Int
ern
at
i
onal
Journ
al of Ele
ctrical
an
d
Co
mput
er
En
gin
eeri
ng
(IJ
E
C
E)
Vo
l.
10
,
No.
2
,
A
pr
il
2020, p
p. 18
14
~
1822
IS
S
N: 20
88
-
8708
,
DOI: 10
.11
591/
ijece
.
v10
i
2
.
pp1814
-
18
22
1814
Journ
al h
om
e
page
:
http:
//
ij
ece.i
aesc
or
e.c
om/i
nd
ex
.ph
p/IJ
ECE
A
hybri
d im
age simi
larity m
ea
su
re based
on a new c
ombin
atio
n
of
diff
er
ent simil
ar
it
y tec
hn
iqu
es
Nisreen
R
yad
h H
amz
a
1
,
Rasha
Ail Dihin
2
,
M
ohamme
d
Ha
s
an
Abdul
ameer
3
1
F
ac
ulty
of
Com
pute
r
Sc
ie
nc
e an
d
Inform
at
ion
T
ec
hnolog
y
,
Univ
ersity
of
Qadisi
yah,
I
raq
2,
3
Depa
rtment
of
Com
pute
r
Sci
en
ce
,
Facu
lty
of
E
duca
t
ion
for
girl
s
,
Univ
ersity
of
Kufa,
Ir
aq
Art
ic
le
In
f
o
ABSTR
A
CT
Art
ic
le
history:
Re
cei
ved
Ma
y
9
, 2
019
Re
vised
Oct
25
,
2019
Accepte
d
Oct
31, 201
9
Im
age
sim
il
arit
y
is
the
degr
ee
of
how
two
image
s
are
sim
ilar
or
dissim
il
ar.
It
computes
the
sim
il
arit
y
degr
ee
b
et
we
en
th
e
int
ensity
pa
tt
ern
s
in
image
s
.
A
new
image
si
m
il
ari
t
y
m
e
asure
named
(
HFEM
M
)
is
proposed
i
n
thi
s
pa
per
.
The
HF
EMM
is
compos
ed
of
t
wo
phase
s.
Pha
se
1,
a
m
odifi
e
d
histogra
m
sim
il
ari
t
y
m
e
asure
(
HSSIM
)
is
m
erg
ed
with
fea
tur
e
sim
il
ari
t
y
m
ea
sur
e
(
FSIM
)
to
ge
t
a
new
m
ea
sure
cal
le
d
(
HFM
)
.
In
p
hase
2,
th
e
resul
te
d
(
HFM
)
is
m
erg
ed
with
err
or
m
ea
sure
(
EMM
)
in
orde
r
to
get
a
ne
w
si
m
il
ari
t
y
m
ea
sure,
whi
ch
is
named
(
HFEM
M
).
Diff
ere
nt
kinde
s
of
nois
es
for
e
x
ample
Gauss
ia
n,
Unifo
rm
,
and
salt
&
p
peppe
r
noise
r
are
used
with
t
he
proposed
m
et
hods.
One
of
the
hum
an
face
da
ta
b
ase
s
(
AT
&
T
)
is
used
in
the
exp
eri
m
ent
s
and
ran
dom
i
m
age
s
are
used
as
well
.
For
th
e
evalua
t
ion,
the
sim
il
ar
ity
p
e
rce
nt
age
und
er
pea
k
k
sign
al
to
noise
ratio
(PS
NR)
is
used.
To
show
the
eff
ec
t
ive
ness
of
the
proposed
m
ea
sure,
a
compar
ision
anong
diffe
ren
t
sim
il
ar
te
chn
ique
such
as
SSIM,
HF
M,
EMM
and
HFEM
M
are
conside
red
.
Th
e
proposed
HFE
MM
ac
h
ie
ved
h
ighe
r
sim
il
ar
ity
result
when
PS
NR was l
ow c
om
par
ed
to
the o
the
r
m
et
hods
.
Ke
yw
or
d
s
:
Feat
ur
e
sim
i
la
rity
ind
ex
m
easur
e F
SI
M
Gau
s
sia
n n
oise
Histo
gr
am
si
m
il
arit
y
m
easur
e
HS
S
IM
Salt
an
d pe
ppe
r no
ise
SSI
M
Un
i
form
n
oise
Copyright
©
202
0
Instit
ut
e
o
f Ad
vanc
ed
Engi
n
ee
r
ing
and
S
cienc
e
.
Al
l
rights re
serv
ed
.
Corres
pond
in
g
Aut
h
or
:
Moh
am
m
ed
H
asan
A
bdulam
eer
,
Dep
a
rtm
ent o
f C
om
pu
te
r
Scie
nce,
Faculty
of E
ducat
ion
for Girl
s
,
Un
i
ver
sit
y o
f Kufa,
Iraq
.
Em
a
il
:
m
oh
a
m
m
ed.
alm
ay
al
i
@uo
kufa.e
du.iq
1.
INTROD
U
CTION
Im
age
si
m
il
ari
ty
in
rece
nt
y
ears
has
bec
om
e
a
key
po
i
nt
i
n
im
age
proces
si
ng
a
pp
l
ic
at
ion
s
for
exam
ple
m
on
it
or
i
ng,
rest
or
at
i
on
a
nd
ot
her
a
pp
li
cat
io
ns
[
1]
.
Sim
il
arit
y
ca
n
be
c
ha
racteri
zed
as
the
c
on
trast
betwee
n
t
wo
im
ages,
a
nd
th
e
sim
il
arit
y
measur
e
is
a
nu
m
erical
diff
e
re
nce
betwee
n
t
wo
dissim
il
ar
i
m
ages
unde
r
com
par
ison
[2
]
.
Wh
e
n
the
two
i
m
ages
m
at
ch
up
to
the
m
axi
m
u
m
si
m
il
ari
ty
,
t
he
si
m
il
arity
deg
ree
betwee
n
tw
o
s
ign
al
s
is
re
quired
t
o
te
st
the
syst
e
m
and
in
order
t
o
m
ake
a
decisi
on
[
3]
.
