Internati
o
nal
Journal of Ele
c
trical
and Computer
Engineering
(IJE
CE)
V
o
l.
6, N
o
. 3
,
Ju
n
e
201
6, p
p
. 1
161
~ 11
67
I
S
SN
: 208
8-8
7
0
8
,
D
O
I
:
10.115
91
/ij
ece.v6
i
3.8
383
1
161
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
/
IJECE
Performance Study of Soft Loca
l Binary Pattern over Local
Binary P
a
tt
ern u
nder Noi
s
y Im
ages
S
a
b
i
na
Ya
s
m
in
,
M
d
.
Ma
s
ud
Ra
na
Department of
Electrical
and
E
l
ectron
i
c Engin
e
e
r
ing,
Rajshahi Univer
sity
of
Engineering &
Technolog
y
,
Rajshahi-620
4, Bang
lad
e
sh
Article Info
A
B
STRAC
T
Article histo
r
y:
Received Aug 1, 2015
Rev
i
sed
No
v 9, 201
5
Accepted Nov 25, 2015
In this paper, the performance of soft
local bin
a
r
y
pattern (SLBP)
method has
been invest
igat
e
d
with edge det
e
ction t
echniqu
es for face re
cogni
tion in cas
e
of nois
y
condition. Various edg
e
detec
tion tech
niques such as Cann
y
,
Robert
and Log m
e
tho
d
s
have be
en u
s
ed with S
L
BP
for recogn
izing
faces
. Th
e
results obtain
e
d using SLBP
with vari
ous edge detection for nois
y
condition
based on image quality
meas
urement shows better rec
ognition rate compared
to the r
e
sults obtain
e
d using local b
i
nar
y
pattern (LBP). Simplified
edg
e
dete
ction
m
e
tho
d
s sim
p
lif
y
th
e i
m
a
ges as a r
e
sult SLBP with ed
ge det
e
c
tio
n
requires
less co
m
putation
tim
e
c
o
m
p
ared with
ed
ge de
te
ction
of
L
B
P.
Keyword:
Local Bina
ry
P
a
ttern
Feature
Ext
r
action
Soft
Local Bi
n
a
ry
Pattern
Qu
ality Measurem
en
t
Filterin
g
Met
h
o
d
s
Copyright ©
201
6 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
:
Sabi
na
Ya
sm
i
n
,
Depa
rtm
e
nt of
Electrical an
d Electronic
Engineering,
R
a
jsha
hi
Uni
v
ersi
t
y
of
En
gi
n
eeri
n
g &
Tec
h
nol
ogy
,
R
a
jsha
hi
62
0
4
,
B
a
n
g
l
a
des
h
.
Em
a
il: syas
mi
n
.
ru
et@g
m
a
il.
co
m
1.
INTRODUCTION
Face recognition face
detection a
nd facial expressi
on rec
o
gnition
have
been a
n
intere
sting area of
researc
h
ove
r
the past fe
w years as a res
u
lt of its gr
owi
ng
usa
g
e i
n
m
a
ny
appl
i
cat
i
o
ns i
n
fi
el
ds s
u
ch as
ai
rl
i
n
es,
ban
k
i
n
g
,
sec
u
ri
t
y
, m
u
l
t
i
m
e
di
a appl
i
cat
i
ons et
c.
Ea
ch i
n
di
vi
dual
c
a
n
be i
d
e
n
t
i
f
i
e
d a
n
d
ve
ri
fi
ed
by
t
h
i
s
m
e
thod. Actually
Face
Recognition
syste
m
works by com
p
aring
unknown facial im
age with known
in
d
i
v
i
d
u
a
ls from
a larg
e d
a
tab
a
se and
then th
e system
return
s a
v
a
lu
e of
si
m
ilarit
y
m
e
a
s
u
r
em
en
t b
e
t
w
een
t
h
e
two
im
ag
es [1]-[9
].
Th
e Facial i
m
ag
e p
r
ep
ro
cessi
n
g
is
v
e
ry im
p
o
r
tant p
a
rt
o
f
face reco
gn
itio
n or face
det
ect
i
on sy
st
em
. One of t
h
e vi
t
a
l
pre
p
r
o
cessi
n
g
t
a
sk
s is i
m
ag
e d
e
no
ising
.
Im
age preprocessing is the
t
echni
q
u
e
o
f
e
nha
nci
n
g
dat
a
im
ages p
r
i
o
r
t
o
com
put
at
i
onal
p
r
oce
ssi
n
g
[2]
.
T
h
ere
are
di
ff
erent
t
y
pes
of
i
m
age
n
o
i
ses
wh
ich
deg
r
ad
e th
e im
a
g
e qu
ality an
d
also
h
a
v
e
th
e filterin
g
tech
n
i
qu
es fo
r rem
o
v
i
n
g
tho
s
e no
ises. An
efficien
t
d
e
n
o
i
sin
g
techn
i
qu
e shou
ld
co
m
p
letely re
m
o
v
e
no
ises
with
ou
t affecting
t
h
e imag
e
q
u
ality. Also
we
h
a
v
e
m
a
n
y
q
u
ality
matrices t
o
m
easu
r
e im
a
g
e qu
ality
[2
]-[4
]
. MSE, PSNR, an
d
SS
IM are so
m
e
u
s
efu
l
and
m
o
st co
mm
o
n
l
y u
s
ed im
ag
e q
u
a
lity m
e
trics in
im
ag
e p
r
o
c
essin
g
[3
]. The SSIM i
n
d
e
x
can
b
e
v
i
ewed as a
q
u
a
lity
m
easu
r
e o
f
on
e of the i
m
ag
es b
e
in
g
co
m
p
ared
pro
v
i
d
e
d
th
e
o
t
her i
m
ag
e is reg
a
rd
ed
as o
f
perfect
q
u
a
lity [1
0
]
. Usin
g th
e
d
i
fferen
t qu
ality m
e
trics we can
easily
m
easu
r
e th
e
q
u
a
lity after
d
e
no
ising
t
h
e no
isy
im
age and afte
r that
we ca
n c
o
m
p
are the
differe
n
ces am
ong them
.
