TELKOM
NIKA Indonesia
n
Journal of
Electrical En
gineering
Vol.12, No.7, July 201
4, pp
. 5438 ~ 54
4
7
DOI: 10.115
9
1
/telkomni
ka.
v
12i7.517
6
5438
Re
cei
v
ed
No
vem
ber 2
0
, 2013; Re
vi
sed
March 13, 20
14; Accepted
April 1, 2014
Image S
e
gmentation Based on
Chaotic Quantum Ant
Colony Algorithm
Li Jiy
i
ng
1
, Dang Jian
w
u
*
2
, Wang Yan
gping
3
, Wan
g
Xiaopeng
4
Schoo
l of Elect
r
onics & Inform
ation En
gi
neer
i
ng,
Lan Z
h
ou J
i
aoto
ng U
n
iver
sit
y
, La
n Z
hou,
Chin
a,
An Nin
g district
w
e
st ro
ad No.
88, Lan Z
h
ou,
Gansu, Chi
na.
T
e
lephon
e No.
:
86+
093
1-4
938
766
*Corres
p
o
n
id
n
g
author, e-ma
i
l
: lj
y76
09@
12
6
.
com
1
, dangj
w
@
mail.lz
jtu.cn
2
A
b
st
r
a
ct
Ant
col
ony al
gorith
m
is
a
new
type of
bio
m
i
m
etic ev
oluti
onary
al
go
rithm, w
h
ic
h h
a
s so
me
outstand
in
g ch
aracteristics
lik
e go
od r
obust
ness, par
al
l
e
l
i
s
m
a
nd
pos
itive
feedb
ack. It has be
en w
i
d
e
l
y
used
in
ma
ny
fields
but
it stil
l h
a
s so
me w
eakn
e
sses, s
u
ch as
slow
-co
n
verg
ence
a
n
d
eas
ily-fal
lin
g i
n
t
o
local
extre
m
e
valu
e. T
her
efo
r
e w
e
pr
op
ose
a
new
a
l
gor
ithm w
h
ich
co
mbin
es th
e q
u
a
n
tum ev
oluti
o
n
a
r
y
alg
o
rith
m
an
d
ant co
lo
ny a
l
g
o
rith
m to
geth
e
r
bas
ed
on
th
ese s
hortco
m
i
ngs, this
n
e
w
type of
al
gorit
h
m
consi
ders the t
w
o quantu
m
bi
t proba
bil
i
ty a
m
p
litu
des as
a
n
ts
’
c
u
rrent l
o
c
a
tion i
n
for
m
ati
on. T
he se
arc
h
in
g
space
w
ill
be
d
oub
led
w
hen
a
n
ts
’
n
u
m
bers
r
e
mai
n
the
sa
me. Both
intro
d
u
c
ing
the
pix
e
l
p
o
int
grad
ie
nt i
n
to
qua
ntu
m
revo
l
v
ing d
oor, a
n
d
dyna
mica
lly
chang
in
g rot
a
tion a
n
g
l
e a
r
e to achi
eve
local-s
earch
i
ng.
Search
ing
by chaotic q
u
a
n
ta
near ca
ndi
dat
es w
i
th optim
a
l
soluti
on is to
improve th
e gl
oba
l opti
m
i
z
a
t
i
on.
Meanw
hi
le, a
d
optin
g a
NAN
D
g
a
te is
to a
c
hiev
e
mutati
n
g
o
perati
on,
a
nd to
avo
i
d
al
gorith
m
’
s
pr
ec
ocity.
Co
mp
ared w
i
t
h
the trad
itio
n
a
l al
gor
ith
m
, the n
e
w
alg
o
rit
h
m
has
a bet
ter pop
ulati
o
n
diversity
and
an
outstanding ability to
overc
o
m
e
the pr
ecoc
ity an
d stagnation
in t
he
opt
imi
z
ation
proc
ess. It has
been
prove
d
that thi
s
improve
d
a
l
g
o
rith
m is effect
ively
to th
e pro
b
le
ms l
i
ke sl
o
w
-converge
nce
and
easi
l
y-fall
i
n
g
into loc
a
l extre
m
e va
lu
e, and
vividly i
n
cre
a
se
t
he spee
d and
accuracy of
the imag
e seg
m
entatio
n.