Si
m
il
arity
m
e
asur
e
m
et
ho
ds
ca
n
be
cl
assified
d
into
:
inf
or
m
at
i
on
the
or
et
ic
al
te
chn
iq
ues
,
an
d
sta
ti
sti
cal
tech
ni
qu
e
s
[4]
.
Seve
ral
stud
ie
s
relat
ed
to
si
m
i
la
rity
t
echn
i
qu
e
s
ha
ve
been
prese
nted
rece
ntly
.
In
1995,
pro
pose
d
a
new
in
f
orm
at
ion
-
theo
reti
c
approac
h
(M
utu
al
I
nfo
rm
ation
)
[5
]
.
In
2004,
the
r
esearche
rs
pr
e
sented
a
scal
e
cal
le
d
(
SSI
M
)
base
d
on
the
in
f
or
m
at
ion
of
t
he
st
ru
c
t
ur
e
[6
]
.
I
n
2014,
i
ntr
oduc
ed
a
new
m
e
asur
e
cal
le
d
(
HSSIM
)
t
hat
ba
sed
on
‘join
t
histo
gra
m
’.
The
m
eas
ur
e
outpe
rfo
r
m
s
sta
t
ist
ic
al
m
easur
e
(
SS
I
M)
[7]
.
On
th
e
oth
er
ha
nd,
i
n
2017
,
the
resea
rch
e
rs
propose
d
m
ea
su
re
sim
il
arity
cal
le
d
(
NMSE
)
de
pe
ned
on
norm
al
iz
ed
m
e
an
squa
re
er
ror
[
8]
.
A
New
li
ke
nes
s
m
easur
e
b
ase
d
on
‘Aff
i
nity
Pr
opa
gatio
n’
intr
oduce
d
in
2018
[
9].
As
we
ll
,
in
20
18
intr
oduce
d
li
ken
ess
me
as
ur
e
based
on
“jo
int
hist
ogram
-
Entropy”
[10].
I
n
2019
,
intrdu
oced
a
hybri
d
m
easur
e
f
or
si
m
il
arity
of
i
m
ages
[
11
]
.
I
n
this
pap
e
r,
pr
opos
e
d
a
hy
br
i
d
m
easur
e
for
i
m
a
ge
si
m
il
arity
cal
le
d
(
HFE
MM
).
The
rest
of
t
his
pa
per
is
or
ga
nized
a
s
f
ollo
w:
the
sim
il
ari
ty
te
chn
iq
ues
i
s
presente
d
i
n
sect
ion
2,
the
r
esearch
m
et
ho
d
is
expl
ai
ned
in
sect
ion
3,
sect
ion
4
pr
e
sents
the
resu
lt
s
and
di
scussions
an
d
the
con
cl
us
i
ons
is
pr
ese
nted
in
se
ct
ion
5.
Evaluation Warning : The document was created with Spire.PDF for Python.
In
t J
Elec
&
C
om
p
En
g
IS
S
N: 20
88
-
8708
A h
y
br
id
im
age simil
ar
it
y m
e
as
ure
ba
se
d o
n a new
c
ombi
nati
on o
f
diff
erent…
(
Ni
sreen R
ya
dh Ha
mza
)
1815
2.
SIM
IL
ARITY
TE
C
HNIQ
UES
Tech
niques
th
at
us
ed
f
or
S
i
m
i
la
rity
can
be
cl
as
sifie
d
into:
the
sta
ti
s
ti
cal
and
the
inf
or
m
at
ion
theo
reti
cal
techn
iq
ues
[4
,
7]
.
2.1. Sta
tistica
l si
mi
larity tec
hniques
It
is
po
ssible
to
ob
ta
in
va
luable
inf
orm
at
ion
fr
om
the
i
m
age
via
cal
culat
ing
'
sta
ti
s
ti
cal
,
m
ea
su
rem
ents'
f
or
e
xam
ple
'
Me
an'
,
‘V
aria
nce’
a
nd
(S
D
)
w
he
re
S
D
m
eans
sta
nder
d
di
vation.
This in
f
or
m
at
i
on can
b
e
u
ti
li
zed to ca
lc
ulate
si
m
il
arity o
f
i
m
age
[
12
]
.
2.1.1.
St
r
uctur
al simi
larit
y me
as
ure
(SSI
M)
In
2004,
the
a
uthors
i
ntrod
uc
ed
a
ne
w
sta
ti
sti
cal
m
easur
e
for
im
age
qu
a
li
ty
ind
ex
cal
le
d
Str
uctu
ral
Si
m
il
a
rit
y
In
de
x
Me
th
od
(
SSIM
)
[6
,
13]
that
util
iz
ed
di
sta
nce
f
unct
io
n
to
m
easur
e
m
ent
the
li
keness
reli
ed
upon stat
ist
ic
al
f
eat
ures
[12].
The
m
easur
e c
an be
disp
la
y i
n (1)
:
(
,
)
=
(
2
+
1
)
(
2
+
2
)
(
2
+
2
+
1
)
(
2
+
2
+
2
)
(1)
Wh
e
re µ
p, µq
are the
'm
eans'
and
σ
2
p
,
σ
2
q
ar
e the
‘v
a
riance
’; σ
pq
is t
he
'
cov
aria
nce'
,
C1
a
nd
C2
are c
ons
ta
nts
(
C1
=
(
R1
P
)
2
,
C2
=
(
R
2P
)
2
,
R1
an
d
R2
a
re c
onsta
nts
,
P
is m
axim
u
m
p
ixel value
).
2.2.2. F
ea
tu
re
simi
larity
m
ea
sure
(
FSI
M)
Feat
ur
e
Sim
il
a
rity
Me
asur
e
is
a
sta
ti
s
ti
cal
m
easur
em
ent
of
im
age
qu
al
it
y
est
i
m
ation
.
FSI
M
offe
red
by
[
14
]
.