Vari
ous m
e
thods
ha
ve
bee
n
used
for e
x
tracting th
e
useful
features
from
face im
ages to face
recognition. Am
ong the m
e
thods
LBP is
m
o
st efficient
for
Face rec
o
gnition. T
h
e l
o
cal
bina
ry pattern
(LBP
)
t
h
at
i
s
i
n
vari
a
n
t
t
o
m
onot
oni
c
g
r
ay
-scal
e t
r
a
n
sf
orm
a
t
i
ons
whi
c
h i
s
very
im
port
a
nt
f
o
r t
e
xt
u
r
e a
n
al
y
s
i
s
.
W
i
t
h
LBP it is po
ssib
le to
d
e
scri
be th
e tex
t
u
r
e an
d shap
e of
a
di
gi
t
a
l
i
m
age [1]
-
[
2
]
.
T
h
i
s
i
s
do
ne
by
di
vi
di
ng
a
n
i
m
ag
e in
to
several sm
all reg
i
o
n
s
fro
m
wh
ich
th
e feat
u
r
es
are ex
tracted (Fig
ure
1
)
.
Face reco
gn
itio
n syste
m
Evaluation Warning : The document was created with Spire.PDF for Python.
I
S
SN
:
2
088
-87
08
IJEC
E
V
o
l
.
6,
No
. 3,
J
u
ne 2
0
1
6
:
11
6
1
– 11
67
1
162
usi
n
g L
o
cal
B
i
nary
Pat
t
e
r
n
(L
B
P
) was i
n
t
r
o
d
u
ced i
n
1
9
96
b
y
Ojal
a et
al
. [
11]
. T
h
e LB
P
ope
rat
o
r i
s
o
n
e
of t
h
e
best
per
f
o
rm
i
ng t
e
xt
u
r
e desc
r
i
pt
ors a
nd i
t
ha
s been wi
del
y
use
d
i
n
vari
ou
s appl
i
cat
i
ons
. Thi
s
o
p
erat
or
wo
rk
s
wi
t
h
t
h
e ei
ght
nei
g
hb
o
r
s o
f
t
h
e cent
e
r
pi
x
e
l
consi
d
eri
n
g
as a t
h
res
h
ol
d [
8
]
-
[
1
1]
. Th
e LB
P i
s
i
nva
ri
ant
t
o
m
onot
oni
c
gra
y
-scal
e t
r
ansf
o
r
m
a
ti
ons w
h
i
c
h i
s
very
i
m
por
t
a
nt
fo
r t
e
xt
u
r
e
anal
y
s
i
s
.
W
i
t
h
LB
P i
t
i
s
poss
i
bl
e t
o
d
e
scri
b
e
th
e t
e
x
t
ure and
shap
e of a
d
i
g
i
t
a
l i
m
ag
e [8
]-[9
]. LBP
h
a
s
b
een
im
p
r
ov
ed
with m
u
ltis
cale an
d
di
ffe
re
nt
t
e
xt
u
r
e pat
t
e
r
n
s s
u
c
h
as
uni
f
o
rm
, ro
t
a
t
i
on i
n
vari
a
n
t
an
d
rot
a
t
i
o
n i
n
vari
a
n
t
u
n
i
f
or
m
pat
t
e
rns [
11]
-[
14]
.
Th
e m
a
j
o
r
d
i
sad
v
a
n
t
ag
es
o
f
LBP is that it
is no
t rob
u
s
t
th
at is sen
s
itiv
e to
no
ise an
d
an
o
t
h
e
r is a
sm
a
ll ch
an
g
e
i
n
th
e inpu
t im
a
g
e
wou
l
d always cau
se on
ly
a sm
all
chan
ge
i
n
t
h
e
out
put
.
B
u
t
i
n
so
ft
hi
st
og
ram
v
e
rsi
o
n, on
e
pix
e
l typ
i
cally
co
n
t
ribu
tes to
m
o
re th
an
o
n
e
bin
[1
5]
-[
1
6
]
.
So
ft Local Binary
Patter
n
is
m
o
re
efficient t
h
an
LBP in ca
se
of
n
o
i
s
y
i
m
ages [15]
-
[
16]
.
In
ou
r st
u
d
y
w
e
have c
o
m
p
ared di
f
f
ere
n
t
n
o
i
sy
im
ages and after rem
oving the noise usi
ng
differe
n
t
filtering m
e
thods for
both LB
P and SLBP
for
face rec
o
gni
tion. Soft Loca
l Binary Pattern can
rec
o
gniz
e the
faces
while the
i
m
age is nois
y but LBP c
oul
d
not that m
eans
recognition
ra
te is ve
ry poor. In section
2, edge
d
e
tectio
n
techn
i
qu
es, im
ag
e q
u
a
lity m
easu
r
e
m
en
t tech
n
i
qu
es
h
a
v
e
b
e
en d
e
scri
b
e
d
th
at
h
a
s
b
een u
s
ed
wit
h
SLPB
an
d LP
B
m
e
t
hod t
o
m
easure t
h
e
qual
i
t
y
i
n
sect
i
on 3
.
Next
, i
n
Sect
i
on
4 an
d 5, LB
P and SLB
P
m
e
t
h
o
d
s
have
bee
n
des
c
ri
be
d. Fi
n
a
l
l
y
, n
u
m
e
ri
cal
val
i
d
at
i
on o
f
SL
B
P
m
e
t
hod c
o
m
p
ared wi
t
h
LB
P i
s
pr
ovi
d
e
d i
n
Sect
i
on 6.