Ke
y
w
ords
: rail
w
a
y signal
ing,
computer i
n
terl
ockin
g
, all-e
l
ec
tronic, reli
abi
lit
y, availa
bil
i
ty, ma
inta
ina
b
il
ity
Copy
right
©
2014 In
stitu
t
e o
f
Ad
van
ced
En
g
i
n
eerin
g and
Scien
ce. All
rig
h
t
s reser
ve
d
.
1. Introduc
tion
Image
seg
m
entation i
s
a
way of extracting the
regi
o
n
of inte
re
st from th
e ba
ckgrou
nd,
or dividin
g
the image int
o
several different sp
ecie
s
1
according to i
t
s releva
nce. It is a com
p
l
e
x
probl
em in th
e field of ima
ge processin
g
. Many
re
se
arche
r
s
have
thought u
p
va
riou
s theo
reti
cal
approa
che
s
f
r
om diffe
rent
are
a
s
of im
age
se
g
m
ent
ation. The t
r
aditional m
e
thod of im
ag
e
segm
entation
is ba
se
d o
n
num
ero
u
s m
e
thod
s li
ke,
t
h
re
shol
d, e
d
ge d
e
tectio
n, re
gion
g
r
owi
ng,
fuzzy theo
ry, mathematical morp
holo
g
y and so
on. Ho
wever, becau
se o
f
the different
appli
c
ation p
u
rpo
s
e
s
an
d image featu
r
e
s
, t
hese m
e
th
ods
still have some limitatio
ns.
With the
constant devel
opment
of the
intelligent
optimization algorithms,
many
resea
r
chers
adopt th
e int
e
lligent
algo
ri
thms i
n
to im
age
se
gment
ations be
ca
u
s
e th
eir com
m
on
advantag
es
li
ke high co
nverge
nce spe
eds
and
glo
b
a
l optimi
z
atio
ns. Fo
r exa
m
ple, Do
cu
me
nts
[1, 2] which a
r
e rel
e
vant to Ant Colony Algorit
hm, Do
cume
nt [3] related with Ge
netic Algo
rith
m
and
Do
cum
e
nts [4, 5]
rele
vant to Particle Swa
r
m
Op
timization
ha
ve pro
p
o
s
ed
different ima
g
e
segm
entation
method
s. Be
cau
s
e th
ese i
n
telligent
alg
o
rithm
s
al
so
have some
weakne
sses li
ke
long-se
archin
g, slow-
conv
erge
nce, the
y
are likely t
o
be premat
ure
whe
n
the
image data
are
large e
nou
gh.
The qua
ntum
computin
g was first propo
sed by
the Ameri
c
an phy
sicist, R.P.Feynman in
1982.
Re
cen
t
ly, some re
sea
r
che
r
s
h
a
ve com
b
ine
d
the qu
ant
um computin
g with p
a
ttern
recognitio
n
, and
come
u
p
with the
re
cog
n
ition al
g
o
rithm b
a
sed
on qu
antum
theory, such
as
quantum
tem
p
late m
a
tchi
ng al
gorith
m
, qua
ntum im
age
re
cog
n
ition al
gorith
m
and
so o
n
. In
Do
cume
nt [6], quantum a
n
t colony alg
o
rithm is u
s
e
d
to solve Q
O
S unicast ro
uting, however,
Do
cume
nt [7] quantum
ad
opts a
n
t col
o
ny algorith
m
to
diag
no
se fa
ults in n
aphth
a
cracke
r. Th
is
pape
r i
s
i
n
spi
r
ed
by
combi
nation
of ant
colo
ny al
g
o
rit
h
m a
nd
qua
n
t
um theo
ry, its ai
m i
s
to
so
lve
the probl
em o
f
data cluste
ri
ng, afte
r whi
c
h the image
will be segme
n
ted.
Evaluation Warning : The document was created with Spire.PDF for Python.
TELKOM
NIKA
ISSN:
2302-4
046
Im
age segm
entation Base
d
on Cha
o
tic Q
uantum
Ant Colon
y
Algorith
m
(Li Jiying
)
5439
2. Quantum
Ant
Colony
Algorithm
The qua
ntu
m
ant colon
y
algorithm (QACA)
was
first pro
p
o
s
e
d
by A.Nara
yanan &
M.Moore
"in
his
pap
er “Qu
antum-ant Al
gorithm
”.
