TELKOM
NIKA
, Vol.13, No
.1, March 2
0
1
5
, pp. 118~1
2
7
ISSN: 1693-6
930,
accredited
A
by DIKTI, De
cree No: 58/DIK
T
I/Kep/2013
DOI
:
10.12928/TELKOMNIKA.v13i1.1264
118
Re
cei
v
ed O
c
t
ober 1
5
, 201
4; Revi
se
d Decem
b
e
r
17, 2014; Accept
ed Ja
nua
ry 6,
2015
Image E
dge Feature Extraction and Refining Based on
Genetic-Ant Colony Algorithm
Xing Zhang*,
Shuai Liu
Dep
a
rtment of Comp
uter Scie
nce an
d Eng
i
n
eeri
ng, Hen
an
Univers
i
t
y
of Urban C
onstructi
on, Ping
di
ngsh
an
467
06
4, Hena
n, Chin
a
*Corres
p
o
ndi
n
g
author, e-ma
i
l
: zhang
x@hnc
j.edu.cn
A
b
st
r
a
ct
Edge
is co
mp
o
s
ed by a c
o
ll
ec
tion of its ne
ar
by pi
xe
ls w
h
ich
has a step c
h
a
nge
or cha
nge
s in roof,
an i
m
a
ge is an
informatio
n
system a
nd
most
of its
informati
on co
mes fro
m
the edges. Thi
s
paper g
i
ves
a
brief overv
i
ew
of the status and
th
e i
m
port
ance of i
m
age
edge
detecti
o
n
and
introd
uc
es the rese
arc
h
status of the
i
m
a
ge
ed
ge
de
tection. After t
hat, it i
n
troduc
es the
bas
ic pr
incip
l
e
an
d th
e
main
steps
of
the
gen
etic alg
o
rit
h
m a
nd a
n
t colo
ny alg
o
rith
m. On t
he ba
sis of these, the pa
per pro
p
o
sed a n
e
w
hybrid
alg
o
rith
m for th
e i
m
a
ge
ed
ge
extraction
an
d
refini
ng,
w
h
ich
combi
ned
the
gen
etic a
l
gor
ith
m
and
a
n
t col
o
ny
alg
o
rith
m. T
h
r
oug
h the
a
nal
ysis of the
ti
me-spe
ed
gr
ap
h
of the
ge
neti
c
alg
o
rith
m
an
d the
ant c
o
lo
ny
alg
o
rith
m, w
e
can fi
nd
the
b
e
st fusio
n
poi
n
t
betw
een
the
gen
etic
alg
o
rit
h
m a
n
d
the
an
t colo
ny
alg
o
rit
h
m.
T
he exp
e
ri
me
n
t
indicat
ed the
prop
os
ed
hybri
d
al
gorith
m
c
a
n make th
e fu
ll
use of the i
m
age
infor
m
ati
o
n,
the si
mu
latio
n
time
is sh
orter, the i
m
a
ge
ed
ge is
mo
re c
o
ntinu
ous, a
nd
preserv
ed th
e
outli
ne
of orig
i
n
a
l
imag
e more co
mp
lete
ly.
Ke
y
w
ords
: image e
d
g
e
dete
c
tion, gen
etic a
l
gorit
hm, a
n
t colo
ny alg
o
rith
m
1.
Introduc
tion
Edge refe
rs t
o
the colle
cti
on of pixels the gr
ay level
of the surro
undin
g
pixels of which
has
step
cha
nge o
r
ro
of chang
e. Edge
is the im
po
rt
ant informati
on in the ima
ge as
well a
s
a
signifi
cant cl
ue of visual
perception.
Actually
, it
can not only
convey most of the image
informatio
n, but it is also
an importa
n
t
f
oundation
of image ana
lysis an
d ma
chin
e vision. An
image is a
n
informatio
n system and a
great deal o
f
its information is p
r
ovide
d
by its cont
ou
r
edge
s. Therefore, edg
e feature
extra
c
tion and d
e
tection play
a significa
n
t
role in image
pro
c
e
ssi
ng [1
]. Although active rese
arch
has b
een m
a
de on e
dge d
e
tection al
go
rithm in the lo
ng
term, the ed
ge dete
c
tion
methods in
cre
a
se day after day an
d nume
r
ou
s edge extra
c
tion
operators ha
ve been co
me up with
grad
ually,
the traditional
Rob
e
rts o
p
e
r
ator and So
bel
operator h
a
ve been rarel
y
used alon
e
[2]. Currentl
y
, an increa
sing numb
e
r
of sch
olars h
a
ve
applie
d bioni
cs algo
rithm in
image edg
e detectio
n
.
In bioni
cs alg
o
rithm, the
e
dge d
e
tectio
n
based
on g
e
netic al
go
rith
m and
that b
a
se
d on
ant colony
al
gorithm
have
their own a
d
v
antage
s a
n
d
disadvanta
g
e
s. In th
e o
p
e
ration
of g
e
netic
algorith
m
, it can be fou
nd t
hat the indivi
dual
s of
the
popul
ation ha
ve a relativel
y
rapid iteration
spe
ed at the
initial stag
e; however, a
s
the algo
rith
m pro
c
e
e
d
s
to the later
p
e
riod,
red
und
ant
iteration e
m
e
r
ge
s obvio
usl
y
and the ite
r
ation
spe
e
d
at this time
is very
slow [3]. As for
ant
colo
ny algorit
hm, in its early phase, the
conver
gen
ce
spee
d of the system is
slo
w
due to the
lack of phe
ro
mone; neve
r
thele
ss,
the converg
e
n
c
e speed
a
c
cele
rate
s o
n
acco
unt of the po
sitive
feedba
ck of t
he alg
o
rithm
at the late
ph
ase
and
ant
colo
ny algo
rit
h
m ha
s
excel
l
ent pa
ralleli
sm
and global search ability.If
the
advan
tages of
geneti
c
algorithm
are to
be integrat
ed
with those of
ant col
ony al
gorithm, the
perfo
rman
ce
of geneti
c
-a
n
t
colony
hybri
d
algo
rithm
can be
imp
r
ov
ed
[4],[5].