FS
IM
com
pu
ta
ti
on
consi
sts
of
tw
o
ph
ase
s:
the
f
irst
phase
cal
c
ulate
s
the
sim
i
la
rity
of
local
(
S
L
)
a
s
fo
ll
ows
:
(
1
)
=
[
(
1
)
]
∝
×
[
(
1
)
]
(2)
Wh
e
re
S
Pc
repr
esents the
P
has
e Co
ngru
e
ncy
si
m
il
ari
ty
(
1
)
=
2
1
(
1
)
×
2
(
1
)
+
1
1
2
(
1
)
×
2
2
(
1
)
+
2
(3)
(
1
)
=
(
1
)
∈
+
∑
(
1
)
,
(
1
)
=
√
1
2
(
1
)
+
1
2
(
1
)
,
(
1
)
=
√
[
2
(
1
)
+
2
]
(4)
Wh
e
re
1
(
1
)
=
∑
(
1
)
,
1
(
1
)
=
∑
(
1
)
,
(
1
)
=
(
1
)
∗
,
(
1
)
=
(
1
)
∗
(5)
S
G
re
pr
ese
nts t
he GM si
m
i
la
rity
(
1
)
=
2
1
(
1
)
×
2
(
1
)
+
1
1
2
(
1
)
×
2
2
(
1
)
+
2
(6)
=
√
2
(
1
)
+
2
(
1
)
(7)
The
sec
ond p
ha
se is to calc
ul
at
e the
F
SI
M
be
tween
F
1(
x
)
a
nd
F
2(
x
):
FSIM
{
F1
(
x
1
)
,
F2
(
x
1
)
}
=
Φ
{
F1
(
1
)
,
F2
(
x
1
)
}
=
∑
∈
Ω
s
l
(
x
1
)
×
pc
(
x
1
)
x
∑
∈
Ω
pc
m
(
x
1
)
x
(8)
Wh
e
re
PC
max
(
x
1
)
is M
A
Xim
um
(
MAX)
bet
ween
'
PC
1
(
x
1
)'
an
d
'
PC
2
(
x
1
)’
[15]
.
2.2.
I
n
fo
r
ma
t
ion
-
th
e
oret
ic
s
im
il
arity
tec
h
niques
Inform
at
ion
-
th
eor
et
ic
al
sim
i
l
arit
y
te
chn
iq
ue
is
use
d
to
get
the
sim
i
lar
it
y
dep
e
nded
on
i
ntensit
y
values
[16
,
17]
.
2.2.1. Hi
st
og
r
am
simi
larit
y me
as
ure
Histo
gr
am
Si
m
il
arit
y
Me
asur
e
is
the
m
easur
em
ent
that
dep
e
nd
s
on
in
form
a
ti
on
the
or
et
ic
al
te
chn
iq
ue
via
us
i
ng
co
nventio
nal
histo
gr
am
and
c
omm
on
histogra
m
H
ij
[
7]
.
H
SSI
M
sugg
e
ste
d
pr
e
viousl
y
as
SS
I
M
Evaluation Warning : The document was created with Spire.PDF for Python.
IS
S
N
:
2088
-
8708
In
t J
Elec
&
C
om
p
En
g,
V
ol.
10
, No
.
2
,
A
pr
i
l 202
0
:
1814
-
1822
1816
scal
e
cannot
be
well
-
i
m
ple
m
ented
under
si
gn
i
fcan
t
noise
.
In
(
HSSIM
)
t
he
resea
rcher
app
li
ed
t
he
co
m
m
on
histo
g
ram
an
d com
bin
ed
it
with the c
onve
nti
on
al
histo
gr
am
as foll
ows
[18
,
19]
:
(
,
)
=
√
∑
∑
[
(
−
)
1
ℎ
+
1
]
2
2
2
(
9)
Wh
e
re,
ℎ
is
t
he
co
nventi
on
al
histo
gr
am
an
d
1
is
a
co
ns
ta
nt
.
L(
p,q
)
ca
n
be
no
rm
alize
by
us
in
g
1
(
,
)
wh
ic
h rep
rese
nt
the m
axi
m
u
m
v
al
ue
of t
he
e
r
ror
est
im
at
e in
ver
y l
ow PS
N
R as f
ollo
ws:
(
6
,
6
)
=
(
,
)
1
(
,
)
,
1
(
,
)
=
1
−
(
6
,
)
(10)
3.
RESEA
R
CH MET
HO
D
The
m
erh
od a
nd
pro
po
se
d
Me
asur
e
(
HF
EM
M) is de
picte
d
in
F
ig
ure
1
.
Figure
1.
The
s
i
m
i
la
rity
b
et
ween
tw
o
im
ages
(v
e
rificat
i
on) b
y t
he
pr
o
pose
d sim
il
arity
m
eas
ur
e
3.
1
.
Ph
as
e
one
In
t
his
phase
,
pr
e
processi
ng
perform
ed
on
bo
t
h
im
ages
(refre
nce
a
nd
in
pu
t
)
to
prepa
re
the
im
ages.
Fu
rt
her
m
or
e,
di
ff
ere
nt
kinds
of
no
ise
a
re a
dd
ed
to
the i
nput
i
m
age f
r
om
a variet
y of
sourc
es [20
,
21
,
22
]
.
3.2.
P
h
as
e
tw
o
In
this
phase,
Mod
ify
Histo
gram
Si
m
il
arit
y
m
easur
e
is
m
erg
e
d
with
Fea
ture
Sim
il
arit
y
m
easur
e
to
get a
new h
y
bri
d
m
easur
e (
H
FM
)
as
in
(
11):
(
,
)
=
√
(
,
)
+
(
,
)
(
1
−
)
(11)
H
(
p,q
)
is
Histogram
Si
m
i
la
rity
m
easur
e but
w
it
h
a
sim
ple c
hange
as
fo
ll
ows
:
1
(
,
)
=
√
∑
∑
[
(
−
)
1
ℎ
+
1
]
2
(12)
Evaluation Warning : The document was created with Spire.PDF for Python.