Th
e
algo
rithm
wa
s u
s
e
d
fo
r
sol
v
ing p
r
oble
m
of
combi
nato
r
ial
optimization
in orde
r to a
c
hieve
go
od
result
s [9]. Th
e qu
antum
bi
t introdu
ce
d
by
each ant in
dicate
s the p
o
sition of ant; the ant’s
fo
rward moving ta
rg
et is sele
cted
by pheromo
n
e
stren
g
th and
the prob
abilit
y of selection
of the vi
sible stru
cture; the algo
rithm
allows qu
ant
um
gate to upd
ate quantu
m
bi
t, then determines th
e
an
ts’ movement
. The phe
ro
mone
stren
g
th is
update
d
by the moved p
o
sition. The
algorith
m
co
nsid
ers the two proba
bility amplitude of
quantum
bit a
s
the
ant
s’
cu
rre
nt lo
cation.
Wh
en
th
e a
n
t
s’ num
be
rs a
r
e the
same,
the searchi
n
g
-
spa
c
e
will be
double
d
; the
mutation op
eration i
s
a
c
hieved by qu
antum gate.
Comp
ared wi
th
traditional
alg
o
rithm, thi
s
a
l
gorithm
ha
s
a better po
pu
lation dive
rsit
y, and effe
ctively overcom
e
the pre
c
o
c
ity and the sta
g
n
a
tion of the ant co
lony alg
o
rithm in opti
m
ization p
r
o
c
ess.
2.1. Quantify
ing the An
t
Colon
y
Space
Assu
me that
each pixel
)
,
(
j
i
in the image i
s
a point, on
e
3D
spa
c
e
wi
ll be built aft
e
r
con
s
id
erin
g t
he g
r
ay
scale, gradient,
an
d
3
3
realm
s
info
rmation.
Ra
n
domly di
strib
u
te
m
ant
s
in this sp
ace. So a quantu
m
bit state is conv
eyed a
s
the followi
n
g
according t
o
the quantu
m
theorie
s.
1
|
0
|
|
β
α
(1)
The
,
respe
c
tively rep
r
esent
prob
ability a
m
plitude of
0
,
1
, and
1
|
|
|
|
2
2
β
α
, th
e
ants’ current l
o
catio
n
s a
r
e i
ndicated by p
r
oba
bility amplitude ju
st as sho
w
n in Fig
u
re 1.
Figure 1. Qua
n
tum Individu
al Tran
sferrin
g
Diag
ram
The n bits qu
anta are e
n
d
o
we
d to the No.
i
Ant.
in
in
i
i
i
i
i
p
sin
cos
...
...
sin
cos
sin
cos
2
2
1
1
(2)
So ea
ch
ant
occupi
es two
po
sition
s in
spa
c
e, P
o
siti
on
)]
cos(
)...
cos(
)
[cos(
2
1
in
i
i
and Positio
n
)]
sin(
)...
sin(
)
[sin(
2
1
in
i
i
.
Comp
ared
wi
th the ordi
na
ry ant colo
ny al
gorith
m
wh
en the p
opul
ation num
be
rs are the
same,
QACA
has
dou
ble
d
the qu
ant
um pop
ulat
io
n se
archin
g
spa
c
e
and v
a
stly increa
sed
conve
r
ge
nce spe
ed.
2.2. Transfer
ring the Qu
a
n
tum An
t’ Position
The tran
sferring-state pro
bability is ca
lc
ulate
d
by the phe
romo
ne intensity and the
heuri
s
tic info
rmation o
n
ant’s moving
proces
s in
Ant Colony Algorithm (ACA), and t
h
e
transfe
rring
p
a
th is dete
r
mi
ned
also. In
QACA, the fo
llowing
Fo
rm
ula i
s
a
dopte
d
to id
entify a
n
t’s
positio
n from
point i to poin
t
j.
Evaluation Warning : The document was created with Spire.PDF for Python.
ISSN: 23
02-4
046
TELKOM
NI
KA
Vol. 12, No. 7, July 201
4: 5438 – 54
47
5440
)]
,
(
[
)]
,
(
)]
,
(
[
)]
,
(
[
0
)
,
(
s
i
s
i
s
i
s
i
s
i
p
k
(
3
)
)
,
(
max
arg
s
i
p
j
k
(4)
Its
Heuris
tic
func
tion is
:
2
1
1(
)
m
ij
k
i
k
j
k
k
qx
x
(5)
q
is the
wei
ght
ing
coefficie
n
t
about the
i
m
age’
s G
r
ay
scale, G
r
adi
e
n
t and th
e
realm
information,
is the phero
m
o
ne den
sity.
