Internati
o
nal
Journal of Ele
c
trical
and Computer
Engineering
(IJE
CE)
V
o
l.
6, N
o
. 5
,
O
c
tob
e
r
201
6, p
p
. 2
158
~216
6
I
S
SN
: 208
8-8
7
0
8
,
D
O
I
:
10.115
91
/ij
ece.v6
i
5.1
122
2
2
158
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
Improved Canny Edges Using Ce
llular Based Particle Swarm
Optimization Technique for
T
a
mil Sign Di
gital
Im
ages
M.
Krishn
ave
n
i, P.
Sub
as
hini, T.T.
Dhi
v
yapr
abh
a
Departm
e
nt o
f
C
o
m
puter Scien
c
e
,
Avinashi
linga
m
Institute
for H
o
m
e
Scienc
e
and
Higher
Educ
ati
on for W
o
m
e
n, I
ndia
Article Info
A
B
STRAC
T
Article histo
r
y:
Received
May 9, 2016
Rev
i
sed
Ju
l 7
,
2
016
Accepte
d
J
u
l 17, 2016
The d
e
velopment of
computer
based
sign lang
uage r
ecognitio
n s
y
stem, for
enabling communication with
hearing
im
paired people,
is an
important
res
earch
ar
ea
th
at f
a
c
e
s
differ
e
n
t
ch
all
e
nges
in t
h
e pre-pro
ces
s
i
n
g
s
t
age
of
image processin
g
, particularly
in
boundar
y
d
e
tection stage. In
edg
e
detection,
the possibilit
y o
f
achiev
i
ng high
qualit
y im
ages signific
a
ntl
y
dep
e
nds on the
fitting
threshold
values,
which
are g
e
ne
ra
ll
y se
lec
t
ed using
can
n
y
m
e
tho
d
,
and these thresh
old valu
es may
var
y
,
b
a
sed on the ty
pe o
f
imag
es and th
e
applications cho
s
en. Th
is
resear
ch work presents
a novel id
ea of
establishing
a h
y
brid p
a
rti
c
le
swarm
optim
ization
algori
t
hm
,
which is a
com
b
ination
of
PSO
with the behavioural patter
n of ce
llular org
a
nism in cann
y
method that
defines an objective to f
i
nd optimal thre
shold values for th
e im
pl
em
entat
i
on
of double thresh
olding h
y
steresis met
hod, which is viewed as
a non-linear
com
p
lex proble
m
. The attem
p
t
to incorpora
t
e t
h
e m
odel has minim
ized th
e
problem of quick convergen
ce
of PSO algorithm which has improved th
e
detection of
br
oken edges
.
Th
e efficiency
of
the proposed
algorithm is
proved through
the
exp
e
rimental observation
,
do
ne in T
a
m
il sign
im
ages to
indicate the better perfo
rmance
of cann
y
oper
a
to
r b
y
in
troducing
new variant
based PSO.
Keyword:
C
a
nny
e
d
ges
Cellu
lar org
a
n
i
sm
Particle swarm op
ti
m
i
zatio
n
Tam
i
l
si
gn l
a
n
gua
ge
Thre
sh
ol
di
ng
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
:
M. Krishn
av
eni,
Depa
rt
m
e
nt
of
C
o
m
put
er Sci
e
nce,
Av
i
n
ash
ilin
g
a
m
In
stitu
te for
Ho
m
e
Scien
ce
an
d Hi
g
h
e
r Edu
catio
n fo
r
Women
,
Co
im
b
a
to
r
e
-
641
043
, Tam
i
l
N
a
d
u
, I
n
d
i
a.
Em
a
il: k
r
ish
n
a
v
e
n
i
.rd
@
g
m
ail
.
co
m
1.
INTRODUCTION
Sign language
is the only reliable tool for
unde
rs
tanding skills am
ong the dea
f
pe
ople
.
Also, t
h
e
p
r
o
f
icien
t
activ
ities o
f
h
e
arin
g
im
p
a
ired
peo
p
l
e can
b
e
im
p
r
o
v
e
d
on
ly b
y
u
s
ing
con
s
i
s
ten
t
SLR syste
m
. Bu
t
t
h
ere i
s
n
o
si
n
g
l
e
st
an
dar
d
fo
rm
of si
g
n
l
a
n
gua
ge,
an
d i
t
v
a
ri
es f
r
om
regi
on
t
o
regi
on
.
Tam
i
l
Si
gn La
ng
ua
g
e
(TSL
) i
s
a
regi
on
base
d si
gn
l
a
ng
ua
ge, c
o
nsi
d
ere
d
t
o
be a
m
o
re usef
ul
m
eans t
o
ha
ve c
o
n
f
i
n
e
d
i
m
pro
v
em
ent
i
n
t
h
ei
r o
w
n bo
u
nda
ry
[
1
]
.
Thi
s
l
a
n
gua
ge
for
dea
f
pe
o
p
l
e
i
s
gai
n
i
n
g
fact
ual
im
por
t
a
nce i
n
t
h
e r
e
gi
o
n
al
com
m
uni
cat
i
on. T
h
e m
a
i
n
object
i
v
e o
f
t
h
i
s
pape
r i
s
t
o
p
r
op
ose a
ro
b
u
st
edge
det
ect
i
o
n
m
e
t
hod,
usi
n
g
hy
bri
d
opt
i
m
i
zati
on t
echni
que
,
whi
c
h c
oul
d be
a fi
ne c
o
nt
ri
b
u
t
i
o
n
f
o
r a
u
t
o
m
a
t
e
d sy
st
em
t
h
at
i
d
ent
i
f
i
e
s an
d
r
ecogn
izes th
e
TSL sign
s.
A
s
th
e f
i
r
s
t step, th
e im
ag
e acq
u
i
r
e
d, u
s
i
n
g
d
i
g
i
t
a
l i
m
ag
es, ar
e resized
in
to
25
6x
256
and converted
into grayscal
e
im
age
and that im
age is preproce
ssed by
opt
i
m
i
zed wei
ght
ed
m
e
di
an noi
s
e
filterin
g
techn
i
q
u
e
[2
]. As a p
r
im
ary an
alys
is o
f
seg
m
en
tatio
n
ph
ase, commo
n
ed
g
e
reco
gn
itio
n
algorith
m
s
su
ch
as So
b
e
l
,
Ro
b
e
rt, Canny an
d
Prew
itt
are ex
perim
e
n
t
ed
, fo
r wh
ich
th
e sco
p
e
o
f
can
n
y
is ex
tended
,
b
y
in
trodu
cing
PSO, and
th
at PSO is i
m
p
r
ov
ised
b
y
in
co
rp
oratin
g
th
e m
o
tili
ty b
e
h
a
v
i
o
u
r of cellu
lar org
a
n
i
sm
called
a n
e
w
varian
t Cellu
lar Particle Swarm Op
ti
m
i
zat
io
n
(CPSO) alg
o
rith
m
.
Th
e ex
perim
e
n
t
s are c
a
rried
out
wi
t
h
1
8
Ta
m
i
l
conso
n
ant
s
and eac
h
rep
r
esent
s
t
h
e St
at
i
c
im
ages of t
h
e pal
m
si
de of
ri
g
h
t
ha
nd
. Fi
g
u
re
1
depi
ct
s t
h
e ge
n
e
rat
e
d c
o
n
s
o
n
a
n
t
s
o
f
Tam
i
l
signs
t
h
at
bel
o
n
g
t
o
TSL
dat
a
s
e
t
.