2.
EDGE DETECTION
TECHNIQUES
Ed
ge det
ect
i
o
n i
s
a basi
c
t
ool
use
d
i
n
i
m
age proc
essi
ng
, basi
cal
l
y
fo
r feat
u
r
e de
t
ect
i
on an
d
ext
r
act
i
o
n,
whi
c
h ai
m
t
o
i
d
ent
i
f
y
poi
nt
s i
n
a
di
gi
t
a
l
im
age whe
r
e
bri
g
ht
n
e
ss of
i
m
age change
s sha
r
pl
y
and
find disc
ontinuities. The purpose of e
dge
de
tection is
signi
ficantly reduci
ng t
h
e am
ount
of
data in a
n
im
age
and
pres
erves the struct
ural
properties for
further im
ag
e processing.
Diffe
r
ent edge
detection techniques are
m
e
nt
i
oned
bel
o
w:
2.
1.
Sobel Oper
ator
The
Sobel
operat
or
per
f
orm
s
a
2-
D spatial gradi
e
nt m
easurem
e
n
t on
an im
age and
so em
phasizes
regi
ons
of
hi
g
h
spat
i
a
l
fre
q
u
e
ncy
t
h
at
cor
r
e
sp
on
d t
o
ed
g
e
s. O
n
e ke
rnel
i
s
sim
p
l
y
t
h
e ot
he
r rot
a
t
e
d b
y
90
o
.
Calculate the edge
s in both
horizont
al and vertical direct
ions
. The
n
c
o
m
b
in
e th
e in
form
at
io
n
in
to
a sin
g
l
e
metric.
2.
2.
Robert Oper
ator
The R
o
bert
s
o
p
erat
or
per
f
o
r
m
s
a sim
p
l
e
, qui
ck t
o
c
o
m
put
e, 2-
D s
p
at
i
a
l
gra
d
i
e
nt
m
easurem
ent
on an
im
age. Pixel values at each
poi
nt in the
output re
pr
ese
n
t
the estim
a
ted absol
u
te
m
a
gnitude
of the
spatia
l
g
r
ad
ien
t
o
f
th
e
in
pu
t im
ag
e at th
at po
in
t
[17
]
.
2.
3.
Prewitt Operator
Th
e
p
r
ewitt
ed
g
e
detecto
r
is
an
ap
pro
p
riate way to
esti
m
a
te
t
h
e m
a
g
n
itu
d
e
an
d
orien
t
ation
of an
ed
g
e
. Th
e prewitt o
p
e
rator is li
m
i
ted
to
8
possib
l
e orien
t
a
tio
n
s
. Th
is grad
ien
t
b
a
sed
ed
ge d
e
tecto
r
is estimated
in
th
e
3
x
3
neig
hb
our
hoo
d
f
o
r
eigh
t d
i
r
e
ctio
n
s
.
A
ll the eig
h
t
co
nvolu
tio
n
m
a
sk
s
ar
e calcu
lated. O
n
e
con
v
o
l
u
t
i
o
n m
a
sk i
s
t
h
en
sel
e
ct
ed,
nam
e
l
y
t
h
at
wi
t
h
t
h
e
l
a
r
g
est
m
odul
e [
1
7
]
.
2.
4.
Canny Operator
Canny operat
or is based on three criteria. T
h
e
Cann
y Ed
ge Detectio
n
Sm
o
o
t
h
e
s th
e imag
e with
a
Gau
s
sian
filter. Th
en co
m
p
u
t
e th
e
grad
ien
t
mag
n
itu
d
e
an
d orien
t
atio
n
u
s
in
g fi
n
ite-d
ifferen
ce app
r
o
x
i
matio
ns
fo
r t
h
e pa
rt
i
a
l
deri
vat
i
v
e
s
.
An
d fi
nal
l
y
appl
y
no
n-m
a
xi
m
a
supp
ressi
o
n
t
o
t
h
e
gra
d
i
e
nt
m
a
gni
t
ude,
use t
h
e
d
oub
le threshold
i
n
g
algo
rithm
to
d
e
tect ed
ges [1
8
]
.
2.
5.
LoG Oper
ator
The Lapl
aci
an m
e
t
hod sea
r
c
h
es f
o
r zer
o cr
ossi
n
g
s i
n
t
h
e
seco
nd
deri
vat
i
v
e
of t
h
e i
m
age to fi
n
d
edge
s. Laplaci
an of
Gaus
sian m
e
thod com
b
ines Ga
ussi
a
n
filtering wit
h
the Lapl
acian for edge det
ection.
After calcu
latin
g seco
nd
-order
d
e
riv
a
tive
of an
im
ag
e, th
e v
a
lu
e of a po
i
n
t wh
ich is
g
r
eater th
an
a sp
ecified
th
resh
o
l
d
an
d
o
n
e
of its
n
e
igh
bors is less than
th
e
n
e
g
a
tiv
e of th
e th
resho
l
d
is called zero-cro
ssing
[1
8
]
.
3.
IMA
G
E Q
U
A
L
ITY ME
AS
URE
MENT
In
recen
t
years, a lo
t o
f
d
e
v
e
l
o
p
m
en
ts h
a
v
e
b
een
m
a
d
e
to
measu
r
e im
ag
e
q
u
a
lity th
at co
rrelate with
p
e
rcei
v
e
d
qu
ality. I
m
ag
e q
u
a
lity is a ch
aracteristic o
f
an
im
ag
e th
at
m
easu
r
es th
e p
e
rceiv
e
d
i
m
ag
e
Evaluation Warning : The document was created with Spire.PDF for Python.
I
J
ECE
I
S
SN
:
208
8-8
7
0
8
Perf
or
ma
nce
S
t
udy
of
Sof
t
L
o
cal
Bi
n
a
ry
Pat
t
ern
over
Loc
a
l
Bi
nary
P
a
t
t
e
rn
u
nde
r N
o
i
s
y
..