In quantum
space, the ant’
s
movem
ent can
be a
c
hie
v
ed by its own quantu
m
bit
s
, while
pha
se chan
gi
ng is u
s
ually
determi
ned b
y
quantum re
volving door,
as defin
ed fol
l
owin
g:
cos
sin
sin
cos
R
(6)
Suppo
se
that
r
p
is
the current pos
i
tion,
d
p
is
the target pos
i
tion,
d
p
attempts
to get
clo
s
e t
o
r
p
by th
e force of
R
,
and to reali
z
e position u
pdati
ng. Ge
n
e
rally, the constructo
r
function
,
S
, the
polling li
st of
and the
qu
a
n
tum revolvin
g doo
r’
s an
gl
e
all dete
r
min
e
ant’s o
p
timization spee
d, there
are
ma
ny
document
s ab
out this i
s
sue [10, 1
1
]
.
,
S
is the
revolving
ang
le’s di
re
ction
that en
sures algo
ri
thm
co
nverge
nce. F
o
r exam
ple,
Figure 2 i
s
t
he
schemati
c
di
agra
m
of quantum revolv
i
ng doo
r, assume that there are
x
individual ants, the
optimal num
ber
is
best
x
. Wh
en
the Fitn
ess
value
)
(
)
(
best
x
f
x
f
, it is
suppo
se
d to i
n
crea
se
probability of current soluti
on 0, namely
to enlarge
2
, if
)
(
exists in the
first and the
third
quad
rant,
should cl
ockwise revolve, however, if
)
(
exists in
the
seco
nd a
nd t
he forth
quad
rant,
sho
u
ld anti
c
lo
ckwise revolve, so
different
quantu
m
co
nversi
on
s a
r
e de
sign
ed
by
different situa
t
ion.
Figure 2. Sch
e
matic Di
ag
ram of Quantu
m
Gate
2.3. Mutage
n
i
c Factor
s
To preve
n
t phero
m
on
e fro
m
early-fallin
g into the local extreme va
lue on the pa
th found
at the initial stage, the m
u
tageni
c fact
ors
achieved
by quantum
NOT g
a
te are introd
uced
into
QACA. First, assum
e
that
the mutation
pro
bab
ility is a con
s
tant, then give
a random
num
b
e
r
Others
s
j
Evaluation Warning : The document was created with Spire.PDF for Python.
TELKOM
NIKA
ISSN:
2302-4
046
Im
age segm
entation Base
d
on Cha
o
tic Q
uantum
Ant Colon
y
Algorith
m
(Li Jiying
)
5441
rang
ed fro
m
0 to 1 to each ant, if this numbe
r
is le
ss than the
co
nstant a
s
sum
ed befo
r
e, then
cho
o
s
e
2
/
n
quant
um bits in th
e ant individ
u
a
ls, so excha
nge the
qua
n
t
um NOT
gat
e with two
probability amplitudes, its prin
ciples are show
n as follows:
i
i
i
i
cos
sin
sin
cos
0
1
1
0
(7)
Actually, the mutating o
peratio
n is t
o
excha
nge
the two qua
ntum bits’ probability
amplitude
s,
make th
e ori
g
inal liabl
e-collap
s
ed
stat
ed "1" coll
ap
se at state "
0
", and vice
versa.
Another muta
tion is the
rot
a
tion of
th
e q
uantum
bit an
gle, the
No.
i
Quantum
bit rotating
a
ngle
is
i
, so the angle will become
i
)
2
(
after mutation, this
rotation does
not c
o
mpare
with the
curre
n
t best i
ndividual afte
r the forward
-rotatin
g
,
it has nothi
ng to
do with its o
p
timal locatio
n
memory a
nd
the locatio
n
, so the po
pul
ation di
versity has bee
n increa
sed, an
d avoided fall
ing
into local extreme values.