In
t J
Elec
&
C
om
p
En
g
IS
S
N: 20
88
-
8708
A h
y
br
id
im
age simil
ar
it
y m
e
as
ure
ba
se
d o
n a new
c
ombi
nati
on o
f
diff
erent…
(
Ni
sreen R
ya
dh Ha
mza
)
1817
H1
(
p,q)
can
be
norm
al
iz
e
by
us
in
g
2
(
,
)
wh
ic
h
represe
nt
the
m
axi
m
u
m
value
of
t
he
er
r
or
est
i
m
at
e
in
sign
ifca
nt
no
is
e as foll
ows:
(
6
,
)
=
1
(
,
)
2
(
,
)
,
(
,
)
=
1
−
(
,
6
)
(13)
FSIM(
p,q)
is Feat
ur
e
Sim
il
ari
ty
Mea
su
re as
in
e
qu
at
io
n (
8
)
.
K
is
ve
ry sm
all co
nst
ant.
3.
3.
Ph
as
e
th
r
ee s
The
oth
e
r
m
ea
su
re
is
the
Er
r
or
Me
an
Me
as
ur
e
(
,
)
bet
ween
im
age
an
d
(d
e
rive
d
f
r
om
Me
an
S
quare
Error
[
23]
)
as
f
ollows
:
=
1
−
[
1
NM
∑
∑
[
p
(
n
,
m
)
−
q
(
n
,
m
)
]
2
/
(
∑
∑
(
,
)
−
1
=
0
−
1
=
0
)
1
2
]
M
−
1
m
=
0
N
−
1
n
=
0
(14)
By
co
m
bin
in
g
the
res
ulted
m
easur
e
from
phase
two
(
HFM
)
an
d
the
E
rror
Me
an
m
easur
e
(
EMM
)
,
we
wil
l
get
a n
e
w
sim
il
arity
m
easur
e
cal
l
ed (
HF
EMM
)
. T
he
(15
)
is s
um
m
aries the n
e
w
m
easur
e
as
for
m
:
(
,
)
=
(
(
,
)
(
1
−
)
+
(
,
)
(
)
)
1
/
2
(15)
Wh
e
re
U
is
ve
ry
sm
al
l
con
sta
nt
,
0
≤
(
,
)
≤
1
.
Finall
y,
pe
rfor
m
the
c
om
pr
asi
on
s.
Al
gorithm
(1
)
sh
ows
the m
eth
od a
nd
pro
posed
m
easur
e
(
HFEMM
) for s
i
m
i
la
rity
.
Algo
rit
hm (
1)
HFEMM
Inp
uts: p
is
refr
ence im
age and q is
noisy
i
m
age, K a
nd
U
a
re
ver
y sm
al
l c
on
sta
nt
Ou
t
pu
ts:
a
nu
m
ber
b
et
wee
n 0 a
nd 1 that
re
pr
ese
nt the
sim
il
arit
y.
-
Step
1: tra
ns
f
orm
the co
lo
ur i
m
ages in
t
o gr
a
ysc
al
e i
m
ages.
-
Step
2: tra
ns
f
orm
the i
m
ages in
to
double
.
-
Step
3: Com
pute
1
(
,
)
=
√
∑
∑
[
(
−
)
1
ℎ
+
1
]
2
-
Step
4: Set
H
2
(
,
)
=
H
1(p,q)
wh
e
n no
ise
is m
axi
m
um
.
-
Step
5: No
rm
alizat
ion
:
H
H=H
1/ H
2
.
-
Step
6: Set
H =1
-
HH
.
-
Step
7: Com
pute
FS
IM
(
,
)
=
∑
(
9
)
.
(
)
x
7
∈
ℶ
∑
(
8
)
x
∈
ℶ
-
Step
8: Com
pute
HFM
=
√
(
,
)
×
+
(
,
)
×
(
1
−
)
-
Step
9: Com
pute
EMM
(
,
)
=
1
−
[
1
IJ
∑
∑
[
p
(
n
,
m
)
−
q
(
n
,
m
)
]
2
/
(
∑
∑
(
,
)
−
1
=
0
−
1
=
0
)
1
2
]
M
−
1
m
=
0
N
−
1
n
=
0
-
Step
10
: C
om
pu
te
HFEM
M
(
,
)
=
(
(
,
)
(
1
−
)
+
(
,
)
(
)
)
1
/
2
-
Step
11
:
Per
for
m
the co
m
par
isons (e
val
uate
the r
es
ults)
-
Step
12
:
En
d o
f
Al
gorithm
Evaluation Warning : The document was created with Spire.PDF for Python.
IS
S
N
:
2088
-
8708
In
t J
Elec
&
C
om
p
En
g,
V
ol.
10
, No
.
2
,
A
pr
i
l 202
0
:
1814
-
1822
1818
4.
RESU
LT
S
AND DI
SCUS
S
ION
S
The
pro
pose
d
m
eaur
es
i
m
ple
m
ented
us
in
g
'M
ATLA
B'
env
i
ronm
ent
and
a
hu
m
an
fa
ce
database
cal
le
d
AT
and
T
was
us
e
d
in
te
sti
ng
the
pro
po
s
ed
m
et
ho
ds.
More
ov
e
r,
di
ff
e
ren
t
ki
nd
s
of
rand
om
i
m
ag
e
are
al
so
use
d
f
or
the
eval
uation.
The
range
o
f
si
m
il
arity
m
e
asur
e
s
am
on
g
(0
a
nd
1).
If
va
lue
is
(
1),
th
en
it
s
disp
la
ys
the
i
deal
m
at
ch
between
t
he
im
a
ges,
Else
if
it
s
value
is
(
0),
then
th
ere
is
no
m
at
ch
betwee
n
i
m
ages [
24]
.
4.1
AT&T d
ata b
as
e
The
AT
&T
is
an
Am
erican
T
el
ephon
e
&
Te
le
gr
am
:
Labo
r
a
tories f
r
om
Ca
m
br
idg
e
c
om
pr
ise
s
a set
of
var
i
ou
s
hum
an
faces
im
ages
wer
e
ta
ken
in
(Apr
il
1992
an
d
A
pri
l
1994)
at
the
data
base
la
b.