.
.
(
S
abi
na
Y
a
sm
i
n
)
1
163
d
e
gr
ad
atio
n
[
4
]. Th
is d
e
g
r
ad
atio
n
of
im
ag
e
b
y
no
ise is
some r
a
n
d
o
m
er
r
o
r. Th
is
N
o
ise
co
u
l
d
b
e
add
e
d
along
with
th
e im
ag
e d
u
ring
cap
t
u
ri
n
g
, transm
i
ttin
g
or during
th
e p
r
o
cessing
[6
]
.
W
e
h
a
v
e
applied
d
i
fferen
t
no
ises
in
im
ag
es
an
d rem
o
v
e
d
t
h
e no
ises u
s
ing
d
i
fferen
t
f
iltering tech
n
i
qu
es and
co
m
p
are usin
g MSE, PSNR, and
SSIM
wh
ich
are co
mm
o
n
l
y u
s
ed
im
ag
e q
u
ality
m
e
trics in
imag
e pro
cessin
g
.
3.
1.
Mean Square
Error (MSE
)
Thi
s
m
easurem
ent
i
s
used t
o
com
put
e an erro
r si
gnal
by
s
ubt
ract
i
ng t
h
e t
e
st
si
gnal
fr
om
t
h
e ori
g
i
n
al
,
and t
h
en c
o
m
puting t
h
e ave
r
a
g
e ene
r
gy
of t
h
e error.
T
h
e
m
ean squa
re e
r
r
o
r
(M
SE) is
the sim
p
lest, and t
h
e
m
o
st widely used im
age qua
lity
m
easurem
ent. MSE
for
two
P×Q m
onochrom
e im
ages,
A is t
h
e
original
im
age and B is the distorte
d im
age and also m
,
n are
t
h
e wi
dt
h a
nd
hei
ght
o
f
t
h
e i
m
age. T
h
en t
h
e
M
S
E i
s
d
e
f
i
n
e
d as th
e fo
llo
w
i
n
g
[3
],
[7].
2
11
1
[(
,
)
(
,
)
]
Q
P
mn
M
SE
A
m
n
B
m
n
PQ
(1)
3.
2.
Ro
ot
Mea
n
Sq
uare Err
o
r
(R
MSE)
The R
oot M
e
a
n
S
q
uare
Er
ro
r
(RM
S
E) is
the
squ
a
re
r
oot
of
M
S
E [
3
]
,
[
7
]
.
2
11
1
[(
,
)
(
,
)
]
Q
P
mn
R
MSE
A
m
n
B
m
n
PQ
(
2
)
3.
3.
Peak
Si
g
n
al
to
N
o
i
s
e R
a
ti
o
(
P
SN
R)
Th
e Peak
Signal to
No
ise Ratio
(PSNR) represen
ts
th
e
v
a
lu
e of th
e no
isy i
m
ag
e with
resp
ect to
the
ori
g
inal im
age. The PSNR is evaluate
d in de
cibels and
is inv
e
rsely propo
rt
io
n
a
l with
th
e
Mean
Squ
a
re
Erro
r
is d
e
f
i
n
e
d as the fo
llo
w
i
n
g
[3
],
[7
].
2
10
l
o
g
1
0
L
PSNR
M
SE
(3)
Wh
ere L is th
e d
y
n
a
m
i
c range of th
e p
i
x
e
l valu
es.
3.
4.
Struc
t
ur
al Si
milarity
Inde
x Me
asureme
nt (SSIM)
The St
r
u
ct
u
r
al
Si
m
i
l
a
ri
t
y
Ind
e
x M
easu
r
em
ent
(SS
I
M
)
i
s
a
not
her
use
f
ul
m
e
t
hod
fo
r m
easuri
ng t
h
e
si
m
ilarit
y
b
e
tween
two
im
ag
es. Th
e SSIM in
d
e
x
can
b
e
v
i
ewed
as a qu
ality
m
easu
r
e o
f
o
n
e
o
f
th
e imag
es
b
e
ing
co
m
p
ared
p
r
ov
id
ed
th
e o
t
h
e
r im
ag
e i
s
reg
a
rd
ed
as
o
f
p
e
rfect qu
ality. Th
e SSIM is d
e
fin
e
d
as th
e
follo
win
g
[3]
,
[
10]
.
(,
)
[
(,
)
]
[
(
,
)
]
[
(,
)
]
SSIM
x
y
l
x
y
c
x
y
s
x
y
(4)
Whe
r
e
0,
0
and
0
cont
rol the
rela
tive signi
ficance of t
h
e thre
e term
s of the inde
x. T
h
e
l
u
m
i
nance,
co
n
t
rast
an
d st
r
u
ct
ural
c
o
m
pone
n
t
s of
t
h
e i
nde
x.
4.
LOCAL
BINARY PATTE
RN
Face rec
o
gnition
system
using Local Binary Pattern
(LBP)
was introduced in
1996
by Oj
ala et al
.
[8]
-
[
1
2
]
.
T
h
e
LB
P o
p
erat
or i
s
o
n
e
of t
h
e be
st
per
f
o
r
m
i
ng t
e
xt
u
r
e de
scri
pt
ors
an
d i
t
has
been
wi
del
y
u
s
ed i
n
vari
ous
ap
pl
i
c
at
i
ons.
I
n
t
h
i
s
pape
r
we
ha
ve
im
pl
em
ent
e
d di
ffe
re
nt
ed
ge det
ection techniques s
u
c
h
as
sobel,
p
r
ewitt, rob
e
rt
, ca
nn
y
an
d log
as
p
r
e
p
r
o
ces
si
ng
be
f
o
re
ap
pl
y
i
ng
LB
P i
n
o
r
de
r t
o
get
b
e
t
t
e
r pe
rf
orm
a
nce
of
LBP.