This p
ape
r
mainly appli
e
s ge
netic-ant
col
ony hyb
r
i
d
algo
rithm i
n
to the imag
e edge
feature
dete
c
tion and
refin
i
ng. Th
roug
h
the analy
s
is
of time-spee
d cu
rve
s
of g
enetic
algo
rit
h
m
and ant colo
ny algorithm,
it can be fou
nd out t
he op
timal combi
n
ation of gene
tic algo
rithm and
ant colo
ny algorithm a
nd
then de
sign
s the al
gorith
m
for image
edge d
e
tecti
on, inclu
d
ing
the
operation
ste
p
s of ge
netic
algorithm
at the early
stag
e; the
integra
t
ion of geneti
c
algo
rithm a
nd
ant colony
al
gorithm
a
c
cording to
the
se
tting co
ndi
tio
n
s
and
the
op
eration
of th
e
later ant
col
o
ny
algorith
m
. At the beginni
n
g
of the prop
ose
d
gen
et
ic-ant colo
ny hybrid algo
rith
m, it adopts the
operation ste
p
s
of gen
etic algorith
m
an
d
end
s it withi
n
the sco
pe of
its gen
etic ite
r
ation
s
. The
n
it
Evaluation Warning : The document was created with Spire.PDF for Python.
TELKOM
NIKA
ISSN:
1693-6
930
Im
age Edge Feature Extraction and Refi
ning Bas
ed on Geneti
c
-Ant
Colony .... (Xing Zhang)
119
transfo
rm
s it
s
current
op
timal sol
u
tio
n
a
s
the m
a
trix dist
ribut
ion of the
i
n
itial phe
rom
one
con
c
e
n
tration
of ant colon
y
algorithm
a
nd op
er
ate
s
ant col
ony al
gorithm. Afte
r the al
gorith
m
end
s, output the edg
e imag
e.
2.
Image Edge Detec
t
ion
Most im
po
rta
n
t inform
atio
n of the
ima
ge exi
s
ts i
n
t
he im
age
ed
ge, which i
s
the mo
st
promi
nent i
n
the ima
ge l
o
cal
ch
ang
es.
It reflect
s
the
feature diffe
rences withi
n
theimage
lo
cal
area
s an
d it is indi
cated a
s
ce
rtain di
scontinuity
of the image info
rmation [6]. Edge mainly e
x
ists
betwe
en obj
e
c
tive and o
b
j
e
ctive, obje
c
tive and ba
ckg
r
oun
d a
s
well as a
r
ea a
nd
area
(in
c
ludi
n
g
different
colo
rs) a
nd it i
s
animp
ortant
found
ation
of the ima
g
e
an
alysis such
a
s
ima
g
e
segm
entation
,
texture fea
t
ures an
d shape fe
at
ure
s
[7]. Edge
is mai
n
ly expre
s
sed
as t
h
e
discontin
uity of imag
e lo
cal cha
r
a
c
teri
stics a
n
d
it
contai
ns the
rel
a
tively se
vere
gray
le
vel
cha
nge
s in t
he imag
e, that is, the pla
c
e
withsi
n
gul
ar chan
ge
s in the sig
nal.
The gray le
vel
cha
nge of si
ngula
r
sig
nal
becom
es in
crea
singly
d
r
a
m
atic alon
g the edg
e. We
generally divide
the edge into
two types: ste
p
and ro
of, which a
r
e indi
cated in Figu
re
1 and Figu
re
2 [8].
(a) Step, n
a
m
ely the ima
ge inten
s
ity has
signi
ficant
differen
c
e
s
betwe
en the
pixel gray
levels of two
discontin
uou
s position
s
[9].
(b)
Ro
of, na
mely the ima
ge inten
s
ity sudde
nly ch
an
ges f
r
om
one
value to a
n
o
t
her a
n
d
then retu
rn
s to the origin
al value after m
a
intaining a
small sched
ule
[10].
Figure 1. Step edge
Figure 2. Roo
f
edge
The imag
e ed
ge dete
c
tion
dire
ctly affects t
he effects
of the entire image p
r
o
c
e
s
sing.Th
e
traditional ed
ge
dete
c
tion method
s
a
r
e norm
a
lly
ba
sed
on su
ch single
featu
r
e asfirst-o
r
de
r or
se
con
d
-o
rd
er derivative of
the edge
nei
ghbo
rho
od a
nd they are n
o
t sen
s
itive to the imag
e with
fuzzy e
dge
s. Beside
s, the
uncertainty of
the
obje
c
t b
ound
ary in th
e image i
s
u
s
ually fuzzin
e
ss
[11].
Evaluation Warning : The document was created with Spire.PDF for Python.
ISSN: 16
93-6
930
TELKOM
NIKA
Vol. 13, No. 1, March 2
015 : 118 – 1
2
7
120
3.