It
com
pr
ise
s
of
10 d
issi
m
il
ar i
m
ages (poses)
of eve
ry p
e
rs
on, im
age size (
92
×
112)
pix
el
s
[
2
5]
.
4.2.
Ev
alu
at
io
n t
he
pr
opose
d me
as
ure
on
AT&T d
atase
t
This
first
of
e
va
luati
on
incl
ud
es
i
m
ple
m
entin
g
t
he
sim
i
la
ri
ty
m
easur
e
under
dif
fer
e
nt
ki
nd
s
of
noise
su
c
h
as
Ga
us
si
ans,
sal
t
an
d
pe
pp
e
r
a
nd
un
i
f
or
m
no
ise
.
Dif
fen
t
ty
pes
of
i
m
ages
from
AT
an
d
T
data
ba
se
are
adopted
a
nd
te
ste
d,
f
or
ex
a
m
ple
an
i
m
age
as
in
the
Figure
2
belo
w
was
te
ste
d
us
in
g
Ga
ussi
ans
noise
.
T
able
1
and
Fi
gure
3
are
show
the
si
m
il
arity
resu
lt
s
between
th
e
pr
op
os
e
d
m
easur
e
a
nd
oth
er
m
easur
es
su
c
h
as
(
SSI
M,
E
EM,
HFM
) under
G
aussian
noise
.
In
a
dd
it
io
n,
t
he
seco
nd
te
st
sh
ows
t
he
im
plem
entat
ion
unde
r
sal
t
and
peppe
r
noise
as
show
n
in
Figure
4.
Tabl
e
2
a
nd
Fi
gure
5
s
how
n
sim
ilarity
resu
lt
be
tween
the
pro
pose
d
m
easur
e
and
an
oth
e
r
m
easur
e
s
unde
sal
t
&
pe
pp
e
r
noise
.
S
o
as
to
show
th
e
ste
adiness
a
nd
ade
quacy
of
the
pro
posed
m
easur
e,
we
a
pp
li
e
d
the
pro
pose
d
m
easur
e
under
a
com
bin
at
ion
of
no
ise
s
(
G
aussian
an
d
U
nif
or
m
no
ise
,
Gau
s
sia
n
a
nd
sal
t
and
pe
ppe
r
noise
)
wh
ic
h
will
be
m
or
e
no
isy
,
and
As
sho
w
in
the
fo
ll
wi
ng
Fig
ur
e
6.
T
he
Fig
ur
e
7
s
hows
the
pe
rfo
rm
ance
unde
r
Ga
ussi
an
no
ise
a
nd
sla
t
an
d
pepp
er
no
ise
Table
3
s
how
n
sim
i
la
rity
resu
lt
be
tween
the
pro
pose
d
m
easur
e
a
nd
a
no
t
her
unde
r
Gau
s
sia
n
a
nd
Un
i
for
m
no
ise
.
Table
4
s
ho
w
the
sim
il
arity
resu
lt
betwee
n
the
pr
opos
e
d
m
easure
and a
no
t
her u
nd
e
r
Ga
us
sia
n
and salt
and
pe
pp
e
r n
oise.
4.3.
Ev
alu
at
io
n t
he
pr
opose
d me
as
ure
by
using di
ffer
en
t
kind
s
of
im
ages
In
t
his
sect
io
n,
we
us
ed
dif
fe
ren
t
kinds
of
i
m
age
to
d
is
pla
y
the
eff
ect
ive
ness
a
nd
ef
fici
ency
of
that
the
pr
opos
e
d
m
easur
e
.
As
il
lustr
at
ed
i
n
Fi
gures
8
-
10.
C
om
par
isons
of
si
m
il
arity
m
e
asur
e
s
un
der
Gau
s
sia
n
and unif
or
m
noise a s
how
n
i
n
Fi
gure
11
Figure
2. O
rigi
nal im
age an
d
no
isy
im
age w
i
th G
a
us
sia
n noi
se
Table
1.
C
om
par
iso
ns
of sim
i
la
rity
m
easur
es
for
s
am
e i
m
ages un
der
G
a
us
s
ia
n
noise
(H
i
gh
e
r no
ise
ra
ti
o
100%
wh
e
n psnr =
100)
Ratio
of
Nois
e
S
S
IM
EMM
HFM
HFE
MM
6
5
%
(PSNR
-
50)
0
.00
2
0
0
.00
0
1
0
.30
7
2
0
.42
9
3
5
2
% (PSNR
-
30)
0
.00
2
0
0
.00
0
8
0
.29
9
0
0
.44
6
5
3
8
% (PSNR
-
10)
0
.00
3
4
0
.08
3
0
0
.39
2
4
0
.53
8
3
3
2
% (PSNR 0)
0
.01
4
1
0
.21
5
5
0
.56
0
5
0
.68
9
1
2
8
% (PSNR 10
)
0
.06
3
3
0
.54
6
9
0
.78
7
1
0
.88
2
4
1
% (PSNR 50
)
0
.98
3
1
0
.99
9
9
0
.99
9
6
1
.00
0
0
Evaluation Warning : The document was created with Spire.PDF for Python.
In
t J
Elec
&
C
om
p
En
g
IS
S
N: 20
88
-
8708
A h
y
br
id
im
age simil
ar
it
y m
e
as
ure
ba
se
d o
n a new
c
ombi
nati
on o
f
diff
erent…
(
Ni
sreen R
ya
dh Ha
mza
)
1819
Figure
3
.
Com
par
is
ons
of
sim
il
arit
y
m
easur
es for sam
e i
m
a
ges
u under
G
a
us
sia
n
noise
Figure
4. O
rigi
nal i
m
age an
d
no
isy
im
age w
i
th salt
and
pe
pper
noise
Table
2.