The proc
ess
of face rec
o
gnition
using LB
P edge
dete
ction is
give
n i
n
Fi
gure
1.
Evaluation Warning : The document was created with Spire.PDF for Python.
I
S
SN
:
2
088
-87
08
IJEC
E
V
o
l
.
6,
No
. 3,
J
u
ne 2
0
1
6
:
11
6
1
– 11
67
1
164
Fig
u
re
1
.
Pro
c
ess of Face Reco
gn
itio
n using
LBP [1
]
Noi
s
e el
i
m
i
n
at
i
on an
d ed
ge
det
ect
i
on are
u
s
ed as pre
p
roc
e
ssi
ng i
n
LB
P
.
Thi
s
LB
P ope
rat
o
r
wo
rk
s
wi
t
h
t
h
e ei
g
h
t
nei
g
hb
o
r
s o
f
t
h
e cent
e
r pi
xel
con
s
i
d
eri
ng a
s
a t
h
res
hol
d.
If
a nei
g
hb
o
r
pi
xel
has a
hi
g
h
e
r
g
r
ay
val
u
e t
h
a
n
t
h
e
cent
e
r
pi
xel
(
o
r t
h
e sam
e
gray
val
u
e) t
h
a
n
a
one i
s
assi
gne
d t
o
t
h
at
pi
xel
,
ot
her
w
i
s
e zer
o. T
h
e
cent
e
r
val
u
e i
s
const
r
uct
e
d b
y
concat
enat
i
n
g t
h
e
bi
na
ry
n
u
m
b
ers fr
om
top l
e
ft
cor
n
e
r
i
n
cl
oc
kwi
s
e
di
rect
i
o
n
.
And
fin
a
lly the d
ecim
a
l v
a
lu
e is pro
d
u
c
ed
b
y
m
u
ltip
lyin
g
th
e t
h
resho
l
d v
a
lues
with
weigh
t
s
g
i
v
e
n
to
the
cor
r
es
po
n
d
i
n
g
pi
xel
s
a
n
d s
u
m
m
i
ng u
p
t
h
e
re
sul
t
,
i
s
cal
l
e
d
L
B
P co
des
[
15]
as sh
o
w
n
i
n
Fi
gu
re
2.
E
x
am
ple
Thres
hol
d
Weights
Bin
a
r
y
Pattern
= 10
001
11
Deci
m
a
l
= 1 +
16
+
32
+
64
+
1
2
8
=
24
1
Fi
gu
re
2.
C
a
l
c
ul
at
i
ng t
h
e
ori
g
i
n
al
LB
P c
o
d
e
Fo
rm
ally, th
en
th
e resu
lting
LBP can b
e
exp
r
essed
i
n
d
ecimal form
d
e
fin
e
d
b
y
th
e
fo
llo
win
g
1
,
0
,2
P
P
PR
c
c
p
c
P
LBP
x
y
s
i
i
(5
)
whe
r
e i
c
and
i
P
are,
res
p
ect
i
v
e
l
y
,
gray
-l
e
v
el
val
u
es
o
f
t
h
e
c
e
nt
ral
pi
xel
a
n
d P
su
rr
o
u
n
d
i
n
g
pi
xel
s
i
n
t
h
e
ci
rcl
e
nei
g
hb
o
r
h
o
o
d
wi
t
h
a ra
di
us
R
o
f
gi
ve
n pi
xel
at
(x
c
,
y
c
)
,
a
n
d
f
unct
i
o
n s
(
x) i
s
de
fi
ne
d as
[
1
4]
.
1,
0
()
0,
if
x
sx
if
x
(6)
5.
SOFT LOCAL BIN
ARY PATTERN (SLBP)
A drawb
a
ck
o
f
th
e lo
cal b
i
n
a
ry p
a
ttern
is
n
o
t
robu
st th
at is
a sm
a
ll ch
an
g
e
in
th
e inpu
t imag
e wou
l
d
always cau
se
o
n
l
y a sm
a
ll c
h
ang
e
in
th
e
ou
tpu
t
. It is sensitiv
e to
n
o
i
se. Bu
t in
so
ft h
i
sto
g
ram
v
e
rsion
,
on
e
pi
xel
t
y
pi
cal
l
y
cont
ri
b
u
t
e
s t
o
m
o
re t
h
a
n
o
n
e
bi
n [
1
5]
,[
1
6
]
.
To i
n
crease t
h
e r
o
bust
n
ess
of t
h
e o
p
erat
or
SLB
P
pr
o
pose
t
h
e t
h
r
e
sh
ol
di
n
g
fu
nc
t
i
on
()
sx
is rep
l
aced
b
y
th
e
fo
llo
win
g
t
w
o fu
zzy
me
m
b
ersh
ip
fun
c
tio
ns:
1,
0,
()
0
.
5
0
.
5
,
1
d
zd
z
fz
d
z
d
d
zd
(7)
Evaluation Warning : The document was created with Spire.PDF for Python.
I
J
ECE
I
S
SN
:
208
8-8
7
0
8
Perf
or
ma
nce
S
t
udy
of
Sof
t
L
o
cal
Bi
n
a
ry
Pat
t
ern
over
Loc
a
l
Bi
nary
P
a
t
t
e
rn
u
nde
r N
o
i
s
y
..
.
.
(
S
abi
na
Y
a
sm
i
n
)
1
165
0,
1
,
()
1
(
)
dd
f
zf
z
(8)
The
param
e
t
e
r d c
ont
rol
s
t
h
e a
m
ount
o
f
f
u
zzi
fi
cat
i
on t
h
e
fu
n
c
t
i
on
per
f
o
r
m
s
. These
t
w
o
fuz
z
i
f
i
cat
i
on
fu
nct
i
o
ns a
r
e
p
l
ot
t
e
d bel
o
w:
Fi
gu
re
3.