2.4. Analy
z
in
g Mechanis
m of Pheromone Upd
a
tin
g
The phe
rom
o
ne inclu
d
e
s
a
n
ts’ cu
rrent optimal
location information level, while gradi
ent
informatio
n i
s
co
ntaine
d in
heuri
s
tic info
rmation in
QA
CA. When
ev
ery ant
compl
e
tes
a
sea
r
c
h
,
ant’s current positio
n will b
e
mappe
d fro
m
the uni
t sp
ace to the op
timal solution
spa
c
e. Wh
e
n
quantum o
p
timization al
go
rithm will han
dle the relati
onship bet
we
en qua
ntum bit and soluti
on
vector, a ra
n
dom ha
s be
en given first
l
y, and
then compa
r
e
with a prob
abili
ty amplitude of
quantum
bit
s
to d
e
termin
e
the
re
sults o
f
the q
uant
u
m
bit tran
sforming, finally
a bin
a
ry
sol
u
tion
has
bee
n go
t, then co
nvert it into a
decim
al solut
i
on vecto
r
.Th
e
ope
ration
i
s
with
a
cert
ain
degree
of ra
ndomn
e
ss, a
nd p
r
on
e to
self-d
eg
rad
a
tion. Given
thi
s
, the
qua
ntu
m
bit p
r
ob
abi
lity
amplitude
ca
n be dire
ctly applie
d as a
solutio
n
vect
or en
codi
ng to avoid the random
ne
ss o
f
the
transfo
rmin
g. Integrating
the fitness functio
n
valu
es
into phe
romone
and
gradi
ent into
the
heuri
s
tic info
rmation make
s the optimu
m
posit
ion of
the pherom
one be
com
e
highe
r; arou
ses
more info
rma
t
ion, and sp
e
eds u
p
t
he se
arch process
and converge
nce.
It is ne
cessa
r
y to update t
he ph
ero
m
on
e on e
a
ch no
de after
ants
finish o
ne
cycle at
t
,
whi
c
h is
sho
w
n a
s
the followin
g
:
n
k
k
ij
ij
ij
t
t
1
1
(8)
In above formula (8
),
is phero
m
on
e attenuatio
n co
e
fficient,
k
ij
is the phe
romo
ne
left on node
s of the No.
k
ant finishes o
n
e
cycle.
3. Chao
tic Q
u
antum
Ant
Colon
y
Algorithm
3.1. The Ov
e
r
v
i
e
w
of
Cha
o
s
Cha
o
s i
s
a
comm
on n
online
a
r p
h
e
nomen
on, it is symb
olized with
ran
domne
ss,
ergo
dicity an
d re
gula
r
ity, it is very
sen
s
itive to
the i
n
itial co
nditio
n
s, it
re
peate
d
ly traverse
s all
states withi
n
a certain
ra
n
ge a
c
cording
to its o
w
n l
a
w. The
s
e
pro
pertie
s
of
ch
a
o
s
ca
n be
m
ade
to optimized
-sea
rch so
ch
aos
can b
e
in
tegrated
with the other alg
o
rithms.
Cha
o
tic va
ria
b
le i
s
u
s
u
a
lly gen
erate
d
b
y
the Lo
gisti
c
. The
ch
ao
s
seq
uen
ce
formula i
s
following:
))
(
1
)(
(
)
1
(
t
x
t
x
t
x
(9)
)
1
(
t
x
is
cha
o
tic v
a
riabl
e,
......
2
,
1
,
0
t
,
is
a c
h
aotic
attrac
tor.
When
=
4
, the
system fall
s into the cha
o
tic state, and g
ener
ates
cha
o
tic variabl
es
rangi
ng from
0 to 1.
Evaluation Warning : The document was created with Spire.PDF for Python.
ISSN: 23
02-4
046
TELKOM
NI
KA
Vol. 12, No. 7, July 201
4: 5438 – 54
47
5442
3.2. The Cha
o
tic Quan
tu
m Ants Sear
ching
Whe
n
the qu
antum ant co
lony algorith
m
begi
n
s
to initialize, the pheromo
ne o
n
each
path is eq
ual.
Therefo
r
e, in
dividual direction is onl
y d
e
termin
ed by the heuri
s
tic i
n
formatio
n, it
is
difficult to quickly find a good sol
u
tion
in a lar
ge nu
mber of ca
n
d
idate
s
, so the conve
r
g
e
n
ce
spe
ed i
s
sl
owed do
wn
at the initial
stag
e. Usi
ng the i
n
itial phe
rom
one di
strib
u
tion form
ed in
the
cha
o
s optimi
z
ation spee
d
s
up t
he co
nverge
nce. In this pape
r,
the improve
d
Tent mapp
ing
mentione
d in Do
cume
nt [14] is adopte
d
to do cha
o
s o
p
timization.
1
5
.
0
))),
1
,
0
(
1
.
0
(
1
(
2
5
.
0
0
),
1
,
0
(
1
.