C
om
par
iso
ns
of sim
i
la
rity
m
easur
es
for
s
am
e i
m
ages un
der
sal
t and pe
pper
nois
e
Ratio
of
Nois
e
S
S
IM
EMM
HFM
HFE
MM
2
7
%
(PSNR 9.9
6
5
0
)
0
.09
3
0
0
.65
1
1
0
.77
3
5
0
.85
1
2
2
2
% (PSNR 15
.37
6
5)
0
.29
6
3
0
.89
7
2
0
.91
7
6
0
.95
3
6
1
8
% (PSNR 20
.42
7
9
)
0
.58
3
5
0
.96
7
6
0
.97
1
6
0
.98
4
9
1
6
% (PSNR 23
.71
3
0
)
0
.72
8
9
0
.98
4
8
0
.98
5
4
0
.99
2
5
1
2
% (PSNR 28
.89
8
3
)
0
.90
6
7
0
.99
5
4
0
.99
5
7
0
.99
7
8
8
% (PSNR 32
.41
6
8
)
0
.95
2
7
0
.99
8
4
0
.99
83
0
.99
9
1
Figure
5. Com
par
is
ons
o
f si
m
il
arit
y
m
easur
es for sam
e i
m
a
ges u
nd
e
r
i
m
pu
lse
no
ise
Evaluation Warning : The document was created with Spire.PDF for Python.
IS
S
N
:
2088
-
8708
In
t J
Elec
&
C
om
p
En
g,
V
ol.
10
, No
.
2
,
A
pr
i
l 202
0
:
1814
-
1822
1820
Figure
6.
(
P
) O
rigin
al
im
age an
d n
oisy i
m
age (q)
c
om
par
iso
ns
of sim
i
la
rity m
easur
es un
de
r
Gau
s
sia
n+
un
i
f
or
m
n
oise
Figure
7.
(P)
ori
gin
al
im
age an
d n
oisy i
m
age (q)
c
om
par
iso
ns
of sim
i
la
rity m
easur
es un
de
r
gaussia
n+salt
a
nd p
e
pper
nois
e
Table
3.
C
om
par
iso
ns
of sim
i
la
rity
m
easur
es
for
s
am
e i
m
ages un
der
G
a
us
s
ia
n
an
d u
nif
orm
n
oise
Ratio
of
Nois
e
S
S
IM
EMM
HFM
HFE
MM
6
5
%
(PSNR
-
50)
0
.00
0
5
0
.00
0
0
0
.33
9
1
0
.45
1
0
5
2
% (PSNR
-
30)
0
.00
0
0
0
.02
0
7
0
.35
1
0
0
.46
7
9
3
8
% (PSNR
-
10)
0
.00
5
3
0
.10
6
9
0
.44
3
1
0
.55
5
5
3
2
% (PSNR 0)
0
.02
1
2
0
.31
4
6
0
.60
1
6
0
.69
7
7
2
8
% (PSNR 10
)
0
.08
8
5
0
.73
2
2
0
.82
3
1
0
.88
7
0
1
0
% (PSNR 30
)
0
.46
3
8
0
.97
2
7
0
.95
4
7
0
.98
2
3
2
% (PSNR 50
)
0
.49
4
5
0
.98
3
4
0
.95
8
0
0
.98
4
0
Table
4.
C
om
par
iso
ns
of
sim
i
la
rity
m
easur
es
for
s
am
e
i
m
ages un
der
Ga
us
s
ia
n
an
d sl
at
and
peppe
r no
ise
Ratio
of
no
ise
S
S
IM
EMM
HFM
HFE
MM
5
2
% (PSNR
-
30)
0
.00
0
0
0
.01
1
8
0
.35
9
6
0
.46
9
5
3
8
% (PSNR
-
10)
0
.00
2
0
0
.00
3
9
0
.34
6
9
0
.45
9
9
3
2
% (PSNR 0)
0
.00
7
2
0
.10
5
0
0
.46
3
0
.56
5
5
2
8
% (PSNR 10
)
0
.02
9
0
0
.29
7
3
0
.62
8
0
0
.70
4
1
8
% (PSNR 50
)
0
.10
4
5
0
.73
1
6
0
.84
3
0
0
.89
3
5
1
% (PSNR 50
)
0
.72
9
5
0
.99
6
4
0
.98
4
9
0
.99
4
7
6
5
%
(PSNR
-
50)
0
.97
7
4
0
.99
9
9
0
.99
9
6
1
.00
0
0
Evaluation Warning : The document was created with Spire.PDF for Python.
In
t J
Elec
&
C
om
p
En
g
IS
S
N: 20
88
-
8708
A h
y
br
id
im
age simil
ar
it
y m
e
as
ure
ba
se
d o
n a new
c
ombi
nati
on o
f
diff
erent…
(
Ni
sreen R
ya
dh Ha
mza
)
1821
Figure
8. (P)
ori
gin
al
im
age an
d n
oisy i
m
age
(
q)
Figure
9. Com
par
is
ons
of
sim
il
arit
y
m
easur
es
unde
r
Gau
s
sia
n
a
nd
s
la
t and
pe
pper
no
ise
Figure
10. Ori
gin
al
Im
age and no
isy
im
age
Figure
11. C
om
par
ison
s
of si
m
il
arity
m
easur
es
unde
r
Gau
s
sia
n
a
nd
unif
or
m
n
oise
5.
CONCL
US
I
O
N
A
hy
br
i
d
sim
ilarity
m
easur
e,
cal
le
d
Histo
gram
Feat
ur
e
Error
Me
a
n
M
e
asur
e
,
ha
s
bee
n
pr
opos
e
d.
The
pro
posed
m
easur
e
is
de
pe
nd
e
d
on
in
f
or
m
at
ion
the
or
et
ic
fe
at
ur
es
an
d
s
ta
ti
sti
ca
l
featur
e
s
.
A
joinit
histo
gr
am
'
with
ori
gin
al
histo
gr
a
m
hav
e
bee
n
us
e
d
as
inf
orm
at
ion
-
the
or
et
ic
too
l,
an
d
F
eat
ure
Me
asur
e
with
Error
Me
a
n
ha
ve
bee
n
us
e
d
as
a
sta
ti
st
ic
al
too
l.
The
pro
po
s
e
d
m
easur
e
has
bee
n
te
ste
d
on
AT
an
d
T
a
nd
dif
fere
nt
ty
pe
s
of
im
ages.