The
t
w
o
f
u
zzi
fi
cat
i
o
n
fu
nct
i
o
ns
0,
1
,
dd
f
and
f
6.
NU
MER
I
C
A
L
VALI
D
ATI
O
N OF SLBP AN
D
LBP
In a
n
al
y
s
i
s
of LB
P and S
L
B
P
we ha
ve com
p
are
d
t
h
at
if we u
s
e edg
e
d
e
tectio
n
m
e
th
od
prio
r to
ap
p
l
y
recognition sys
t
em
then the efficiency will be enha
nce
d
. S
o
we ha
ve
use
d
som
e
edge det
ection m
e
thods
and
com
p
ared the results. From
T
a
ble 1 to Ta
ble 5 we ha
ve shown the c
o
m
p
arative data between LBP and SLBP.
In
pa
pe
r [16] mean
classi
fic
a
tion e
r
ror rate
s we
re c
o
m
p
ared
with
LBP and SLB
P
, pa
pe
r [13] showe
d
the face
recognition rat
e
of LBP with diffe
rent pa
ra
meters. But
in our experim
e
nt we have
disc
usse
d about the face
reco
g
n
i
t
i
on c
o
m
p
ari
s
on
rat
e
bet
w
ee
n LB
P
a
n
d
SLB
P
.
For a
n
al
y
s
i
s
w
e
have
use
d
O
R
L dat
a
base [
5
]
whe
r
e t
e
n
d
i
ffere
nt
im
ages of 2
5
6
g
r
ay
l
e
vel
s
wi
t
h
a
resol
u
tion
of
92x112
pixels of each of 40 di
stinct pers
ons.
In this e
xpe
riment we
proce
ss the sam
p
le image
wi
t
h
di
ffe
re
nt
edge
det
ect
i
o
n
m
e
t
hod
s t
h
e
n
cal
cul
a
t
e
t
h
e
LB
P aft
e
r
t
h
at
com
p
are u
s
i
n
g C
h
i
Sq
ua
re
m
e
t
hod
from
database
face im
ages to
measur
e
face
recognition rate and t
h
e sam
e
manner is
use
d
for SLBP.
Differe
n
t
edge
detection
m
e
thods are
use
d
first then the diffe
rent
noises a
r
e applied and
check the recognition rate.
After rem
o
v
a
l
o
f
no
ises t
h
e reco
gn
itio
n rate i
s
an
alized. Sam
p
le i
m
ag
e o
f
o
r
i
g
in
al an
d noisy are d
e
p
i
ct
belo
w:
(a)
(
b
)
(c)
(d)
(e)
Figure
4. The
ori
g
inal im
age(a) a
n
d no
isy imag
e (G
au
ssian
-
(
a
), Po
isso
n(b
)
, Salt & Pepp
er (
c
)
an
d Sp
eck
l
e(d
))
Tabl
e
1. C
o
m
p
ari
s
o
n
of
S
obel
ed
ge
det
ect
i
o
n
f
o
r
LB
P a
n
d
S
L
B
P
L
B
P(
1,
8) Noise
M
e
thods
Recognition Rate
of noisy
im
age
Rate after
Denoising
SLBP
Recognition Rate
of noisy
im
age
Rate after
Denoising
93.
98%
Gaussian
54.
57%
75.
59%
89.
32%
81.
61%
83.
62%
Salt & Pepper
32.
25%
82.
92%
81.
61%
84.
98%
Poisson
81.
95%
83.
78%
88.
57%
89.
84%
Speckle 34.
34%
73.
91%
67.
44%
75.
01%
Using Sobel edge
detection face
recognition rate is 93.98% for LB
P a
n
d 89.32%
for SLBP. But if
th
e im
ag
es are no
isy wit
h
d
i
fferen
t techn
i
ques th
en
r
ecognitio
n
rate will
b
e
redu
ced
and
after
rem
o
v
i
n
g
th
e
n
o
i
ses th
e
recog
n
ition
rate are enh
a
n
c
ed
. Table 1
sho
w
s th
e
co
m
p
ariso
n
d
a
t
a
.
Evaluation Warning : The document was created with Spire.PDF for Python.
I
S
SN
:
2
088
-87
08
IJEC
E
V
o
l
.
6,
No
. 3,
J
u
ne 2
0
1
6
:
11
6
1
– 11
67
1
166
Tab
l
e
2
.
C
o
m
p
ar
ison
o
f
Canny ed
g
e
d
e
tection
f
o
r
LBP and
SLBP
L
B
P(
1,
8)
Noise
M
e
thods
Recognition Rate of
noisy
im
age
Rate after
Denoising
SLBP
Recognition Rate
of noisy
im
age
Rate after
Denoising
91.
42%
Gaussian
0%
55.
76%
91.
58%
4.
09%
63.
28%
Salt & Pepper
68.
42%
76.
26%
79.
57%
84.
42%
Poisson
70.
38%
82.
58%
79.
07%
88.
24%
Speckle 0%
17.
08%
4.
3%
24%
In Ta
bl
e
2 C
a
nny
e
dge
det
e
ct
i
on m
e
t
hod i
s
use
d
a
nd
has
got
t
h
e com
p
ari
s
o
n
t
a
bl
e
of
fres
h
i
m
age
r
ecogn
itio
n
r
a
t
e
of
LBP and SLBP, r
a
te
of
n
o
i
sy im
age and
noise
re
m
oved im
age respectively. T
h
e
data
sh
ows th
at SLBP reco
gn
itio
n rate is
h
i
gh
er th
an LBP ev
en
for th
e no
isy i
m
ag
es.