Feasibilit
y
A
n
aly
s
is of Int
e
gratio
n of
Genetic Alg
o
rithm & An
t Colon
y
Alg
o
rithm
3.1. Principle of Gen
e
tic
Algorithm
As a
rando
m an
d hi
gh-efficient gl
ob
al se
a
r
ch
m
e
thod to
sol
v
e problem
s,
gen
etic
algorith
m
ne
eds n
o
prio
r kno
w
le
dge of
its search
space, instead
, it appr
oximates the opti
m
al
solutio
n
step
by step through the inf
o
rmat
io
n exchang
e betwe
en individual
s, by rando
mly
gene
rating
th
e initial po
pul
ation in th
e search
sp
ace
and
with the
help of th
e o
b
jective fun
c
ti
on
[12].
(i)
The
code
s ca
n p
e
rfo
r
m
co
rrespon
di
ng
codi
ng
on
the
pro
b
lem
s
to
be
settled. Since
the image
gray level value
is bet
wee
n
0
-
255, the
c
o
d
e
of every chromosome i
s
an 8-digit bin
a
ry
cod
e
.
(ii) If the
po
pulation
mod
e
l ha
s ex
ce
ssive
po
pulati
ons, the
am
ount of
cal
c
u
l
ation of
every ge
ne
ra
tion of fitne
s
s valu
e i
s
h
u
ge; ho
weve
r,
if the
popul
ation i
s
small
,
it is
difficult
to
demon
strate spe
c
ie
s diversity. Therefo
r
e,
set the pop
ulation num
b
e
r re
asona
bly.
(iii) Fitne
s
s functio
n
is th
e stand
ard to
evaluate e
v
ery individu
al and it nee
ds to b
e
sele
cted to
show th
e evol
ution tre
nd.
Here, choo
se
the fitness f
unctio
n
a
s
th
e ba
sis for
a
n
t
colo
ny para
m
eter sel
e
ctio
n
.
(iv) Co
mplete
ly copy the i
ndividual
s wit
h
high fitne
s
s fun
c
tion in
the popul
atio
n to the
next-gen
eration p
opul
atio
n. Set a
ra
n
dom fu
ncti
o
n
and
sele
ct t
he
copi
ed i
n
dividual
s
with
th
e
widely-used
sele
ction
met
hod, n
a
mely
Roul
ette wh
e
e
l sele
ction i
n
the
ant
col
ony alg
o
rithm
to
mak
e
it as
the s
e
lec
t
ed operator.
(v)Cro
ssover
use
s
the met
hod
of si
ngle
-
point cro
s
sover. Select two ch
romo
so
m
e
s. Set
a rand
om fun
c
tion to determine a com
m
on point of
these two chrom
o
some
s randomly a
s
the
cro
s
sove
r poi
nt of these chrom
o
some
s. After t
he cro
s
sover, keep
the positio
ns
of the node
s
of
these t
w
o
chrom
o
some
s from the
source p
o
int
to the cro
s
sover poi
nt u
n
ch
ang
ed th
e
interchan
ge the nod
e po
sitions fro
m
the cro
s
sove
r poi
nt to the destination poi
nt.
(vi) Mutation
determi
ne
s the local
sea
r
ching ability of geneti
c
ope
ra
tion.
The flowchart
of genetic al
gorit
hm i
s
as
follows[13],[14].
Figure 3. The
flowch
art of geneti
c
algo
ri
thm
Evaluation Warning : The document was created with Spire.PDF for Python.
TELKOM
NIKA
ISSN:
1693-6
930
Im
age Edge Feature Extraction and Refi
ning Bas
ed on Geneti
c
-Ant
Colony .... (Xing Zhang)
121
3.2. Concr
e
te Realization
of Ant
Colo
n
y
Algorithm
Many
kind
s
of ants relea
s
e
phe
rom
o
n
e
an
d p
h
e
r
o
m
one
directi
on in
food
foragi
ng.
Den
eub
ourg explain
s
this kind of
path selectio
n
mo
d
e
ba
sed
on
self-organi
zati
on with
a
sim
p
le
test model. In
this mode, a
bridg
e
with two bypa
sse
s
of equal len
g
th sep
a
rate
s the ant ne
st and
the food. At the begi
nni
ng, the ant
s sele
ct fr
om
these t
w
o
paths
acco
rd
ing to the
same
prob
ability sin
c
e the
r
e is n
o
pheromon
e distrib
u
tion
in
the paths. Wi
th the introdu
ction of ra
ndo
m
fluctuation, m
o
re a
n
ts will
cho
o
se one
path,
whi
c
h i
n
crea
se
s the
pheromon
e con
c
e
n
tration
of
this path; therefore, it will attract
more ants to select this path.
In Dene
ubou
rg’s exp
e
rim
ent, the left-over
phe
romo
ne co
ncentra
tion on the path is i
n
dire
ct propo
rti
on to the
nu
mber
of ant
s
and n
o
p
hero
m
one vol
a
tilization is taken
into a
c
count.
In
this sim
p
lified
model, the a
n
t choo
se
s a
path acco
rdi
n
g to the total quantity of the ants p
a
ssin
g
this path. Assume that
i
A
and
i
B
are the num
bers of ants to sele
ct path
A
and
B
re
spe
c
ti
v
e
ly
after the
ith
ant
cro
s
se
s the
b
r
idge, th
en th
e proba
bilitie
s of th
e
(1
)
it
h
than
t
t
o
ch
oo
se p
a
t
h
A
and
B
are
()
1
()
()
n
i
AB
nn
ii
kA
PP
kA
k
B
(1)
Formul
a(1) h
a
s
qua
ntize
d
this sel
e
cti
on m
ode. P
a
ram
e
ter
n
determin
e
s the
non
-
linearity of th
e sel
e
ctio
n fu
nction.