The
or
ga
ni
z
a
t
i
on
of t
h
e
pa
per i
s
as fol
l
o
ws:
Sect
i
on 2 i
n
t
r
o
d
u
ces o
v
er e
d
g
e
prese
r
vi
ng
d
e
t
a
i
l
s
t
h
rou
gh
t
h
res
hol
di
n
g
conce
p
t
s
,
usi
n
g
canny
Evaluation Warning : The document was created with Spire.PDF for Python.
I
J
ECE
I
S
SN
:
208
8-8
7
0
8
Im
pr
oved Canny E
d
ges
Using Cellular
B
a
s
e
d P
a
rticle
Sw
arm
O
p
timiz
ation Tec
hni
que
....
(
M
. Kris
hnaveni)
2
159
ope
rat
o
r. Sect
i
on
3 ex
pl
ai
ns t
h
e p
r
o
p
o
se
d h
y
b
ri
d ca
nny
e
d
ge det
ect
i
o
n m
e
t
h
o
d
usi
ng
PS
O,
by
i
n
t
r
o
d
u
c
i
ng a
n
e
w v
a
rian
t.
Sectio
n 4 illu
strates th
e resu
lts an
d
an
alysis, b
a
sed
on
ev
alu
a
tion
assessm
en
ts of the
expe
ri
m
e
nt
al
resul
t
s
. Sect
i
o
n
5 c
oncl
ude
s t
h
e fi
n
d
i
n
gs a
n
d
t
h
e sc
ope
o
f
t
h
e resea
r
ch
w
o
r
k
i
n
TSLR
.
Fi
gu
re
1.
M
a
n
u
al
l
y
Gene
rat
e
d Tam
i
l
Si
gn L
a
ng
ua
ge
Dat
a
s
e
t
(co
n
s
ona
nt
s)
2.
BOUNDARY DETEC
TION USING THRESHOLDING
An e
dge i
s
de
fi
ned
base
d o
n
t
h
e swi
f
t
cha
n
g
e
of i
n
t
e
nsi
t
y
o
f
an i
m
age. Edge det
ect
i
o
n i
s
a pr
ocess o
f
find
ing
th
e sh
arp
co
n
t
rast b
a
sed
on
th
e in
tensities o
f
an
imag
e, b
y
redu
ci
n
g
th
e am
o
u
n
t
o
f
d
a
ta in
an
imag
e,
whi
l
e
p
r
ese
r
vi
ng i
m
port
a
nt
st
ruct
ural
feat
ur
es of t
h
at
i
m
ag
e [3]
.
T
h
e m
e
tho
d
s
fo
r
det
ect
i
ng t
h
e
ed
ges
d
e
pe
n
d
o
n
t
h
e co
m
p
u
t
atio
n
of im
ag
e g
r
ad
ien
t
s and
th
e typ
e
o
f
filter u
s
ed
to
calcu
late g
r
ad
ien
t
esti
m
a
tes i
n
th
e
ho
ri
zo
nt
al
a
n
d
vert
i
cal
di
rect
i
ons
. T
h
e
r
ef
ore
,
t
h
e t
h
res
h
ol
d
val
u
e
i
s
t
h
e
o
n
e
w
h
i
c
h
deci
de
s w
h
et
he
r e
d
ge
s are
prese
n
t
or
not
at
an im
age poi
nt
[4]
.
The cr
u
c
i
a
l
pro
b
l
e
m
i
s
t
h
e i
ssue of ch
oosi
n
g t
h
e t
h
re
shol
ds.
A com
m
onl
y
use
d
m
e
t
hod f
o
r
fi
n
d
i
n
g t
h
e
app
r
op
ri
at
e t
h
r
e
sh
ol
d i
s
ca
nn
y
m
e
t
hod, i
n
whi
c
h t
h
e
res
u
l
t
a
nt
out
put
c
o
nt
ai
ns
t
h
i
n
ed
ges
,
an
d
t
h
e ed
ge pi
xel
s
are l
i
nke
d
usi
ng e
d
ge t
r
ac
ki
ng
pr
oce
d
ure
wi
t
h
hy
st
ere
s
i
s
t
h
res
hol
di
n
g
m
e
t
h
o
d
[
5
]. I
t
v
a
r
i
es t
h
e th
resho
l
d
by tr
ack
ing
th
e ed
g
e
on
ce and
f
i
nd
ing
th
e
calcu
lated
th
r
e
sh
o
l
d
as th
e second
resu
ltan
t
argumen
t. Th
is m
e
t
h
od
is th
erefore b
e
tter at su
pp
r
e
ssi
ng
no
ise, an
d
m
o
r
e
lik
ely to detect true weak
edge
s [
6
]
.
The
r
ef
ore
,
t
h
e o
p
e
r
at
or
fi
n
d
s i
t
s
desi
g
n
t
o
be b
e
st
edge
det
ect
or
. Fo
r i
m
pl
em
ent
a
t
i
on, t
h
e
di
scret
e
ap
pro
x
i
m
a
tio
n
to Gau
ssian
fu
n
c
tion is
do
ne with
1
.
4
,
an
d
t
h
e S
obel
o
p
e
r
at
or
i
s
per
f
o
rm
ed t
o
fi
n
d
t
h
e
approxim
a
te a
b
sol
u
te
gra
d
ient
m
a
gnitude at
each
poi
nt wit
h
the
convol
ution of m
a
sk G
x
and
G
y
.
Ho
we
ver
,
t
h
i
s
pr
obl
em
of cho
o
si
ng a
p
p
r
o
p
ri
at
e t
h
re
sh
ol
d val
u
es m
a
y vary
o
v
er t
h
e im
age [6]
.
There
f
ore,
t
o
im
pro
v
e t
h
e c
a
nny
e
d
ges, t
h
i
s
pa
pe
r p
r
o
pos
es t
o
fi
n
d
t
h
e o
p
t
i
m
al
t
h
res
hol
d
val
u
es, by
im
pl
em
ent
i
ng nat
u
re-i
n
s
pi
re
d
com
put
i
ng t
e
chni
que
, i
n
c
o
m
b
i
n
at
i
on
wi
t
h
be
ha
vi
o
u
ral
pat
t
e
rn
o
f
ce
l
l
u
l
a
r
or
ga
ni
sm
.
3.
IMPROVED CANN
Y EDGES USING PROP
OSED CELLULAR PSO
Th
ou
g
h
t
h
e ap
pr
o
p
ri
at
e t
h
res
hol
d val
u
e
s
i
s
cho
s
en i
n
ca
nn
y
m
e
t
hod
, t
h
e edge
det
a
i
l
s
t
h
at
have t
o
b
e
prese
r
ved a
r
e
m
o
re in sign
im
ages, and they va
ry
accordi
ng t
o
different im
ag
es. T
h
e propose
d
method
d
e
fi
n
e
s an
o
b
jectiv
e to
ch
oose th
e
o
p
tim
al
th
resho
l
d
val
u
es,
usi
n
g
C
P
SO,
w
h
i
c
h
is
also c
o
m
p
ared with
classical PSO-base
d
canny e
d
ge detecti
on
[7
]
.