We
co
nclu
de
d
that
the
ne
w
m
easur
e
ga
ve
bette
r
pe
rform
ance
(m
or
e
si
m
i
la
rit
y)
than
the
ot
her
sim
i
la
rity
m
easur
es
su
c
h
as
(
SSIM
,
E
MM
)
unde
r
di
ff
e
re
nt
ki
nd
s
of
no
ise
wh
e
n
power
of
no
ise
is
hi
gh
l
y
hig
h.
The
propose
d
m
easur
e
can
be
us
ed
i
n
a
f
undam
ental
issue
in
real
-
w
or
l
d
app
li
cat
io
ns
.
Su
c
h
as
ca
n
be
em
plo
ye
d
in
f
ound
sim
l
arit
y
and
dif
f
eren
t
betwee
n
i
m
age,
ve
rifi
cat
ion
,
recog
niti
on
(f
a
ce, iris, a
nd
othe
r patt
ern
rec
ogniti
on syst
em
s).
ACKN
OWLE
DGE
MENTS
We
would
li
ke
to
tha
nks
de
par
tm
ent
of
com
pu
te
r
sci
e
nce
in
fac
ulty
of
ed
ucati
on
for
girlss
i
n
un
i
ver
sit
y
of
kufa
,
a
nd
Fac
ulty
of
Com
pu
te
r
Scie
ncec
an
d
inf
orbm
ation
t
echnolo
gy,
U
ni
ver
sit
y
of
Qa
di
siy
ah
for
thei
r
s
uppo
rt.
Evaluation Warning : The document was created with Spire.PDF for Python.
IS
S
N
:
2088
-
8708
In
t J
Elec
&
C
om
p
En
g,
V
ol.
10
, No
.
2
,
A
pr
i
l 202
0
:
1814
-
1822
1822
REFERE
NCE
S
[1]
S.
Farisa,
C.
H
avi
an
a,
and
D.
Kurniadi
,
“
Aver
age
Hashing
for
Perc
ept
ua
l
Im
age
Sim
il
ari
t
y
in
Mobile
Phone
Applic
a
ti
on,
”
J.
Tele
mat.
Informati
cs
,
vol
.
4
,
no
.
1,
pp
.
12
–
18
,
20
16.
[2]
D.
Li
n
,
“
An Info
rm
at
ion
-
The
or
etic
Def
ini
t
ion
of
Sim
il
ari
t
y
,
”
Icml
,
pp
.
296
–
304
,
1
998.
[3]
K.
Shin,
“
An
a
lterna
t
ive
appr
oa
c
h
to
m
ea
sure
si
m
il
ari
t
y
bet
we
en
two
de
te
rm
ini
sti
c
tr
ansie
nt
signa
l
s,”
J
.
Sound
V
ib.
,
vol.
371
,
pp
.
434
–
445,
2016
.
[4]
A.
F.
Hass
an,
D.
Cai
-
l
in,
and
Z.
M.
Hus
sain,
“
A
n
informati
on
-
th
eor
etic
image
qu
al
ity
m
e
asure
:
C
om
par
ison
with
stat
isti
ca
l
sim
il
ar
ity
,
”
J. Comput.
Sci
.
,
vo
l. 10, no.
11,
pp
.
2269
–
22
83,
2014
.
[5]
P.
A.
Viola,
“
m
assac
husett
s
inst
it
ute
o
f
technol
og
y
al
ignment
b
y
Maximi
za
t
io
n
of
Mutual
Inf
orm
at
ion,
”
Arti
f
.
Inte
ll.
,
no.
1548,
1995.
[6]
Zhou
W
ang
and
I.
Alan
C.
Bovik,
Fell
ow,
“
Im
age
Quali
t
y
As
sessment:
Fr
om
Err
or
Visib
il
ity
to
Struc
tur
al
Sim
il
ari
t
y
,
”
IEEE
Tr
ans.
I
mage
Proce
ss
.
,
vol. 13
,
no
.
4
,
pp
.
1
–
14
,
2004.
[7]
A.
F.
Hass
an,
“
An
Inform
at
ion
-
The
ore
ti
c
Me
asure
for
Face
R
ecogniti
on :
Com
p
ari
son
with
Stru
ct
ura
l
Sim
il
arit
y,
”
Int.
J. A
d
v. R
es
.
Arti
f. Intell.
,
vol
.
3,
no.
11,
pp.
7
–
13,
2014
.
[8]
R.
A.
Dihin
and
Z.
M.
Hus
sain
,
“
Face
Re
cognitio
n
over
FF
T
-
OF
DM
Com
pute
r
Network
s,”
Int
.
J.
Appl
.
Eng.
Res
.
,
vol.
12
,
pp
.
1476
4
–
14773,
2017
.
[9]
Akash,
O.
M;
Ahm
ad,
Sharifa
h
Sakina
h
Bin
t
i
Sy
ed
;
Azm
i
,
Mohd
Sanusi
Bin,
“
A
New
Sim
il
ari
ty
Mea
sur
e
Base
d
Affinity
Propag
at
ion
for
Dat
a
Cluste
ring
,
”
Ad
vanc
ed
Sci
ence
Le
tters,
Volum
e
24,
Num
ber
2,
p
p.
1130
-
1133(4
),
Februa
r
y
2018
.
[10]
M.
A.
Alja
nabi,
N.
A.
Shnain,
and
S.
F.
Lu,
“
A
n
image
sim
il
arit
y
m
ea
sure
base
d
on
joi
nt
histogra
m
-
Ent
rop
y
fo
r
fac
e
re
cogni
t
ion
,
”
2017
3rd
IEE
E
Int.
Conf
.
Comput.
Commun.
ICCC
2017
,
vol.
2018
-
Janua
,
no.
Dec
ember,
pp.
1626
–
16
31
,
2018.
[11]
Mohammed
Ab
dula
m
ee
r
Alja
n
abi
,
Hua
zhong
and
W
uhan,
“
D
esign
of
a
h
y
br
id
m
ea
sure
for
image
sim
il
arit
y:
a
sta
ti
sti
ca
l
,
al
g
e
bra
ic, and
in
for
m
at
ion
-
the
or
et
i
c
appr
oa
ch
,
”
European
Journal
o
f
R
emote Se
nsin
g
,
2019
.