Tabl
e
3. C
o
m
p
ari
s
o
n
of
R
o
be
rt
ed
ge
det
ect
i
o
n
f
o
r
LB
P a
n
d S
L
B
P
L
B
P(
1,
8)
Noise
M
e
thods
Recognition Rate of
noisy
im
age
Rate after
Denoising
SLBP
Recognition Rate
of noisy
im
age
Rate after
Denoising
88.
93%
Gaussian
0%
43.
78%
92.
48%
22.
79%
66.
18%
Salt & Pepper
0.
35%
73.
35%
55.
67%
84.
19%
Poisson
34.
66%
67.
10%
67.
18%
88.
57%
Speckle 0%
27.
56%
60.
86%
78.
76%
In Tab
l
e 3 Rob
e
rt edg
e
d
e
tectio
n
m
e
th
o
d
i
s
u
s
ed and
t
h
e reco
gn
itio
n rate o
f
SLB
P
is
h
i
gh
er th
an
LB
P, S
L
B
P
s
h
ows
t
h
e
bet
t
e
r
reco
g
n
i
t
i
on
rat
e
o
f
i
n
case
of
noi
sy
i
m
age an
d
noi
se
rem
ove
d i
m
ages.
Tabl
e
4. C
o
m
p
ari
s
o
n
of
Pre
w
i
t
t
edge
det
ect
i
o
n
fo
r LB
P
an
d
SLB
P
L
B
P(
1,
8) Noise
M
e
thods
Recognition Rate
of noisy
im
age
Rate after
Denoising
SLBP
Recognition Rate
of noisy
im
age
Rate after
Denoising
92.
15%
Gaussian
60.
47%
76.
78%
88.
27%
48.
48%
84.
01%
Salt & Pepper
2.
08%
82.
89%
3.
9%
92.
25%
Poisson
73.
03%
81.
39%
70.
33%
87.
93%
Speckle 43.
74%
75.
33%
76.
88%
82.
40%
In Ta
ble 4 Pre
w
itt edge dete
ction m
e
thod is used
a
nd the
recognition ra
te of LBP is highe
r tha
n
SLBP, B
u
t SLBP shows th
e
b
e
tter recogn
itio
n rate
o
f
in
case of
n
o
i
sy im
a
g
e an
d no
ise
rem
o
v
e
d
im
ag
es.
Tabl
e
5. C
o
m
p
ari
s
o
n
of
Lo
g e
dge
det
ect
i
o
n
f
o
r
LB
P a
n
d SL
B
P
L
B
P(
1,
8) Noise
M
e
thods
Recognition Rate
of noisy
im
age
Rate after
Denoising
SLBP
Recognition Rate
of noisy
im
age
Rate after
Denoising
88.
87%
Gaussian
0%
53.
76%
90.
10%
0.
5% 65.
17%
Salt & Pepper
57.
80%
80.
03%
68.
84%
84.
49%
Poisson
65.
59%
80.
65%
88.
33%
90.
14%
Speckle 0%
18.
85%
66.
72%
87.
74%
In Ta
ble 5 L
o
g edge
detection
m
e
thod is use
d
and th
e rec
o
gnition rate of
SLBP is highe
r
than LBP
,
SLBP sh
ows th
e
b
e
tter recogn
itio
n
rate
o
f
in
cas
e
of
noisy
im
age and
noise rem
oved images.
7.
CO
NCL
USI
O
N
In t
h
i
s
pape
r,
we ha
ve ana
l
y
zed soft
l
o
c
a
l
bi
nary
pat
t
ern (
S
LB
P
)
m
e
t
h
o
d
wi
t
h
e
d
ge det
ect
i
o
n
tech
n
i
qu
es fo
r face reco
gn
itio
n in case
of
n
o
i
sy con
d
ition
.
It is seen that if ed
g
e
d
e
tectio
n
m
e
th
o
d
i
s
u
s
ed
bef
o
re a
ppl
y
i
n
g
LB
P o
r
SL
B
P
, t
h
e t
i
m
e
com
p
l
e
xi
t
y
wil
l
be reduce
d
.
It
i
s
al
so obs
erve
d t
h
at
so
m
e
edge
detection techniques showe
d
better result for face
recognition
using LBP and
SL
BP. Soft Local
Binary
Pat
t
e
rn
wi
t
h
e
dge
det
ect
i
on
m
e
t
hods
s
u
ch
as C
a
n
n
y
,
R
o
b
e
rt
an
d
Lo
g
sh
ow
t
h
e
bet
t
e
r
r
e
sul
t
wi
t
h
c
o
m
p
are
t
o
LB
P. B
u
t
i
n
case of n
o
i
s
y
i
m
ages, LB
P w
i
t
h
edge det
ect
i
on m
e
t
hods p
r
o
v
i
d
es l
e
ss ef
fi
ci
ent
resul
t
whe
r
ea
s
SLBP p
r
ov
id
es
b
e
tter resu
lt
for n
o
i
sy
im
ag
es.
Evaluation Warning : The document was created with Spire.PDF for Python.
I
J
ECE
I
S
SN
:
208
8-8
7
0
8
Perf
or
ma
nce
S
t
udy
of
Sof
t
L
o
cal
Bi
n
a
ry
Pat
t
ern
over
Loc
a
l
Bi
nary
P
a
t
t
e
rn
u
nde
r N
o
i
s
y
..
.
.
(
S
abi
na
Y
a
sm
i
n
)
1
167
REFERE
NC
ES
[1]
S
.
Yas
m
in and
M
.
Rana
,
”Effi
cien
t Lo
cal
Bin
a
r
y
P
a
t
t
er
n with Modified
Edg
e
Det
ection Technique for
Face
Recognition,” in
Proceed
ing
of Internat
io
nal Conferen
ce
on Ele
c
tri
c
al
Engine
ering
and Informatio
n
Communication Technology
(
I
CEEICT)
,
May
,
2
1
-23, Dhak
a, Bangladesh, 2015.