In t
h
i
s
resea
r
ch w
o
rk
, t
h
e
pr
op
ose
d
C
P
S
O
al
go
ri
t
h
m
i
s
appl
i
e
d
i
n
t
h
e ca
nny
m
e
t
hod,
t
h
at
d
e
fi
nes a
n
ob
je
ct
i
v
e, t
o
fi
n
d
opt
i
m
al
t
h
resh
ol
d
val
u
es
f
o
r
t
h
e i
m
pl
em
ent
a
t
i
on
of
do
u
b
l
e
t
h
resh
o
l
di
ng
hy
st
ere
s
i
s
m
e
t
hod,
w
h
i
c
h i
s
vi
ewe
d
as
a n
o
n
-
l
i
n
ear
c
o
m
p
l
e
x pr
obl
em
[
8
]
.
3.
1.
Particle sw
ar
m op
timiz
a
tio
n
w
i
th
fibro
b
last
beh
a
viour
Particle Swarm
Op
ti
m
i
za
tio
n
(PSO) is a n
a
ture
i
n
sp
ired
algor
ith
m
p
r
op
o
s
ed
b
y
K
e
nn
ed
y and
Eber
ha
rt
(
1
9
9
5
)
.
Thi
s
al
go
ri
t
h
m
i
s
devel
o
ped
by
t
h
e c
h
aract
eri
s
t
i
c
s o
b
ser
v
e
d
f
r
om
bi
r
d
s fl
ocki
ng
and
fi
sh
scho
o
ling
.
Each
p
a
rticle
(p
i
)
has i
t
s
o
w
n
po
si
t
i
on (
x
i
) and
v
e
lo
city (v
i
) in
the ev
oluti
o
n
a
ry
re
gio
n
.
A
set o
f
p
a
ram
e
ters such
as in
ertia
weigh
t
(
ω
), constriction c
o
efficients
(K), a
cceleration fac
t
ors (
φ
1
,
φ
2
), velo
city
ran
g
e [
-
vm
ax,
+
vm
ax]
and r
a
nd
om
num
bers (
r
1
,r
2
) enforce the swa
r
m to migrate in the n -
dim
e
nsional
problem space, and e
v
aluated using the fitness functio
n until the pre
d
eterm
i
ned criteria (s) or the m
a
xi
m
u
m
2 4
5 4
2
4 9
1
2
9
4
5
12
15
12
5
4 9
1
2
9
4
2 4
5 4
2
-
1
0
1
-
2
0
2
-
1
0
1
1 2
1
0 0
0
-1
-2
-1
Discrete
app
r
oxi
m
a
t
i
on
1
.
4
.
G
x =
G
y
=
Evaluation Warning : The document was created with Spire.PDF for Python.
I
S
SN
:
2
088
-87
08
I
J
ECE
Vo
l. 6
,
N
o
. 5
,
O
c
tob
e
r
20
16
:
215
8
–
21
66
2
160
n
u
m
b
e
r of iteratio
n
s
(t
) is m
e
t. Th
e obj
ective is to
find
op
timal so
lu
tio
n
(i
n
d
i
v
i
du
al b
e
st
(P
best
) and s
o
ci
al best
(G
best
)
)
by
u
p
d
a
t
i
ng t
h
e
vel
o
c
i
t
y
(v
i
)
co
ord
i
nates to
po
sition (
x
i
) e
quat
i
ons
ove
r t
i
m
e whi
c
h are
re
pre
s
ent
e
d i
n
eq. 1
a
n
d
e
q
. 2 [9]
.
∗
∗
∗
∗
∗
(1)
w
h
er
e
2
|
2
4
,
0
.
5
ran
d
()
i
s
a
ran
dom
fu
nct
i
o
n t
h
at
ge
ne
rat
e
s a
di
st
ri
b
u
t
e
d
ra
n
dom
num
ber
w
h
i
c
h l
i
e
s
wi
t
h
i
n
a
ra
nge
o
f
0
and
1
.
X
ij
(t
+
1
)
= X
ij
(t
)
+ V
ij
(
t
+1
)
(2
)
In a fe
w attem
p
ts, it is observed that the quick
conve
r
ge
nc
e of PS
O algorith
m
cannot fi
nd m
u
ltiple
optim
al
solutions in a single search
space problem
. This intricacy is c
ontrolled by incorporating the motilit
y
b
e
h
a
v
i
or
of fi
brob
last org
a
n
i
sm
[1
0
]
. Th
e m
o
tility facto
r
(
σ
)
a
c
t
s
a
s
an
en
er
g
e
tic
f
o
r
c
e
th
at e
n
ab
le
s
th
e sw
a
r
m
to prolife
r
ate a
r
ound t
h
e entire prob
lem
space and the
particles are acceler
ated towa
rds
the best ca
ndidate
so
lu
tion
s
wh
ich
h
a
v
e
b
e
tter
fitn
ess
v
a
l
u
es. Th
e fun
c
tio
nal
i
t
y
of s
w
arm
h
a
s bee
n
p
r
o
g
re
ssed
wi
t
h
m
i
gr
at
i
o
n
factor (
σ
) wh
i
c
h
is
app
lied
i
n
n
e
ighb
or
hood
p
a
r
ticles that in
cr
eases th
e d
e
g
r
ee
o
f
in
t
e
r
actio
n am
o
n
g
t
h
e
po
p
u
l
a
t
i
on
[
1
1
]
. Hence
f
ort
h
,
t
h
e ra
nd
om
l
y
gene
rat
e
d
col
l
agen
en
ha
nced
t
h
e m
ovi
n
g
s
p
eed
o
f
part
i
c
l
e
s, a
n
d
t
h
e ef
fi
ci
ency
of C
P
SO
al
g
o
r
i
t
h
m
t
o
obt
ai
n
opt
i
m
al
sol
u
t
i
on ha
s bee
n
i
m
pr
o
v
ed
. T
h
e al
go
ri
t
h
m
1 gi
ve
s t
h
e
st
eps
fo
r i
m
pl
em
ent
a
t
i
on
of C
PSO
t
ech
ni
q
u
e
.
Algo
rithm 1.
Cellula
r Pa
rticle Swa
rm
Optimiza
tio
n
(CPSO)
Step 1:
In
itiali
ze th
e p
a
rticles o
f
po
pu
latio
n
size M
i
(i
= 1,
2,…
,
n
)
wi
t
h
r
a
nd
om
l
y
gener
a
t
e
d p
o
si
t
i
on
x
i
and
v
e
lo
city v
i
i
n
t
h
e
n-
di
m
e
nsi
onal
searc
h
spac
e.
Beg
i
n
Step 2: Repe
at
Eval
uat
e
t
h
e
o
b
ject
i
v
e f
u
nct
i
o
n
o
f
e
v
ery
pa
r
t
i
c
l
e
usi
n
g
st
an
dar
d
be
nchm
ark
fu
nct
i
o
n.
Step 3:
Co
m
p
are th
e fitn
ess
v
a
lu
e
of each
particle F (M
i
)
wi
t
h
t
h
e
val
u
e
of i
n
di
vi
d
u
al
b
e
st
P
best
.
If
th
e
cu
rr
en
t
v
a
lu
e of a p
a
rt
icle F (M
i
) is
b
e
tter th
an P
be
st
, th
en
th
e
v
a
lu
e
o
f
M
i
is set
to
P
best
an
d then
t
h
e
p
o
s
ition of
a
cu
rren
t p
a
rticle
x
i
is assign
ed to
P
i
in t
h
e
proble
m
space.
