
80
[14]
Petrovic V,
Xyd
eas
C.
T
h
e
effects of s
ensor
no
ise
i
n
pix
e
l-l
e
ve
l i
m
a
g
e
fusio
n
perfor
m
a
n
ce
.
Procee
din
g
s of
the
T
h
ird Internatio
nal C
onfer
ence o
n
Image
F
u
sion. 200
0; 2: 14–1
9.
Evaluation Warning : The document was created with Spire.PDF for Python.