Whe
n
n
is big
ger, the
pheromo
ne concentratio
n
of one p
a
th is
only a little higher tha
n
tha
t
of the other path,
then the prob
ability for the next a
n
t to choo
se the
p
r
e
v
io
us
pa
th is
b
i
gg
er
. Pa
r
a
me
ter
k
refle
c
ts the
attra
c
tion of the u
n
m
arked
path.
If
k
is bigge
r,
then it ha
s h
i
gher
re
quire
ments
on the
necessa
ry
p
hero
m
on
e co
nce
n
tration t
o
pe
rform
no
n-
rand
omi
z
ed sele
ction.
T
h
i
s
kind of
probability
ex
p
r
ession fo
rm
is a
c
tually in
ferre
d fro
m
the
actual ant p
a
th sele
ction
experiment.
T
he para
m
e
t
ers suitable
for experim
e
n
t are
2
n
and
20
k
. If
&2
0
ii
i
AB
A
, then
1
A
P
, if
ii
A
B
but
20
i
A
, then
0.5
A
P
. T
he dyn
a
mi
c
sele
ction result of the system is
dete
r
mi
ned by the followin
g
formul
as
1
1,
,
iA
i
iA
A
if
P
A
A
if
P
1
1,
,
iA
i
iA
Bi
f
P
B
Bi
f
P
(2)
Her
e
,
ii
A
Bi
and
is
a rand
om vari
able unifo
rml
y
distributed
within [0,1].
Figure 4. Prin
ciple of ant p
a
th sea
r
ch
This expe
rim
ent desig
ned
on equi-a
rm
bri
dge
s ca
n also be exte
nded to non
-equi-arm
bridg
e
s.
As i
ndicated i
n
F
i
g.4, the m
e
thod fo
r th
e a
n
t to
sele
ct t
he p
a
th i
s
b
a
s
ically the
sa
me
with the abov
e pro
c
e
ss. Among the ant
s from the
an
t nest, the ants passin
g
by the shorte
r p
a
th
Evaluation Warning : The document was created with Spire.PDF for Python.
ISSN: 16
93-6
930
TELKOM
NIKA
Vol. 13, No. 1, March 2
015 : 118 – 1
2
7
122
will re
ach the
food sou
r
ce
and g
o
ba
ck
to the ne
st e
a
rlie
r. Then
the ph
ero
m
on
e on the
sh
orter
path will be e
nhan
ce
d at a sho
r
ter time,
cau
s
in
g more
ants to cho
o
s
e this
sho
r
te
r path [15].
The above
circum
stan
ce
s haven’t taken
the cons
tituti
on, place
m
en
t and volatilization of
pheromo
ne. There are va
riou
s diversifi
ed act
ual
ph
erom
one left-over metho
d
s in the n
a
tu
ral
environ
ment.
The research p
e
rson
ne
l has
onl
y
obtaine
d so
me funda
me
ntal feature
s
of
pheromo
ne throu
gh expe
riment, incl
u
d
ing, vola
tilization rate, a
b
so
rption
rat
e
and diffu
si
on
coeffici
ent. T
he research
prove
s
that
p
hero
m
on
e ca
n be
pre
s
e
r
v
ed on
the
pat
h for
a lon
g
ti
me
,
whi
c
h ha
s a
clo
s
e relation
ship
with the
type,
popula
t
ion rule
and
environm
ent
al co
ndition
s
of
ants.
3.3. Feasibilit
y
Analy
s
is
Geneti
c
algo
rithm is a universal metho
d
to
solve opti
m
ization p
r
o
b
l
ems, whi
c
h h
a
s be
en
widely used in variou
s optimization p
r
ob
lems. Ho
wev
e
r, becau
se o
f
its str
ong un
iversality, it h
a
s
relatively ba
d
flexibility. For la
ck of infini
te appl
i
c
ation
ran
ge, alth
o
ugh it
can
en
sure that it
h
a
s
global
convergen
ce ability on
m
o
st pr
obl
ems, l
o
cal d
e
gene
ration
is
difficult to
avo
i
d an
d it fail
s t
o
pro
c
e
ss th
e feedb
ack info
rmation of the
system
. The
calculation e
fficiency redu
ce
s wh
en hu
ge
inefficient an
d red
unda
nt iteration
s
re
p
eat wh
e
n
se
arching to a
certai
n deg
ree. Ant colo
ny
algorith
m
a
c
h
i
eves th
e
purpose to
conv
erge
to th
e
o
p
timal p
a
th t
h
rou
g
h
the
a
c
cumulatio
n
and
updatin
g
of pheromo
ne and
it
h
a
s distrib
u
tivity
,
paralleli
sm and
gl
obal
sea
r
ching
ab
ility;
neverthel
ess,
it is sho
r
t of
pherom
o
ne
a
t
the e
a
rly
sta
ge a
n
d
its
sol
v
ing efficie
n
cy is
not hi
gh.
In
orde
r to
co
m
p
lement
ea
ch
other’
s
adva
n
tage
s of
th
e
s
e t
w
o
algo
rithms, it
can
b
e
foun
d fro
m
the
resea
r
ch of these two al
g
o
rithm
s
that they pre
s
e
n
t the Spee
d-Ti
me Cu
rves i
n
Fig. 5 in sol
v
ing
probl
em
s [16].