Step 4:
Id
en
tify th
e n
e
ig
hb
orh
ood
b
e
st (
G
be
st
) p
a
rticle in
t
h
e po
pu
latio
n. If th
e curren
t
v
a
lu
e of a p
a
rt
icle F
(M
i
) is b
e
tter th
an
F
(G
best
), th
en
the v
a
lu
e
o
f
M
i
is set to
G
best
. The index val
u
e of a
current
particle x
i
is
assi
gne
d t
o
P
g
.
Step 5:
Update
the velocity
and position of
par
ticle accordi
n
g to t
h
e
following e
quation:
ij
(t
+
1
)
=
ω
* K
*
V
ij
(t
)
+ (
φ
1
* R
1
* (
P
best
– x
p
(t
)
) +
σ
* (
φ
2
* R
2
* (
G
best
-p
g
(t
)
)
)
(3
)
X
ij
(t
+
1
)
= X
ij
(t
)
+
V
ij
(t
+
1
)
(4
)
w
h
er
e
V
ij
(t)
=
vel
o
city of
j
th
p
a
rticle in
i
th
iteratio
n
at ti
m
e
t
X
ij
(t)
=
p
o
s
ition of
j
th
p
a
rticle in
i
th
iteratio
n at
ti
m
e
t
R
1
, R
2
= r
a
ndom
n
u
m
b
er
s lies b
e
tw
een
0
and 1
G
best
= N
e
i
g
hbor
hoo
d (
s
o
c
ial)
p
a
r
ticle b
e
st g
x
P ,
p
g
=
In
de
x
v
a
l
u
e o
f
c
u
r
r
e
n
t
and
nei
g
h
b
o
r
h
oo
d
pa
rt
i
c
l
e
2
|
2
4
,
σ
= m
o
til
ity facto
r
Evaluation Warning : The document was created with Spire.PDF for Python.
I
J
ECE
I
S
SN
:
208
8-8
7
0
8
Im
pr
oved Canny E
d
ges
Using Cellular
B
a
s
e
d P
a
rticle
Sw
arm
O
p
timiz
ation Tec
hni
que
....
(
M
. Kris
hnaveni)
2
161
w
h
er
e
ψ
=
10
4
-1
a
sat
= 1.1
|
|
|
|
|
|
w
h
er
e
p
c
an
d d
c
are
po
sitiv
e co
n
s
tants
||c|| = fro
m
0
to 0
.
20
8
||
ω
|| = f
r
o
m
0
.
33
to 1
f
i
= i
th
direction of a
vector
f
x
=
po
sitio
n of
a p
a
rticle
Until
term
in
atio
n cond
itio
n(s) / m
a
x
i
m
u
m
i
t
e
r
atio
n (s) is m
e
t.
End
3.
2.
Cellular base
d PSO for im
proved c
a
nny
edges
Cellu
lar p
a
rticle swarm
o
p
timizati
on al
g
o
ri
t
h
m
i
s
im
pl
em
ent
e
d t
o
fi
nd t
h
e
opt
i
m
al
sol
u
t
i
ons (l
ow L
and
Hi
g
h
H
)
t
h
at
can be ap
pl
i
e
d f
o
r se
g
m
ent
a
t
i
on
of t
h
e re
gional sign language i
m
ages. The ra
nge
s of
t
h
res
hol
ds a
r
e
est
i
m
a
t
e
d t
h
ro
ug
h
hi
st
o
g
ram
anal
y
s
i
s
of
t
h
e si
gn l
a
ng
ua
g
e
dat
a
set
.
T
h
e
pr
oce
d
u
r
e
of
C
PSO
opt
i
m
i
zed cann
y
m
e
t
hod t
o
pe
rf
orm
edge
det
ect
i
on i
s
desc
ri
bed
bel
o
w
i
n
al
go
ri
t
h
m
2.
A
l
go
rit
h
m 2:
Step 1:
I
n
p
u
t
a
re
gi
o
n
al
si
g
n
l
a
ng
ua
ge i
m
ages I
(u
,v
).
Step 2:
Sm
o
o
t
h
i
ng
- Blurri
n
g
o
f
t
h
e im
ag
e t
o
rem
o
v
e
n
o
i
se b
y
app
l
yin
g
a
Gau
s
sian
filter. Th
e co
nvo
lu
tio
n
of
an
im
ag
e with a core of
Gaussian
filter
u
s
i
n
g stand
a
rd
d
e
v
i
atio
n
o
f
σ
=
1.
4 i
s
s
h
o
w
n i
n
e
quat
i
o
n
(5
)
gi
ve
n
bel
o
w:
2
4542;
4912
94;
5
1215
12
5;
4
91294;
24542;
(5)
Step
3:
Determin
atio
n
o
f
grad
ien
t
s - Find
ed
g
e
s b
y
d
e
termin
in
g
g
r
ad
ients
o
f
th
e
inpu
t i
m
ag
es
to
i
d
entify
th
e
varie
d
intensit
y of the im
age. Gra
d
ients at each pi
xel
ha
ve found using
Sobel
operat
or.
It can be im
plemented
to
approx
im
ate
th
e
g
r
ad
ien
t
i
n
th
e x-
a
n
d
y
-
di
rect
i
o
ns
res
p
ect
i
v
el
y
i
n
t
h
e sm
oot
hed
i
m
age,
by
a
ppl
y
i
ng
t
h
e
ker
n
el
s gi
ve
n
i
n
e
quat
i
o
n (
6
) and
(
7
)
101;
202;
10
1;
(6
)
121;
0
00;
1
2
1;
(7)
whe
r
e,
G
x
an
d G
y
are gra
d
ie
nts in t
h
e
x a
n
d y directions
res
p
ectively;
The
gra
d
i
e
nt
m
a
gni
t
ude
s ca
n be
det
e
rm
i
n
ed by
a
p
pl
y
i
ng
an E
u
cl
i
d
ea
n d
i
st
ance m
easure, usi
ng t
h
e
Pyth
ago
r
as law as sh
own
in eq
u
a
tion
(8
).
|
|
(8)
Step 4:
N
o
n-
m
a
xim
a
supp
r
e
ssi
on
–
A
gra
d
i
e
nt
i
m
age h
a
s bl
u
rre
d e
d
g
e
s, w
h
i
c
h
i
s
t
r
ansf
o
r
m
e
d i
n
t
o
s
h
ar
p
edge
s,
by
pre
s
ervi
ng al
l
l
o
cal
m
a
xim
a
and e
l
im
i
n
at
i
ng
ot
he
r pi
xel
s
i
n
t
h
e
im
age. Thi
s
pr
ocess c
o
n
s
i
s
t
s
of t
h
e
follo
win
g
t
h
ree
steps:
The g
r
a
d
i
e
nt
di
rect
i
o
n
θ
i
s
ro
un
de
d o
f
f
t
o
t
h
e neare
s
t
45
◦
, it corresponds t
o
the 8 c
o
nnect
e
d
nei
g
hb
o
u
r
h
oo
d
pi
xel
val
u
es
of
t
h
e i
m
age,
C
o
m
p
are t
h
e gra
d
i
e
nt
m
a
gni
t
ude o
f
t
h
e c
u
r
r
ent
pi
xel
valu
e with
th
e
g
r
ad
ien
t
m
a
g
n
itu
d
e
of th
e p
i
x
e
ls
foun
d in
bo
th
po
sitiv
e an
d th
e
n
e
g
a
tiv
e d
i
rectio
n
,
and
If th
e
g
r
ad
ien
t
m
a
g
n
itu
d
e
o
f
th
e cu
rren
t p
i
xel is larg
est, th
en
p
r
eserv
e
th
e v
a
l
u
e o
f
t
h
e cu
rren
t p
i
x
e
l to
i
m
p
r
ov
e th
e edg
e
streng
th
of
th
e im
age. Otherwise, the
pixel values
ar
e s
u
p
p
r
esse
d
i
n
t
h
e gi
ve
n gra
d
i
e
nt
im
ages.