[2]
S.
Ya
smi
n
a
nd M.
Ra
na
,
”
C
ompa
ra
t
i
v
e
St
udy
of De
noi
si
ng T
e
c
hni
que
s
for Fa
c
i
a
l
Ima
g
e
usi
ng Qua
lity
M
eas
urem
ent,
”
in
Proceedin
g
of In
ternational Conference
on
Electrica
l
Engin
eering
and Information
Communication Technology
(
I
CEEICT)
,
May
,
2
1
-23, Dhak
a,
Bangladesh, 2015.
[3]
Yusra A.
,
et al
., “Co
m
parison of Image Quality
A
ssess
ment: PS
NR,
HVS,
S
S
I
M
,
UIQI,
”
International Journal of
Scien
tifi
c
&
Engineering
Rese
arc
h
,
vol/issue: 3(8)
, 2012
.
ISSN 2229-5518.
[4]
J. Patil and S. Ja
dhav, “
A
Com
p
arativ
e Stud
y
of I
m
age Denoising Techn
i
ques,
”
International Journ
a
l of Innovative
Research
in S
c
ience, En
g
i
neerin
g and Techno
log
y
,
vo
l/issue: 2(3)
, 2013
.
[5]
http://www.cl.cam.ac.uk
/Resear
ch/D
TG/attarch
iv
e/pub/d
a
ta/att_faces.
[6]
Vija
ya
lakshm
i. A,
et a
l
.
, “Image Denoising f
o
r differ
e
nt
noise models b
y
v
a
rious
filters:
A Brief Survey,”
International
Jo
urnal of Emerging Trends
&
Te
chnology in Co
mputer Science (
I
JETTCS
)
,
vol/issue: 3(6), 2014.
ISSN 2278- 6856.
[7]
C. Srivastav
a
,
et
al
., “
P
erform
an
ce Com
p
arison
of Various Filter
s
and W
a
velet
T
r
ansform
for Image De-Noising,
”
IOSR Journal of Computer Engineering (
I
OSR-JCE
), vol/issue: 10(1), pp. 55-
6, 2013. e-ISSN: 2278-0661, p
-
ISSN: 2278-8727.
[8]
T. Ahonen
,
et
al
.,
“
F
ace re
cogn
ition wi
th Loc
a
l
Binar
y
P
a
t
t
erns
,”
Ma
chine Vision Group
,
University
of Oulu,
Finland, 2004.
[9]
T. Ahonen,
et al.
, “
F
ace d
e
script
i
on with Loca
l Bi
nar
y
P
a
tt
erns: Applic
ation to F
a
c
e
Recogn
ition
,
”
Machine Vis
i
on
Group
, University
of Oulu
, Fin
l
and, 2006
.
[10]
R. Dosselmann and X. D. Yang,
“
A
F
o
rm
al As
ses
s
m
ent of the S
t
ructur
al S
i
m
i
l
a
r
i
t
y
Index
,
”
Te
ch
nica
l Repor
t TR
-
CS 2008-2 September, 2008.
[11]
M.
Oja
l
a,
et
al
.
,
“
A
com
p
arativ
e s
t
ud
y
o
f
tex
t
u
r
e m
eas
ures
wi
th classification
based on
featur
e distributions,”
Pattern
Recognition
, vo
l. 29, pp.
51-59, 1996
.
[12]
T. Oj
ala
,
et a
l
.
,
“
M
ultiresolution
gra
y
-sca
le
and r
o
tation
inv
a
rian
t
textu
r
e
cl
assi
fi
cation
wi
th loca
l binar
y
p
a
tt
erns,
”
IEEE
T
r
ans. Pat
t
ern Ana
l
,
Mach
. Intell
., vo
l/issue: 24(7), pp. 971–
987, 2002
.
[13]
Anagha V.
M.
,
et al.
, “
A
New Technique for LBP
Method to Im
pr
ove Face Recog
n
ition,”
International Journal of
Emerging Techn
o
logy and Adva
nced Engin
eerin
g
, vol/issue: 1(1
)
, 2011. Websit
e: www.ijetae.co
m (ISSN 2250
-
2459).
[14]
T. Ahonen
,
et a
l
.
, “Rotation Inv
a
rian
t Image Descri
ption with
Local B
i
nar
y
Pattern
Histogram Fourier Features,”
A
.-B. Salberg
, J.Y. Hardeberg, a
nd R. Jenssen
(Eds.), Spring
er-Verlag
Berl
in Heidelberg
, pp
. 61–
70, 2009
.
[15]
D.
Huang,
et al.
,
“
L
ocal
Bin
a
r
y
P
a
tt
erns
and
Its
A
pplic
ation
to
F
a
c
i
al
Im
age Ana
l
ys
is
: A S
u
rve
y
,
”
.
[16]
T. Ahonen
and
M. Pietik
¨
a
in
en, “Soft Hi
stograms for Local Binar
y
Patterns,”
Machine
Vision
Group, Infotech
Oulu
,
WWW home
pa
ge
: htt
p
:
//www.
e
e
.
oul
u.
fi
/mvg.
[17]
N. Senthilkumar
an and R
.
R
a
jes
h
, “Edge Detecti
on Techniques f
o
r Image Segmentation
- A Surv
ey
,”
Pr
oc
eeding
s
of the Internatio
nal Conference
on Managing Next
Gen
e
ration
Software Applications (
M
NGSA-
08)
, pp.749-760,
2008.
[18]
G. T. Shrivaksh
a
n, and C. Chandrasekar
, “A C
o
mparison
of various Edge Detection
Techniqu
es used in Image
Processing,”
IJC
S
I International
Journal of Computer Science Is
sues
, vol/issue:
9(5), 2012. ISSN (Online): 169
4-
0814.
Evaluation Warning : The document was created with Spire.PDF for Python.