Figure 5. Speed-time
curve
of ant co
lony
algorithm a
n
d
geneti
c
alg
o
rithm
The genetic algorithm ha
s a
fast solving
speed at the early solving peri
od (
durin
g
0
~
a
tt
), but it
can be
seen that after
a
t
, the
speed d
r
ops quickly. On
the contrary,
the ant
colony algo
ri
thm (during
0
~
a
tt
) is
slow in its searching due to the
relative sho
r
tage of
pher
omone
at the initial
searching a
nd t
he long pher
omone
accumulation time; however,
when
the p
heromo
ne is
accumulated to a certa
i
n deg
ree
(
a
fter
a
t
), it accelerates its
conver
gence
to the o
p
timal solution. The b
a
sic
principle of t
he integ
r
ate
d
algo
rithm
is:
gene
rate the
initial phero
m
one conce
n
tration th
ro
ugh g
enetic algorithm be
fore the o
p
tima
l
integration p
o
int ( point
a
); fully use the rapid
ness, r
and
omness and fast conve
r
ge
nce of
genetic algorithm and after the optimal point,
utilize the parallelism,
positiv
e feedback and
high solving
efficiency to
seek t
he optimal solution
to the proble
m
.
Evaluation Warning : The document was created with Spire.PDF for Python.
TELKOM
NIKA
ISSN:
1693-6
930
Im
age Edge Feature Extraction and Refi
ning Bas
ed on Geneti
c
-Ant
Colony .... (Xing Zhang)
123
4.
Strateg
y
of
Mutual Inte
g
r
ation of
Ge
netic Algo
rithm and Ant
Colon
y
Algorithm
It is the
key
for the
algo
ri
thm of this p
aper to
see
k
the o
p
timal
con
n
e
c
tion ti
me for
geneti
c
algo
ri
thm and ant colo
ny algorit
hm and the
followin
g
dyn
a
mic integ
r
ati
on strate
gy can
achi
eve the
purp
o
se of
d
y
namically m
u
tual in
teg
r
at
ion of g
eneti
c
al
gorith
m
and
ant colo
ny
algorith
m
.
At the begin
n
ing of gen
etic-a
nt colo
ny hybr
id alg
o
ri
thm, cond
uct
genetic al
go
rithm till
the set termi
n
ation co
nditio
n
. Tran
sfer th
e cu
rre
nt
opti
m
al sol
u
tion from ge
netic
a
l
gorithm a
s
th
e
distrib
u
tion
m
a
trix of the
ini
t
ial phe
rom
o
n
e
con
c
ent
rati
on
of ant col
o
ny
algo
rithm. Implement ant
colo
ny algorit
hm until the optimal solutio
n
is so
ught.
In this pap
er, max-min
ant-colony
algorithm
(M
MAS) is ad
opted to up
date the
pheromo
ne. Whe
n
setting
the
initial ph
erom
one of
a
n
t col
ony alg
o
rithm,
set th
e initial valu
e
of
the pheromo
ne on every p
a
th as the ma
ximum value
ma
x
t
and the initial value
0
can be
set as a
fixed numbe
r, namely 100.
Acco
rdi
ng to the distri
but
io
n of the initial phero
m
on
e tran
sferred fro
m
the previou
s
geneti
c
algo
ri
thm, we ca
n set the initial value of the phero
m
on
e in resou
r
ce
i
as:
0
()
()
sG
ii
tt
(3)
In Formul
a (3),
0
is a
con
s
tant of phero
m
one, eq
ual
to
ma
x
t
in MMAS.
()
G
i
t
is the
pheromo
ne value tra
n
sfe
r
red from the
se
con
d
-b
est
solutio
n
obtai
ned from th
e
early gen
etic
algorith
m
.
The re
alizatio
n step
s of ge
netic-ant col
o
ny hybrid alg
o
rithm are as
follows:
(i)Th
e
co
nst
r
uction of the i
n
itial populati
o
n
We have exp
r
esse
d
eve
r
y
paramete
r
with
16-digit bi
nary
codi
ng
and
we
have
had
a
grou
p of para
m
eters (4 pa
rameters) fo
r every indivi
d
ual; therefore
,
we indicate
every individ
ual
in the initial populatio
n with 64digit bina
ry codi
n
g
. 1 to 16digit is th
e value of pherom
one imp
a
ct
fac
t
or
of state
transfe
r p
r
ob
ability; 17 to 32digit is th
e
value of the i
m
pact fa
ctor
o
f
heuri
s
tic
dire
ction fun
c
tion, 33 to 48digit is the val
ue of the phe
romo
ne volatilization
coefficient
while 49
to 64digit is the numb
e
r of
iterations
N
.
(ii)Th
e de
sign
of objective functio
n
Ran
domly pu
t
m
artific
i
al ants
in
n
pixels a
nd evaluate
the effects o
f
the final edge
extraction im
age ge
ne
rate
d by every in
dividual
with
obje
c
tive function. When
a co
rre
sp
ond
ing
grou
p of para
m
eters to the individual obt
ains t
he final edge ima
ge throu
gh edg
e extraction a
n
d
if
the
iterative edge
im
age has small differen
c
e with
the origi
nal i
m
age, na
mel
y
that the two
image
s are b
a
si
cally the same, it prove
s
that
the ed
ge of the obj
ective obje
c
t has b
een fou
nd.
Therefore, we introdu
ce th
e simila
rity
2
s
td
value of two i
m
age
s for ev
aluation.
The
obje
c
tive functio
n
i
s
e
x
presse
d a
s
2(
)
Objv
std
u
. He
re,
u
is th
e
energy difference
betwe
en the i
t
erative edge
image an
d the origin
al ima
ge, namely
12
uI
I
.
(iii)The sel
e
ct
ion of the next pixel to visit
To every artificial ant,
k
G
is the colle
ction of node
s wh
i
c
h haven’t
been visited
before
.
k
S
is the collecti
on of node
s
whi
c
h allo
w to be vi
sited i
n
the next step acco
rdin
g
to proba
bility
transfe
r fo
rm
ula.
k
tabu
is the
search
tab
u
table, whi
c
h
i
s
the
colle
ction
of node
s whi
c
h
have
been visite
d before.