Step 5:
Do
u
b
l
e
t
h
res
hol
di
n
g
-
The res
u
l
t
a
nt
edge pi
xel
s
m
a
y
cont
ai
n
noi
se or i
r
rel
e
va
n
t
val
u
es. T
h
e c
a
nny
m
e
t
hod
uses
d
o
u
b
l
e
t
h
res
hol
di
n
g
(
h
i
g
h
T1
and
l
o
w T
2
)
va
l
u
es t
o
f
u
rt
he
r
sup
p
r
ess t
h
e
n
o
i
s
e co
nt
e
n
t
as
wel
l
as
prese
r
ve t
h
e
t
r
ue i
m
age. E
d
ge
pi
xel
s
(T
) l
a
rge
r
t
h
an
hi
g
h
t
h
res
hol
d
(T
1)
are
m
a
rked
as st
r
o
n
g
.
E
d
g
e
pi
x
e
l
s
(T) sm
al
l
e
r t
h
an l
o
w t
h
res
h
ol
d (T
2) a
r
e rem
ove
d. E
dge
pi
xels (T
) that fa
ll between T
1
and T
2
are c
o
nsidere
d
as wea
k
.
Sel
e
c
t
i
on
of
t
h
resh
o
l
d val
u
es
(T
1 a
n
d
T
2
)
can
be
con
s
i
d
ere
d
as
a n
onl
i
n
ear
co
m
p
l
e
x pr
obl
em
. He
re
,
Evaluation Warning : The document was created with Spire.PDF for Python.
I
S
SN
:
2
088
-87
08
I
J
ECE
Vo
l. 6
,
N
o
. 5
,
O
c
tob
e
r
20
16
:
215
8
–
21
66
2
162
Cellu
lar Particle Swarm
Op
timizatio
n
(C
PS
O)
al
g
o
ri
t
h
m
i
s
i
m
pl
em
ent
e
d t
o
c
h
oose
t
h
r
e
sh
ol
ds
f
o
r
a
gi
ve
n
regi
onal
si
gn
l
a
ng
ua
ge i
m
ages.
Pr
ocedure
1 and
2
are done to
achieve
step 5.
Step 6:
Trac
ki
ng
o
f
e
dges
i
n
t
h
e ch
ose
n
i
m
ages ca
n
be
per
f
o
rm
ed by
usi
n
g
hy
st
eresi
s
t
h
r
e
sh
ol
di
n
g
t
e
c
h
ni
q
u
e
- Th
e seg
m
en
ted
im
ag
e is ob
tain
ed
as fi
n
a
l
ou
tpu
t
im
ag
e and
it is
u
s
ed fo
r
furth
e
r an
alysis.
Procedure 1: To
find the
r
a
nge of
th
reshold values
using his
t
ogr
a
m
method:
Step 1:
A re
gi
onal
si
g
n
l
a
n
g
u
age i
m
age da
t
a
set
i
s
gi
ven a
s
i
n
p
u
t
an
d t
h
e
edges
o
f
si
g
n
l
a
ng
uage i
m
ages are
pre
d
et
erm
i
ned
as R
e
gi
o
n
o
f
I
n
t
e
rest
(R
O
I
)
i
n
t
h
e dat
a
set
.
Step 3:
Inpu
t imag
e is conv
erted
in
to gray scal
e im
age to re
duce
the
size of the
im
age.
Step 4:
Re
peat : Identify t
h
e
histogram
h (z
) of a
n
im
age to be
segm
ented.
Step 5:
Th
e
p
r
o
b
a
b
ility o
f
a pix
e
l v
a
lue is
rep
r
esen
ted in
t
h
e equ
a
tio
n (9
):
P (z)
= p (z/backgr
o
und
pixel)
P (backgr
o
und
pi
xel)
+ p (z/object
pixel
)
P (
obj
ect
pi
xel
)
or
or
(9)
w
h
er
e
p
b
(
z
),
p
o
(
z
)
-
p
r
o
b
a
b
ility d
i
strib
u
tion
s
of
b
a
ck
gro
und
an
d ob
j
ect
p
i
xels
μ
b
,
μ
o
-
m
eans of
t
h
e
di
st
ri
b
u
t
i
ons
σ
b
,
σ
o
- st
an
dar
d
devi
at
i
o
ns of
t
h
e di
st
ri
b
u
t
i
o
ns
P
b
,
P
o
- a-prio
ri p
r
ob
ab
ilities o
f
b
a
ckg
r
o
und
an
d obj
ect p
i
x
e
ls.
Step 6:
Th
e
p
r
o
b
a
b
ility o
f
m
i
sclassificatio
n
o
f
an obj
ect
p
i
x
e
l as
b
a
ck
grou
nd
is exp
r
essed
in eqn
(1
0):
(10)
Step 7:
The math
em
at
ical eq
u
a
tio
n
for th
e p
r
ob
ab
ility o
f
b
ackgroun
d
p
i
x
e
ls, in
co
rrectly classified
as
o
b
j
e
ct
p
i
x
e
ls, is rep
r
esen
ted in
eqn
.
11
:
(11
)
Step 8:
The m
a
t
h
em
ati
cal
for
m
ul
a for t
h
res
hol
d sel
ect
i
on
i
s
obt
ai
ne
d
by
m
i
nim
i
zi
ng t
h
e
abo
v
e e
x
p
r
ess
i
on a
s
d
e
no
ted in
eqn
.
12
:
(12)
Step 9:
U
n
til th
e thresh
o
l
d
v
a
lu
es is
o
b
t
ained
for th
e im
ag
es fo
und
in reg
i
o
n
a
l
sign
languag
e
d
a
taset.
Procedure2 :
To find
optim
a
l thres
ho
ld va
lues (Lo
w
L a
n
d
High
H)
Step 1:
In
itializatio
n
of swarm
an
d
p
a
ram
e
ters - A
po
pu
latio
n
o
f
thresh
o
l
d
p
a
rticles says p
,
of size
m
(0.0
<=n
<
=1
.0
)
are g
e
n
e
rated
with
rando
m
p
o
s
itio
n
x
p
a
n
d
r
a
nd
om
vel
o
ci
t
y
v
p
.
Step 2:
Fi
t
n
es
s eval
uat
i
o
n -
Each
part
i
c
l
e
i
s
eval
uat
e
d wi
t
h
st
an
dar
d
be
nchm
ark f
u
nct
i
on
(g
ri
ewa
n
k
)
whi
c
h
h
a
s co
n
tinuo
us, d
i
ff
er
en
tiab
l
e, non
-
s
ep
ar
ab
l
e
, scalab
le an
d m
u
lt
i
m
o
d
a
l pr
op
er
ties
f
o
r
ch
oo
si
n
g
t
h
e
op
ti
m
a
l
t
h
res
hol
d
val
u
e
s
. T
h
e m
a
t
h
em
at
i
cal
equat
i
o
n
of
g
r
i
e
wa
nk
f
u
nct
i
o
n
i
s
re
p
r
es
ent
e
d i
n
e
q
n
.