0
(
2
1
t
t
t
t
t
x
rand
x
x
rand
x
x
(
1
0
)
Her
e
,
1
0
k
t
, the f
i
rst
path
of
quantum
div
e
rsity
pop
ula
t
ion is initiali
zed
by
cha
o
tic varia
b
les, an
d the quantum bit i
s
followi
ng:
),
2
sin(
),
2
cos(
t
i
t
i
x
x
n
i
1
(11)
1
n
path
s
will b
e
gen
erate
d
a
c
cordi
ng th
e
method
abov
e, find the
o
p
t
imal path
s
from
them, and initialize the p
h
e
r
omo
ne on th
ese o
p
timal p
a
ths.
Con
s
id
erin
g t
he g
r
adi
ent i
s
the imp
o
rtan
t feat
ure
of th
e imag
e e
dge
s, this pap
er
adju
s
ts
the step of
by using the g
r
adie
n
t functi
on, it is
most likely on the i
m
age ed
ge when the pixel
gradi
ent valu
e o
c
curs,
we
must m
a
ke
small an
gle
to
prevent
the
e
x
treme p
o
ints. If the value
of
the gradient
of the pixel i
s
smalle
r, a
large
r
a
ngul
a
r
will
cro
ss
the area
within the im
ag
e,
d
d
r
r
β
α
β
α
、
、
,
prob
ability a
m
plitude a
n
d
the target
quantum
b
i
t for the current bit
r
e
spec
tively. W
h
en
r
d
d
r
β
α
β
α
A
,
0
A
, the directio
n of
is
)
sgn(
-
A
,while,
0
A
,the directio
n of
can be plu
s
or min
u
s. S
o
dynamic a
d
j
ustment of d
e
sig
n
is sho
w
n belo
w
:
min
max
min
0
)
(
)
sgn(
f
f
f
x
f
A
(
1
2
)
)
(
x
f
is
an i
ndividu
al fitness fu
n
c
tion,
)
(
x
f
is a
g
r
a
d
ient valu
e at
the p
o
int of
x
. The
locatio
n
of th
e initial
optim
al solution
wil
l
be fo
und
by
, sea
r
che
s
th
e optim
al
sol
u
tion in
the
current near l
o
cation, If a better sol
u
tion
is found,
it will replace the original optim
al solution. T
he
positive fe
ed
back
of a
n
t col
ony
algo
rithm p
h
e
r
o
m
one
is u
s
e
d
to
avoid
cha
o
s sea
r
ching
blindn
ess, a
n
d
imp
r
ove
se
arch
efficien
cy. Distu
r
ba
nce 10
ch
aotic se
que
nce i
s
use
d
in
this
pape
r, thus,
different indiv
i
dual
s are su
bjecte
d to
dif
f
erent di
sturb
ance by
fatn
esse
s. a si
ng
le
and fixed
-
si
ze qu
antum
rotation g
a
te
has be
en avoided, ants popul
ation
i
s
preve
n
ted
f
r
om
falling into local extreme v
a
lue
s
. The seque
nce pert
u
rbatio
n ampl
itude is
t
i
e
0
,
0
is the
controlling-parameter,
t
is the iteration
s
.
)
1
(
t
x
i
i
(
1
3
)
3.3. Algorith
m
Flo
w
Char
t
Step 1
:
Initializin
g ch
ao
s, chaoti
c
varia
b
les a
r
e g
e
n
e
rated
ch
aotic varia
b
les
a
c
cordi
n
g
to the Formu
l
a (11), an
d initializing p
a
ths ar
e gene
rated acco
rdi
ng to the Formula (10). T
he
initial populati
on si
ze is
m
、
iteration maxim
u
m numb
e
r o
f
is
IterMax
、
attenuation co
efficient
of
pheromo
ne is
、
the pherom
one upd
ate index is
、
heu
ri
stic informatio
n of updating the index
is
、
mutation probability is
m
p
,A maximum phero
m
on
e is
max
.
Step 2:
cal
c
ulating the fitness of ea
ch
individual
qu
antum value
according to
(3), (4),
and calculatin
g the heuri
s
ti
c inform
ation
according to (5).
Evaluation Warning : The document was created with Spire.PDF for Python.