0
arg
m
a
x
(
,
)
(
,
)
,
0,
j
a
llowe
d
ij
ij
i
f
q
q
Ci
j
otherwise
(4)
In Form
ula
(4
),
ij
C
is a
state
variabl
e
which
is o
b
taine
d
from the
tran
sfer form
ula of
ant
colo
ny
,
0
q
is a consta
nt with its value rang
e within0
~
1 a
nd
q
is a ran
d
o
m
numbe
r an
d
(0
,1
]
q
,it
is ge
nerated
rand
omly wh
en the a
n
t co
lony sele
cts the next pix
e
l. Whe
n
q
is smaller than or
equal to
0
q
,
t
r
a
n
sf
er
by
sel
e
ct
ing
t
he n
e
x
t
sho
r
t
e
st
pix
e
l;
whe
n
q
is
b
i
gg
er
th
an
0
q
,
sel
e
ct
according to
Formul
a (5
).
Evaluation Warning : The document was created with Spire.PDF for Python.
ISSN: 16
93-6
930
TELKOM
NIKA
Vol. 13, No. 1, March 2
015 : 118 – 1
2
7
124
[(
)
]
[
(
)
]
,
[(
)
]
[
(
)
]
()
0,
k
ij
ij
k
k
ij
ij
ij
sa
l
l
o
w
e
d
tt
if
j
a
l
l
owe
d
tt
pt
ot
h
e
rw
i
s
e
(5)
In Formul
a (5),
is the
phe
romo
ne h
euri
s
tic fa
ctor,
is the impa
ct fa
ctor of h
e
u
r
istic
dire
ction fun
c
tion;
allow
e
d
k
is the next pixel allowed to visit a
nd
()
k
ij
pt
is the probability that an
t
k
mo
ve
s
fr
om pixe
l
i
to pixel
j
at moment
t
.
.
(iv) Up
dating
of the phero
m
one con
c
en
tration of every pixel
The phe
rom
o
ne on the pat
h betwee
n
two pixels is not
accumulate
d
infinitely. This pape
r
update
s
the pheromo
ne with MMAS by setting a pheromo
ne scop
e
mi
n
m
ax
[,
]
to pre
v
ent the
algorith
m
fro
m
being
trap
ped in l
o
cal
conve
r
ge
nc
e
becau
se of
the infinite in
cre
a
se of lo
cal
pheromo
ne.
(a)The up
dati
ng mechani
sm of global p
hero
m
on
e
(1
)
(
)
ij
ij
tQ
(6)
Her
e
,
is
the pheromo
nevo
l
atilization co
effici
ent
with
its valu
e ra
nge
within0
~
1.
Q
refers to the value of the initial pherom
one.
ma
x
Q
in order t
o
increa
se the sea
r
ching scop
e at
the begin
n
ing
.
(b)The up
dati
ng mechani
sm of local ph
erom
one
For the
kth
ant, the phe
rom
o
n
e
increme
n
t
ij
on the path it passe
s by is:
(1
)
(
)
(
1
)
ij
ij
ij
tt
t
(7)
mi
n
,(
,
)
(1
)
0,
k
ij
n
t
he
op
ti
mal
p
a
Q
if
i
j
o
T
t
ot
herw
i
s
e
th
(8)
Her
e
,
is the pheromone vol
a
t
ilization coefficient
and
1
is the ph
ero
m
o
ne coefficie
n
t
remai
ned bet
wee
n
iteratio
nst and itera
t
ions
1
t
.
Q
is a consta
nt of th
e remaini
ng
trajecto
ry
numbe
r.
mi
n
k
T
is the
sho
r
te
st ope
ration time
af
ter the a
n
t ha
s go
ne th
rou
gh all the
wal
k
ing
step
s.
In the initial moment
s,
0
ij
.
(v) If it meet
s the
esta
bli
s
he
d en
d
co
ndition
s,
exit the alg
o
rith
m, otherwise, return to
Step (3).
De
cod
e
the i
n
itial pop
ulati
on (namely
e
v
ery
individua
l) ra
ndo
mly g
enerated fro
m
gen
etic
algorith
m
. Th
en
con
d
u
c
t a
n
t col
ony
alg
o
rithm
ev
oluti
on o
n
th
e e
d
ge extractio
n
acco
rdi
ng to
a
corre
s
p
ondin
g
group
of pa
ramete
rs to e
v
ery individua
l. Finally, get the final resul
t
of an iteratio
n,
namely the image ed
ge d
e
tection resul
t.
The re
alizatio
n flowchart of
the algorit
hm
of this paper
is as follo
ws [17].
Evaluation Warning : The document was created with Spire.PDF for Python.
TELKOM
NIKA
ISSN:
1693-6
930
Im
age Edge Feature Extraction and Refi
ning Bas
ed on Geneti
c
-Ant
Colony .... (Xing Zhang)
125
Figure 6. Flowchart of gen
etic-a
nt colo
n
y
hybrid algo
rithm
5.
Experiment
Simulation & Analy
s
is
Simulate this experime
n
tal
result and d
a
ta in the en
vironme
n
t of Wind
ows 7
with Intel
(R) Co
re2
C
P
U
1.33G a
nd
a memory of
2.5G thro
ugh
Matlab R2
01
2a.