13
.
∑
П
cos
√
1
(13)
Step 3:
Fi
nd
th
e so
cial best p
a
r
ticle -
Based
on
th
e ev
aluatio
n
of
f
itn
ess f
u
n
c
tion
(
m
in
i
m
a)
, a n
e
ighb
or
hood
best
(
S
oci
a
l
)
g
be
s
t
is chosen for each iteration.
Step 4:
C
o
m
p
are the
previ
o
us
value
g
besti-1
of th
e
p
a
rticle wi
th
th
e curren
t
particle (g
best
) valu
e.
If (
g
best-1
< g
best
)
Set G
best
=
g
best
;
Else
Set G
best.
= g
best-
1
;
Evaluation Warning : The document was created with Spire.PDF for Python.
I
J
ECE
I
S
SN
:
208
8-8
7
0
8
Im
pr
oved Canny E
d
ges
Using Cellular
B
a
s
e
d P
a
rticle
Sw
arm
O
p
timiz
ation Tec
hni
que
....
(
M
. Kris
hnaveni)
2
163
Step 5:
Fi
n
d
t
h
e so
cial b
e
st p
a
rticle - Id
en
tify th
e n
e
i
g
h
bou
rho
o
d
(social) p
a
rticle in
th
e
p
opu
latio
n
and
assi
gn
t
o
i
nde
x
va
ri
abl
e
g.
Step
6:
Velo
ci
ty an
d po
sition up
datin
g equ
a
tio
n
- Th
e v
e
l
o
c
ity and
p
o
sitio
n of a p
a
rticle are upd
ated
u
s
ing
t
h
e f
o
l
l
o
wi
n
g
m
a
t
h
em
at
i
c
al
gi
ven i
n
e
q
n
.
14
an
d
15:
V
ij
(t
+
1
)
=
ω
* K
*
V
ij
(t
)
+ (
φ
1
*
R
1
* (
P
best
– x
p
(t
)
) +
σ
* (
φ
2
* R
2
* (
G
best
-p
g
(t
)
)
)
(1
4)
X
ij
(t
+
1
)
= X
ij
(t
)
+ V
ij
(
t
+1
)
(1
5)
w
h
er
e
V
ij
=
v
e
lo
city of
j
th
p
a
rticle at i
th
iteratio
n
X
ij
= Po
sition
o
f
j
th
p
a
rticle at
i
th
iteratio
n
0.729
(i
.e)
ra
nd
om
( )
f
unct
i
o
n
ge
ner
a
t
e
a di
st
ri
but
e
d
ran
d
o
m
num
ber
w
h
i
c
h
l
i
e
s fr
om
0 t
o
1
ϕ
1
=
ϕ
2
= 2.0
R
1
, R
2
= ra
n
d
o
m
nu
m
b
er l
i
e
s
bet
w
ee
n
0 a
n
d
1
P
best
= Perso
n
al
(ind
iv
idu
a
l) b
e
st
G
best =
Neigh
bou
rho
o
d
(social) p
a
rticle b
e
st ,
σ
= m
o
tili
ty facto
r
Step 7:
R
e
peat
t
h
e st
eps fr
om
3 t
o
7 ei
t
h
er t
h
e m
a
xim
u
m
num
ber of i
t
e
rat
i
ons
or p
r
e
d
et
erm
i
ned con
d
i
t
i
ons i
s
to
b
e
m
e
t.
Step 8:
Continuous e
vol
ution of s
w
arm
in problem
space – Cellular
Par
ticle Swarm
Optim
ization (CPSO)
algo
rithm
offe
r
G
best
can
di
dat
e
sol
u
t
i
o
n
(l
ow
an
d
hi
g
h
t
h
res
hol
d
val
u
es
) a
f
t
e
r t
h
e
fi
t
n
ess
execut
i
o
n
o
f
1
0
0
0
0
iteratio
n
s
.
Step 9:
T
h
e re
sultant thres
h
o
l
ds (hi
gh
[T
1
]
and l
o
w [T
2
])
are applied for tracking
o
f
ed
ges i
n
re
gi
o
n
al
si
gn
l
a
ng
uage
i
m
age dat
a
set
by
usi
n
g
hy
st
ere
s
i
s
t
h
res
h
ol
di
n
g
.
Th
is effo
rt
h
a
s u
lti
m
a
tel
y
i
m
p
r
ov
ed
t
h
e temp
eram
en
t o
f
PSO al
g
o
rith
m
t
o
yield
h
i
gh
qu
ality resu
lts
i
n
si
gn i
m
ages. The su
bject
i
v
e eval
uat
i
on
re
sul
t
of can
ny
m
e
t
hod,
PS
O b
a
sed can
ny
and C
PSO
base
d
canny
are
gi
ve
n i
n
Fi
gu
re
2.
4.
R
E
SU
LTS AN
D ANA
LY
SIS
The ex
peri
m
e
nt
al
obse
r
vat
i
on i
s
dem
o
n
s
t
r
at
ed t
o
i
n
d
i
cat
e how t
h
e new va
ri
an
t
al
gori
t
h
m
out
perform
s
the Classical PS
O on all evaluated m
e
trics.
The ex
perim
e
ntal param
e
ters set were the sa
m
e
for
bot
h PS
O a
nd
C
PSO.
The
pa
rt
i
c
l
e
si
ze was
set
t
o
1
01 a
n
d
t
h
e i
n
ert
i
a
wei
ght
was set
t
o
0.
72
9.
The e
x
e
c
ut
i
o
n
o
f
bo
th PSO and
CPSO
were t
e
rm
in
ated
with 10
000
fitn
ess
ev
alu
a
tion
iteratio
n
s
.
Th
is effo
rt h
a
s u
lti
m
a
tely
i
m
p
r
ov
ed
th
e temp
eram
en
t o
f
PSO algo
rith
m
t
o
yield
h
i
g
h
qu
ality resu
lts
i
n
si
gn i
m
ages. The su
bject
i
v
e eval
uat
i
on
re
sul
t
of can
ny
m
e
t
hod, P
S
O
base
d can
ny
and C
P
S
O
base
d can
ny
are
gi
ve
n i
n
Fi
gu
re
2.
Th
e
propo
sed
work is im
p
l
emen
ted
in MATLAB and
Java and
it is tested
i
n
m
a
n
u
a
lly g
e
n
e
rated
Tam
i
l
Si
gn La
ng
ua
ge
(TSL
)
dat
a
set
s
wi
t
h
l
i
m
i
t
a
t
i
on o
f
si
ngl
e
ha
n
d
ed
si
gns
a
n
d
bl
ac
k
back
g
r
o
u
n
d
i
m
ages
.
Each
sign
is cap
t
ured with ten d
i
fferen
t si
g
n
e
rs.
It is
seen that cellular PSO has
fo
und bett
er a
v
era
g
e
sol
u
tions
th
an
t
r
ad
ition
a
l PSO
b
a
sed on si
m
ilarit
y
in
dex
an
d p
e
arso
n co
rrelatio
n coefficien
t m
e
tri
c
s fo
r
wh
ich
t
h
e eqn
(1
6)
an
d
(
1
7
)
whi
c
h a
r
e
gi
ve
n
bel
o
w a
n
d
t
h
e com
p
ari
s
o
n
o
f
t
h
e
res
u
l
t
s
i
s
sho
w
n i
n
Ta
bl
e 1.