TELKOM
NIKA
ISSN:
2302-4
046
Im
age segm
entation Base
d
on Cha
o
tic Q
uantum
Ant Colon
y
Algorith
m
(Li Jiying
)
5443
Step 3:
Repl
acin
g
with
cu
rre
nt po
sition,
wh
en
ant
th
e
location
is superi
o
r to th
e
optimal
locatio
n
of its memo
ry, repla
c
ing with
the cu
rrent global optim
al location,
whe
n
the cu
rre
nt
global o
p
tima
l location i
s
b
e
tter than tha
t
of
the global optimal positi
on before the
search.
Step 4:
Upd
a
t
ing ant locati
on with qu
ant
um revolving
door.
Step 5:
Asce
nding the indi
viduals a
c
cording their fitness value
,
m
u
tation-o
p
e
r
a
t
ing the
after
2
/
m
individu
als’ mutation
prob
ability mutation by the quantum gat
e.
Step 6:
Cha
o
s qua
ntum sea
r
ching n
e
a
r the cu
rre
nt optimal solut
i
on acco
rding
to (13),
if optimal solu
tion exists, re
placi
ng ste
p
3 to get the optimal solutio
n
.
Step 7:
Upd
a
t
ing the phero
m
one on the
node
s a
c
cord
ing to (8).
Step 8:
Retu
rning to Step
2 cycle
-
calcu
l
ating,
until the conve
r
g
e
n
ce
conditio
n
or th
e
maximum limi
t
of number o
f
iteration is g
o
t.
Step 9:
Tag
g
i
ng the no
de
)
,
(
j
i
as the g
oal,
if its phero
m
one
s meet
max
ij
afte
r
rea
c
hin
g
o
r
a sp
ecifie
d
numbe
r of it
eration
s
of
cycling time
s. Otherwise taggin
g
it as the
backg
rou
nd, and completi
ng the image
segm
entation
.
4. Analy
z
ing
the Algo
rith
m’s Conv
ergence
Theorem 1:
The p
opul
ation sequ
en
ce {
0
t
,
X
t
} of cha
o
tic qu
antum
ant col
ony
algorith
m
(CQACA) i
s
a limited
and ho
mogen
eou
s
Markov ch
ain
.
In CQACA, quantum bits
have been a
dopted,
altho
ugh the sp
ace populatio
n spa
c
e i
s
theoreti
c
ally i
n
finite, yet in
this pa
per th
e po
pulatio
n
size i
s
a p
r
e
determi
ned
value, the
spa
t
ial
dimen
s
ion
is 3, so the
p
opulatio
n is l
i
mited.
Ho
we
ver, qua
ntum
mutation,
chaotic qu
ant
um
sea
r
ching
an
d ph
ero
m
one
upd
ating all
have n
o
thin
g to d
o
with
the num
be
r
of iteration
s
,
so
1
t
X
is only relate
d
to
},
,
f
)
(
{max(
*
*
S
A
A
f
A
, not ab
o
u
t the n
u
mbe
r
of ite
r
ation
s
.
So as me
ntioned, the
pop
ulation
seq
u
e
n
ce
{
0
t
,
X
t
} of
cha
o
tic q
uantum
ant colony al
gorith
m
(CQACA) i
s
a
limited and h
o
moge
neo
us
Markov ch
ain
.
Theorem 2
:
CQACA
conv
erges to the globally opt
imal solution
with probability 1.
Suppo
se that
S
is the states spa
c
e,
f
is the global
optimal sol
u
tion, and orde
r
},
S
A
,
)
A
(
f
{max(
A
*
*
f
if and only if
1
}
S
)
0
(
A
A
)
t
(
A
{
p
0
*
t
lim
, s
o
the
algorith
m
is converg
ent.
Assu
me the i
ndividual
state transitio
n probability is:
)
X
X
(
p
)
t
(
p
i
t
j
1
t
ij
Becau
s
e the
algorith
m
use
s
the optimal
ret
ention strategy, there
are two type
s of the
transition pr
obabilities.
(1) Whe
n
*
*
A
j
,
A
i
,
for any
,
0
t
there is
)
X
(
f
)
X
(
f
t
1
t
,
so
)
t
(
p
ij
0
;
(2) Whe
n
*
*
A
j
,
A
i
,
}
S
)
0
(
A
|
i
)
t
(
A
{
p
)
t
(
p
p
0
i
t
)
t
(
p
)
t
(
P
)
t
(
p
)
t
(
P
p
ij
A
iA
j
i
ij
A
iA
j
i
1
t
**
**
Bec
a
us
e,
1
)
t
(
p
)
t
(
p
*
*
A
j
ij
A
j
ij
Evaluation Warning : The document was created with Spire.PDF for Python.