For the
co
nvenien
ce of
compa
r
ison, compa
r
e the
geneti
c
-a
nt colony hybri
d
algorith
m
prop
osed in this pa
per
with single g
e
n
e
tic algo
rithm
and singl
e a
n
t colony alg
o
rithm with t
he
same p
a
ram
e
ters. Th
e al
gorithm p
a
ra
meters are
set as follows:
the populati
on si
ze of ge
netic
algorith
m
is
g
nm
, the crossover probability is
0.8
c
p
, the mutati
on probability
is
0.0
5
m
p
,the si
ze
of
ant colony
al
gorithm
is
hn
m
, the ph
ero
m
on
e
heu
ri
stic fa
ctor i
s
3
, impact
factor of heu
ri
stic directio
n functio
n
is
1
an
d the phero
m
one volatilizat
ion coeffici
ent
i
s
0.
1
The exp
e
rim
ent will
co
nd
uct 5
anal
og
simul
a
tion
s to the e
d
g
e
dete
c
tion
of every
algorith
m
an
d ch
oo
se the
optimal effe
cts a
s
the
fin
a
l re
sults. Fi
gure
7 refle
c
t
s
the
simulati
on
results by different alg
o
rith
ms.
It can be se
en from the
above expe
ri
ment
simul
a
tion that the image ed
ge feature
extraction
ba
sed
on g
eneti
c
-a
nt colony
hybrid
algo
rithm ha
s a
hig
her l
e
vel in it
s obj
ective eff
e
cts
and excell
ent
continuity, re
fining and sm
oothne
ss.
It can demo
n
st
ra
te the main edge features
o
f
the image; co
ntains the m
a
in edge i
n
formation of
the
image an
d its edg
e line
s
are q
u
ite refi
ning
and sm
ooth. In one word, it has re
ached
the expecte
d effects.
Evaluation Warning : The document was created with Spire.PDF for Python.
ISSN: 16
93-6
930
TELKOM
NIKA
Vol. 13, No. 1, March 2
015 : 118 – 1
2
7
126
(a) O
r
igin
al image
(b) Gen
e
tic algo
rithm
(c)Ant colony algo
rith
m (d)Gen
etic-a
nt colo
ny algorith
m
Figure 7. Image edg
e featu
r
e extra
c
tion
result
s by different algo
rithm
s
6.
Conclu
sion
As the mo
st
fundame
n
tal
feature, ed
ge
ca
rri
e
s
the
majority of th
e image i
n
formation.
This pa
pe
r a
nalyze
s
the
advantag
es
and di
sadv
an
tages of ge
n
e
tic algo
rithm
and ant col
ony
algorith
m
; integrate
s
the a
d
vant
age
s that genetic alg
o
rithm ha
s fast conve
r
ge
n
c
e speed in t
he
early pha
se a
nd that the ant co
lony algorithm has stro
ng global sea
r
chi
ng ability and appli
e
s it in
image e
dge
feature extra
c
tion a
nd de
tection.As
sh
own i
n
the e
x
perime
n
t si
mulation
result,
geneti
c
-a
nt colony hybrid
algorith
m
ca
n detect
the
image ed
ge
quickly and
accurately with
obviou
s
edg
e
details, ex
ce
llent co
ntinuity and co
mple
te edge exp
r
ession to
rea
c
h the
expe
cted
objec
tive.
Referen
ces
[1]
Anhar R, Pa
la
i
aha
nkote S, C
hee SC, C
h
e
w
LT
. A
Robust
Arbitrar
y T
e
xt Detectio
n S
y
stem for Natura
l
Scene Imag
es.
Expert Systems w
i
th Applicati
ons.
201
4; 41(
18): 802
7-8
048
.
[2
]
Yu
a
n
Y, H
a
o
bo L
,
X
i
ao
qi
a
n
g
L. Se
mi
-Sup
e
r
vise
d
Ch
an
ge
De
te
cti
o
n Me
thod
fo
r Mul
t
i
-
T
e
mp
o
r
al
Hy
pe
r
spectral Images.
Neuro co
mp
uting.
20
15; 14
8(19): 36
3-3
7
5
.
[3]
Oscar C, Evelia L, Jose S,
Patricia M, Fevri
e
r V. Ne
w
ap
p
r
oach
Usin
g A
n
t Col
o
n
y
Opti
mizatio
n
w
i
t
h
Ant Set Partiti
on for
F
u
zz
y
Contro
l D
e
sig
n
Ap
pli
ed t
o
T
he Ball A
nd
Beam S
y
stem.
Informatio
n
Scienc
es
. 201
5; 294(1
0
): 203
-215.
[4]
RM Rizk-Allah,
Elsay
e
d MZ, Ah
med AES.
Hy
bridizing Ant
Colony
Op
timization
w
i
th Fir
e
fly
A
l
gorithm
for Unconstrained Opti
miz
a
tion Problems.
A
ppli
ed M
a
the
m
atics an
d C
o
mputatio
n
. 20
13;
224(
1): 47
3-
483.
[5]
T
i
mur K, Mehmet BY, Mehmet B. An Ant Co
lon
y
Opti
mizatio
n
Algor
i
t
hm for Load
Bala
ncin
g i
n
Parall
el M
a
chi
nes
w
i
t
h
Se
q
uenc
e-De
pe
nd
ent Setu
p T
i
mes.
Co
mp
uters
& Operatio
ns
Rese
arch
.
201
2; 39(6): 12
25-1
235.
Evaluation Warning : The document was created with Spire.PDF for Python.