(16)
whe
r
e,
µ
x
= m
e
an
of x, µ
y
=
mean
of y and c
1
= (k
1
* L)
2 ,
c
2
=
(k
2
* L)
2
K
1
=
0.01; k
2
=
0.
0
3
by
de
fa
ul
t
, L =t
he
dy
na
m
i
c range
o
f
t
h
e pi
xel
val
u
es;
B
y
defa
ul
t
L=
25
5
= va
riance
of
x ,
= vari
a
n
ce
of
y
,
σ
xy
= covariance
of x and y
(1
7)
whe
r
e, x
i
= in
ten
s
ity v
a
lu
es of i
th
pi
xel
i
n
im
age x, y
i
= i
n
ten
s
ity v
a
lu
es o
f
i
th
pi
xel
i
n
im
age y
, x
m
= m
ean
in
ten
s
ity v
a
lu
es of im
ag
e x
,
y
m
=
mean
in
tensity v
a
lu
es
o
f
imag
e y.
Wh
en th
ere is a
v
a
lue of
r
=1
, i
f
th
e
two
im
ages are abs
o
lutely identical,
r
= 0
if th
ey
are co
m
p
letely
u
n
c
o
r
related
an
d
r
= -1
if th
ey are co
m
p
letel
y
an
ti-
correlated.
Evaluation Warning : The document was created with Spire.PDF for Python.
I
S
SN
:
2
088
-87
08
I
J
ECE
Vo
l. 6
,
N
o
. 5
,
O
c
tob
e
r
20
16
:
215
8
–
21
66
2
164
Fi
gu
re
2.
Vi
s
u
al
C
o
m
p
ari
s
on
of
C
a
n
n
y
an
d
Opt
i
m
i
zed C
a
nny
wi
t
h
PS
O a
n
d
C
P
S
O
Fi
gu
re
3.
Ed
ge
det
ect
ed
f
r
om
Han
d
Si
g
n
Im
age
usi
n
g C
a
nn
y
(a),
PS
O
bas
e
d C
a
nny
(b
),
and
C
e
l
l
u
l
a
r
P
S
O
base
d C
a
n
n
y
(
c
)
5.
CO
NCL
USI
O
N
Edg
e
d
e
tection is th
e m
a
in
seg
m
en
tatio
n
p
r
o
cess carried
ou
t in
fi
n
d
i
ng
th
e bou
nd
ary of th
e
o
b
j
ect
s
with
in
th
e im
a
g
es. Can
n
y
edg
e
d
e
tection
is co
n
s
i
d
ered
to th
e b
e
st edg
e
d
e
tectio
n
m
e
t
h
od
an
d
it seem
s
to
pr
o
duce fal
s
e
det
ect
i
on i
n
n
o
i
s
y
envi
r
o
nm
ent
.
A
n
o
b
ject
i
v
e i
s
pro
p
o
se
d t
o
de
vel
o
p an o
p
t
i
m
i
zed
edg
e
detector t
h
at increases the loc
a
lizati
on accuracy of edge
de
tection, es
pecia
lly in real time digital im
ages. The
goal
o
f
im
pro
v
ed ca
nny
ed
g
e
s i
s
achi
e
ved
by
i
n
t
r
od
uci
n
g a hy
bri
d
o
p
t
i
m
i
zati
on t
ech
ni
q
u
e t
o
fi
n
d
opt
i
m
al
th
resh
o
l
d
v
a
l
u
es wh
ich
is a co
m
b
in
atio
n
o
f
PSO
with
cellu
lar o
r
g
a
n
i
sm in
sp
ired
fro
m
fib
r
ob
last. Th
e
pr
o
pose
d
a
p
pr
oach
has i
m
prove
d t
h
e ca
n
n
y
det
ect
o
r
t
o
c
o
nnect
t
h
e
sh
ort
ed
ge c
o
nt
o
u
rs
i
n
t
o
l
o
nge
r c
o
nt
o
u
r
s
,
an
d th
e ex
p
e
ri
men
t
al resu
lts
v
a
lid
ate its po
ten
tiality with
d
i
fferen
t
m
e
tri
c
s. Th
e resu
lts o
b
t
ai
n
e
d
will lead
to
hi
g
h
er acc
ura
c
y
of t
h
e cl
a
ssi
fi
cat
i
on m
e
t
h
o
d
, a
d
o
p
t
e
d
fo
r t
h
e
deve
l
opm
ent
of T
a
m
i
l
si
gn l
a
ngu
a
g
e
recognition sys
t
em
.
Evaluation Warning : The document was created with Spire.PDF for Python.
I
J
ECE
I
S
SN
:
208
8-8
7
0
8
Im
pr
oved Canny E
d
ges
Using Cellular
B
a
s
e
d P
a
rticle
Sw
arm
O
p
timiz
ation Tec
hni
que
....
(
M
. Kris
hnaveni)
2
165
Tabl
e
1.
C
o
m
p
ari
s
on
o
f
t
h
e t
h
e
pr
o
pose
d
m
e
t
h
o
d
base
d
on
pe
rf
orm
a
nce
m
e
t
r
i
c
s
ACKNOWLE
DGE
M
ENTS
Th
is wo
rk
is su
pp
orted
b
y
th
e p
r
o
j
ect en
titled
“Inv
est
i
g
a
tio
n
on
v
i
ew b
a
sed
m
e
th
o
d
o
l
og
ical
app
r
oaches
f
o
r Tam
i
l
Si
gn
l
a
ng
uage R
e
c
o
g
n
i
t
i
on”
(
N
o
.
F. 3
0
-
4
2/
2
0
1
4
(B
SR
))
fu
n
d
ed
by
U
G
C
– B
S
R
Research Start
-Up- Grant
.
REFERE
NC
ES
[1]
P
.
Ghate,
et al.
,
“An Introduction to the Signing S
y
stem for Indian Languages
,
”
Part II - Additional Signs. Bombay
,
Ali Yavar
Jung
National Institute fo
r
the Hear
ing
Handicapped
,
1
990.
[2]
M. Krishnaveni,
et al.
, “Efficien
t Removal of I
m
pulse Noise in Ta
mil Sign
Language Digital I
m
ages using PSO
Based Weighted
Median Filter,”
International
Journal of Applie
d Engin
eering
Research (
I
JAER)
,
vol.
10, pp
.
40474-40480, 2
015.
[3]
N. Salm
an, “
I
mage Segm
enta
tio
n
Based on Watershed and
Edge Detection Tech
niques,”
The International Arab
Journal of Infor
m
ation Techno
lo
gy,
vo
l. 3, pp. 10
4-110, 2006
.
[4]
S. Jansi and P. Subashini,
“Optimized Adaptive
Thresholding based Edge
Detection Method for
MRI Brain Imag
es,”
International Jo
urnal of
Computer Applications
,
vol. 51
, pp
. 1-8
,
2012.
[5]
J. Cann
y
,
“A C
o
mputational
Approach to
Edg
e
Dete
ct
ion
,”
I
EEE Transactio
n on Patt
ern A
nalysis and Ma
chine
Intell
igen
ce
, vol. 8, pp. 679-698,
1986.
[6]
H. Pan,
et al.
,
“
P
arti
cle
Swarm
Optim
izat
ion for
Function
Optim
iz
ation
in Nois
y
E
nvironm
ent,”
Jo
urnal
of
Applied
Mathematics an
d Computation
,
vol. 181
, pp
. 908
–919, 2006
.
[7]
L. I. Hong-qi,
et al.