[12]
D.
Brunet,
E
.
R.
Vrs
cay
,
and
Z.
W
ang,
“
On
the
m
a
the
m
at
i
ca
l
prope
rt
ie
s
o
f
the
stru
ct
ur
al
sim
il
ari
t
y
index
,
”
IEE
E
Tr
ans.
Image
Proc
ess
.
On,
vol. 21, no. 4, p
p.
1488
–
1499
,
2
012.
[13]
R.
Dos
selmann
and
X.
D.
Yan
g,
“
A
Form
al
As
sessment
of
th
e
Struct
ur
al
Si
m
il
ari
t
y
Ind
ex,”
Te
chn
ical
R
ep
ort
TRCS 2008
-
2
S
e
pte
m
ber
,
2008
.
[14]
L.
Zh
ang,
L
.
Zh
ang,
X.
Mou,
a
nd
D.
Zha
ng
,
“
FS
IM:
A
fea
tur
e
sim
il
arit
y
ind
e
x
for
image
qu
a
li
t
y
assess
m
ent
,
”
IEE
E
Tr
ans.
Image
Proc
ess.
,
vol. 20, no. 8, pp. 23
78
–
2386,
2011
.
[15]
U.
Sara
,
M.
Ak
te
r,
and
M.
S.
Uddin,
“
Im
age
Quali
t
y
As
sess
m
ent
through
F
SIM
,
SS
IM,
M
SE
and
PS
NR
-
A
Com
par
at
ive Stu
d
y
,
”
J. Comput.
Comm
un.
,
vol
.
0
7,
no
.
03
,
pp
.
8
–
18,
2019
.
[16]
F.
M.
Altuf
ai
l
i,
“
A
Corre
la
ti
v
e
I
nform
at
ion
-
Theoret
i
c
Mea
sure
f
or
Im
age
Sim
il
a
rity
,
”
Global
Jo
urnal
of
Pure
an
d
Appl
ie
d
Math
emati
cs
,
vol
.
13
,
no
.
2
,
pp
.
15
9
–
169
,
2017.
[17]
R.
Soundara
r
aja
n
and
A.
C.
Bovik,
“
Surve
y
of
informat
ion
the
or
y
in
visu
al
qu
al
i
t
y
asses
sm
ent
,
”
Signa
l,
Image
Vi
d
eo Proce
ss
.
,
vol. 7, no
.
3
,
pp
.
391
–
401
,
2013.
[18]
T.
Ka
la
ise
lvi,
P.
Sriramakri
shna
n,
and
K.
Som
asunda
ram,
“
Brai
n
abnor
m
al
i
t
y
de
t
ec
t
io
n
from
MRI
of
hum
an
he
ad
sca
ns
using
the
bil
atera
l
s
y
m
m
et
r
y
p
rope
rt
y
&
histogra
m
sim
i
la
rity
m
ea
sures
,
”
Int.
Comput.
Sci
.
Eng
.
Conf.
,
no.
oc
tope
r
2016
,
pp
.
1
–
6
,
2016
.
[19]
N.
R.
Ham
za
,
H
.
R.
Moham
m
ed,
and
Z
.
M.
Hus
sain,
“
A
joi
nt
hi
stogram
-
2D
cor
rel
a
ti
on
m
ea
sur
e
for
inc
om
plete
image
sim
il
ar
ity
,
”
Int
.
J. A
ppl
.
En
g.
R
es.
,
vol
.
12
,
no.
22
,
pp
.
1275
5
–
12761,
2017
.
[20]
Charl
es
Boncel
et
,
"Im
age
Noi
se
Models
,
"
In
Alan
C.
Bovik.
Handbook
of
Im
age
and
V
ide
o
Proce
ss
ing.
Aca
demic
Press
,
2005.
[21]
Abd
Al
-
sala
m
S
el
ami,
Am
ee
n
&
Fadhi
l,
Ahm
ed.
“
A
Stud
y
of
th
e
Eff
e
ct
s
of
G
aussian
Noise
on
Im
age
Feat
ure
s,
”
Kirkuk
Univer
sit
y
Journ
al
/
Sc
ie
nt
i
fic
Stu
d
ie
s (199
2
-
0849).
11
.
152
–
169,
2016
.
[22]
F.
M.
Altufa
il
i
,
H.
R.
Mohammed,
Z.
M.
Hus
sain,
“
A
Noise
-
Resista
nt
H
y
b
rid
Mea
sure
for
Im
age
s
Sim
il
ari
t
y
”
,
20
16.
[23]
Z.
W
ang
and
A.
C.
Bov
ik,
"M
ean
square
d
err
or:
love
it
or
l
ea
ve
i
t
?
A
n
ew
look
at
signal
fidelity
m
ea
sures,"
IEEE
Signal
Proce
ss
in
g
Magazine
,
200
9
.
[24]
T.
Kalaisel
vi
,
P.
Sriramakri
shnan,
and
P.
Naga
raj
a
,
“
Brai
n
Tumor
Boundar
y
De
te
c
ti
on
b
y
Edg
e
Indic
ation
Ma
p
Us
ing
Bi
-
Modal
Fuzz
y
Histogra
m
Thre
sholding
Te
chn
ique
from
MRI
T2
-
W
ei
ghte
d
Scans,
”
Int
.
J
.
Image,
Gr
aph.
Signal
Proce
ss
.
,
vol.
8
,
no
.
9
,
pp
.
51
–
59,
2016
.
[25]
AT
&
T
La
bora
tories
Cambridge,
“
AT
&
T
f
ac
e
da
ta
base
(f
orm
erly
‘th
e
ORL
Data
base
of
Face
s’)
”
,
2002
,
htt
p://ww
w.c
l.cam
.
ac
.
uk
/re
sea
rch
/dt
g/attarchi
v
e/ f
ac
ed
at
ab
ase
.
h
tml
Evaluation Warning : The document was created with Spire.PDF for Python.