TELKOM
NIKA
ISSN:
1693-6
930
Im
age Edge Feature Extraction and Refi
ning Bas
ed on Geneti
c
-Ant
Colony .... (Xing Zhang)
127
[6]
JM Prats-Mont
alb
án, A F
e
rr
e
r
. Statistical Pr
oc
ess C
ontro
l
Based
on
Mu
ltivariat
e
Imag
e
Anal
ys
is: A
Ne
w
Pr
opos
al for
Monitor
i
ng and Defect
D
e
tection.
C
o
mp
uters & Ch
e
m
ica
l
Eng
i
ne
eri
n
g
. 201
4;
71(
4)
:
501-
511.
[7]
Gajanan
KB, Vijay
HM.
Blind
Meth
od for
Rescal
i
n
g
D
e
tection
an
d R
e
scale F
a
ct
or
Estimation
in
Digita
l
Images
Using Peri
od
i
c
Properties of
Interpolati
on.
AEU - Internati
ona
l Journ
a
l of
Electronics
and C
o
mmun
ic
ations
. 20
14; 6
8
(7): 644-
65
2.
[8]
Ron
ghu
a S, Li
pin
g
Q, Liche
ng J, Rustam S,
Yang
ya
ng
L. Chan
ge D
e
tection in SAR
Images b
y
Artificial Immu
ne Mu
lti-Obj
e
c
t
ive Cl
usteri
ng.
Engi
ne
erin
g A
pplic
atio
ns of
Artificial Int
e
lli
g
ence.
201
4;
31(5): 53-
67.
[9]
Ian W
,
Nichol
as B, David S. A Performance Ev
alu
a
tion
of statistical T
e
sts for Edge Detectio
n in
T
e
xtured Imag
es.
Computer
Visio
n
and I
m
a
ge Un
dersta
ndi
ng.
201
4; 122(
5): 115-1
30.
[10]
Hazem
MAO. Semi-fragile Wa
termarki
ng f
o
r Gra
y
scal
e
Im
age
Auth
entic
a
t
ion
an
d T
a
mper D
e
tectio
n
Based
o
n
An
Adjuste
d
E
x
pa
nde
d-Bit Mu
ltis
cale
Quantiz
ati
on-Bas
ed T
e
c
hni
que.
J
our
na
l of V
i
sua
l
Co
mmun
icati
o
n and I
m
ag
e R
epres
entati
o
n
. 201
4; 25(5): 10
64-1
081.
[11]
Hans
han L, J
unch
a
i G. Research o
n
T
a
rget
Photoel
ectri
c
it
y
T
r
ack Method a
nd Impr
oved Imag
e
Processi
ng Ar
i
t
hmetic i
n
D
y
n
a
mic T
a
rgets
Detectio
n S
y
st
em.
Optik - Int
e
rnati
ona
l J
our
nal
for L
i
g
h
t
and El
ectron O
p
tics
. 2014; 1
2
5
(14): 35
90-
35
95.
[12]
Pour
ya
H, Ma
hrokh GS. Efficient
Contrast
Enha
ncem
ent
of Images U
s
ing
H
y
bri
d
A
n
t Col
o
n
y
Optimisatio
n
, Genetic Alg
o
rit
h
m, and Sim
u
l
a
ted An
nea
lin
g.
Digita
l
Sig
n
a
l Process
i
ng
.
2013; 2
3
(3):
879-
893.
[13]
Arash G, SMR
Kazemi, F
a
rh
ad M, Moh
a
m
m
ad MN
. A C
o
oper
ative A
n
t Colo
n
y
Optimi
zation-Ge
netic
Algorit
hm Ap
p
r
oach
for C
o
n
s
truction
of E
nerg
y
Dem
a
n
d
F
o
rec
a
stin
g
Kno
w
l
e
dg
e-B
a
sed
E
x
p
e
rt
S
y
stems.
Kno
w
ledge-B
a
se
d Systems
. 20
13
; 39(2): 194-2
0
6
.
[14]
Miros
ł
a
w
T
.
An
t Colo
n
y
Optim
i
zatio
n
Al
gorith
m
Appl
ie
d to S
h
ip St
eeri
ng
C
ontrol.
Pr
oced
i
a
Co
mpute
r
Scienc
e
. 201
4; 35: 83-92.
[15]
Ehsan
M, Ma
hmou
d M, Av
i
C, Sar
a
M,
Gr
aham
C. Efficient T
r
ansit
Sched
ule
D
e
si
gn
of T
i
min
g
Points: A Co
mpariso
n
of A
n
t Col
o
n
y
a
n
d
Genetic Al
go
rithms.
T
r
ansp
o
rtation
Rese
a
r
ch Part B:
Method
olo
g
ic
al
. 2012; 46(
1): 217-2
34.
[16]
Ali RN, Moh
a
mmad HF
, Seye
d T
A
N. A
Fuzz
y
V
end
or
Mana
ged Inv
e
ntor
y
of Multi-Item Economi
c
Order Quantit
y Model u
n
d
e
r Shortag
e
: An An
t Colon
y
Optim
i
zatio
n
Algor
ith
m
.
Internation
a
l
Journ
a
l of
Producti
on Eco
n
o
m
ics
. 20
14; 155(
7): 259-
27
1.
[17]
Sener A, G Mirac B, Adil B. H
y
brid
izin
g An
t
Colon
y
Opti
mizatio
n
via G
enetic Al
gor
ith
m
for Mixe
d-
Mode
l Assemb
l
y
Lin
e
Bal
anci
ng Prob
lem
w
i
th Sequ
ence
Dep
end
ent Set
up T
i
mes bet
w
een T
a
sks.
Appl
ied S
o
ft Computi
n
g
. 20
1
3
;13(1): 57
4-5
89.
Evaluation Warning : The document was created with Spire.PDF for Python.