, “An Impr
oved PSO-based of Harmony
S
earch for
Com
p
lic
ated
Optim
i
z
a
tion
P
r
oblem
s
,
”
International Jo
urnal of Hy
brid
Information
Technology
, vol. 1
,
pp. 91-98
, 2008
.
[8]
G. Coath and S. Halgamuge, “A Comparison of
Constraint-H
and
ling Methods for the Application
of Particle Swarm
Optim
izatio
n to
Constrained
Nonline
a
r Optim
iz
a
tion Probl
em
s,”
IEEE Congress on Evo
l
utionary Computation
, v
o
l.
4, pp
. 2419–242
5, 2003
.
[9]
C. R. Eberh
a
rt
and Y. Shi, “
P
a
r
ticl
e
Swarm
Optim
i
zation: Developments, App
lications and
Resources,”
IE
EE
Xplore
, vol. 1
,
p
p
. 81-86
, 2001
.
doi: http://dx.do
i.org/10.1109
/CEC.2001.934374
[10]
H. Stebbings, “Cell Motility
,
”
En
cyclopedia
of Life Scien
ces,
pp. 1
-
6, 2001
.
[11]
C. J. Dallon an
d A. J. Sherratt, “
A
Mathem
ati
cal Model for Fibroblast and C
o
llag
e
n Orientation,”
Bu
lle
tin of
Mathematical Biology
, vol. 60
, p
p
.101-129, 1998
.
T
a
m
i
l constants
(
E
ach sign with 10
differ
e
nt signer
s
)
Si
m
ila
rity Index
(m
e
a
n
)
Pearson correla
tion coef
f
i
cient
(m
e
a
n
)
Canny
PSO
optim
ized
canny
CPSO
optim
ized
canny
Canny
PSO
optim
ized
canny
CPSO
optim
ized
canny
1.
5750
3
1.
5752
2
3.
3551
4
0.
7139
0.
868
1.
1186
1.
0735
2
1.
0737
2
2.
4370
6
0.
4585
0.
6082
0.
7793
1.
4473
9
1.
4476
4
3.
2678
0.
7026
0.
8778
1.
0844
1.
4929
1.
4930
7
3.
1572
1
0.
5259
0.
6666
0.
8917
1.
2132
4
1.
5116
3
3.
7415
4
0.
4583
0.
5498
0.
7342
1.
3297
9
1.
4486
6
3.
5851
1
0.
7213
1.
0103
1.
458
1.
4353
6
1.
3425
8
3.
4457
8
0.
5948
0.
8514
1.
0131
1.
2232
3
1.
2819
8
2.
9311
6
0.
6134
0.
7649
1.
0048
1.
1094
2
1.
1096
1
2.
4100
7
0.
4802
0.
6278
0.
8325
0.
9320
3
0.
9341
2.
1909
8
0.
3101
0.
4441
0.
6134
1.
0508
6
1.
0776
5
2.
7044
2
0.
33
0.
5148
0.
8615
1.
0280
8
1.
0282
9
2.
3565
6
0.
4201
0.
5866
0.
8375
0.
4568
9
0.
4570
4
1.
2112
9
0.
3133
0.
4491
0.
5777
1.
0977
9
1.
0416
2
2.
1094
8
0.
554
0.
7892
1.
1656
0.
7318
8
0.
7320
2
1.
6816
5
0.
3161
0.
4448
0.
6104
0.
7868
6
0.
8376
9
2.
0851
8
0.
5021
0.
7609
0.
9824
1.
0613
4
1.
0618
5
2.
4127
4
0.
4191
0.
6127
0.
8114
1.
1266
9
1.
2616
4
2.
5747
9
0.
4303
66
0.
7424
0.
9927
Evaluation Warning : The document was created with Spire.PDF for Python.
I
S
SN
:
2
088
-87
08
I
JECE
Vo
l. 6
,
N
o
. 5
,
O
c
tob
e
r
20
16
:
215
8
–
21
66
2
166
BIOGRAP
HI
ES OF
AUTH
ORS
Dr
.
M.
K
r
ishnave
ni
is
an
As
s
i
s
t
ant P
r
ofes
s
o
r in th
e Dep
a
r
t
m
e
nt of Com
p
uter S
c
ienc
e,
Avinashilingam
Universit
y
for
W
o
m
e
n, Coim
bato
re. She has
7
y
e
ars of research exper
i
ence
working as
a
res
earch
er
in Nava
l
Res
ear
ch Boa
r
d
,
Defen
c
e
Res
e
a
r
ch Deve
lopm
ent Organis
a
tion,
Ministr
y
of Def
e
nce
.
She was been a consult
a
nt
for m
ilitar
y
r
e
la
ted appli
c
a
tion p
r
ojec
t. She has
published sever
a
l papers in her research spec
ialization with an overall coun
t of 61 both
nation
a
ll
y
and i
n
terna
tiona
ll
y. S
h
e focuses on re
s
earch and
expe
rim
e
ntal pro
ces
s
e
s
with the us
e
of digital
techno
logies such
as MATLAB, Com
puter Science, B
i
ologica
l computation along with
m
achine in
tel
lig
ence
and rea
l
-t
i
m
e ph
y
s
ic
al co
m
putation. S
h
e
als
o
acts
as
a r
e
s
ource pers
on an
d
holds nation
a
l
and intern
ational
funded research
projects which
come under University
Gran
ts
Commission (India)
and Dep
a
rtme
nt of Science
and Technolo
g
y
.
Dr
.
P.
Subash
ini
rec
e
iv
ed a B
.
S
c
. (M
a
t
hem
a
ti
cs
) Degree
from
Bharath
i
ar Uni
v
ers
i
t
y
,
Tam
i
l
Nadu, India in 1
990 and M.C.A
.
Degree from th
e
sa
me University
in 1993. She h
a
s also receiv
e
d
Degrees of
M.Phil. and Ph
.D. r
e
spectively
in
C
o
mputer Scien
c
e in
the
y
e
ars 1
998 and 2009
respectively
fro
m Avinashilingam University
f
o
r Women, Tamil Nadu, India. From 1994 to
2007, she worked for the Dep
a
rtm
e
nt of Com
put
er Scien
c
e,
Avinashilingam
Universit
y
for
W
o
m
e
n, where
s
h
e is
cur
r
ent
l
y
in th
e pos
it
io
n
of Professor. She has also
held short-ter
m
appointments at
several institutio
n
s around the State. She is a member of the editorial board of
several in
terna
t
i
onal journals
, an
d reviews
s
e
vera
l conferen
ces
. S
h
e has
been invi
ted as
s
p
eaker
to sever
a
l workshops and org
a
nized
intern
ationa
l workshops and
special sessions
at
confer
ences
.
T.
T.
Dhivy
a
prabha
has com
p
l
e
ted M.Sc.
,
M.Phil. in Bhara
t
hiar University
. She has 4 y
e
ars of
teaching
experience. She is pursu
ing Ph.D. (Com
put
er Science) p
r
ogramme
in the Department of
Com
puter S
c
ien
ce,
Avinas
hil
i
ng
am
Univers
i
t
y
f
o
r W
o
m
e
n, Coim
batore. H
e
r a
r
eas
of in
ter
e
s
t
includ
e Com
put
ation
a
l In
tel
lig
en
ce,
Optim
iz
ation
Te
chnique
and
I
m
age Processing
.
Evaluation Warning : The document was created with Spire.PDF for Python.