ISSN: 23
02-4
046
TELKOM
NI
KA
Vol. 12, No. 7, July 201
4: 5438 – 54
47
5444
We get,
))
t
(
p
)
t
(
p
)(
t
(
p
)
t
(
p
*
*
*
A
j
ij
A
j
ij
A
i
i
=
)
t
(
p
)
t
(
p
)
t
(
p
)
t
(
p
ij
A
iA
j
i
ij
A
iA
j
i
**
**
Form the firs
t c
a
se,
,
)
t
(
p
)
t
(
p
ij
A
iA
j
i
**
=0
And,
)
t
(
p
)
t
(
p
)
t
(
p
ij
A
iA
j
i
**
1
t
p
)
t
(
p
)
t
(
p
)
t
(
p
ij
A
iA
j
i
**
So,
t
1
t
p
p
0
,
Bec
a
us
e,
t
t
A
i
i
t
*
t
t
p
lim
1
)
t
(
P
lim
1
}
f
f
{
lim
.
Then,
1
}
f
f
{
lim
*
t
t
.
End.
Conclu
sion
: the CQA
C
A converges to the global
opti
m
al solution with probabilit
y 1.
5. Simulatio
n
and Results
In order to verify the feasibility of CQACA
algorithm f
o
r image segm
entation, thi
s
papers
has
segm
ent
ed LENA im
age wh
ose size is 25
6*
25
6 respe
c
tively via Sobel
operator, Ca
nny
operator, Ant Colony Alg
o
rithm an
d Cha
o
tic Qu
a
n
tum Ant Co
lony Algorith
m
of the in
the
Matlab7.0
si
mulation envi
r
onm
ent and
on the platform
that is equ
ipped
with the Intel (R) Core
(TM)
Du
o CPU, domin
an
t freque
ncy
1.83G
Hz,
9
8
7
MHz, 0.99
GB memo
ry. The results are
sho
w
n
in
Fi
gure
4. Initi
a
lizin
g p
a
ra
meters a
r
e:
Ant col
ony
size
m
is
40
,
th
e evap
oratio
n
coeffici
ent
is 0.7,
is 0.5,
is 0.5, the maximum numbe
r o
f
iterations
IterMax
is 100,
m
p
is
0.02, the phe
romo
ne
s max
max
is 0.9, chaoti
c
distu
r
ba
nce
factor
0
is 1.6. Quantum an
ts are
rand
omly dist
ributed
at the
initializing
st
age, t
he ph
e
r
omo
ne at th
e optimal
sol
u
tion begi
ns
to
increase duri
ng sear
ching process, until achieve the ma
ximum at last, so the ants will be
con
c
e
n
trated
at these
nod
e
s
ju
st
as sho
w
s in
Figu
re
3,
wh
en
the
i
m
age ed
ge is being ch
ecked,
the image
ed
ge point i
s
th
e optimal
sol
u
tion, so
the
pheromo
ne d
ensity is m
u
ch highe
r, an
d
the
pheromo
ne d
ensity matrix
will be obtained, then
the
image edge
point can b
e
extracted, the
quantum a
n
ts po
sition tra
n
sferrin
g
dia
g
ram i
s
sh
ow
n as Fig
u
re
4. Figure 5 i
s
the picture from
Evaluation Warning : The document was created with Spire.PDF for Python.
TELKOM
NIKA
ISSN:
2302-4
046
Im
age segm
entation Base
d
on Cha
o
tic Q
uantum
Ant Colon
y
Algorith
m
(Li Jiying
)
5445
edge
segme
n
t
ation, Figu
re
(a
) i
s
the
result of
bei
ng e
dge-ch
ecke
d by
Soble ope
rator, Figu
re (b)
Figure (a
) is t
he re
sult of b
e
ing ed
ge
-ch
e
cked
by
Ca
nny ope
rator,
Figure (c) i
s
the re
sult of
ant
algorith
m
, and Figure (d) i
s
the result of chaotic
qu
a
n
tum ant algorithm. It is proved that Soble
operator
ha
s
lost some l
o
w level of gray in so
me
d
e
tails, the Canny
ope
rator i
s
b
e
tter, but it h
a
s
some
too mu
ch d
e
tails. Fi
gure
(c) i
s
th
e re
sult of A
C
A an
d Figu
re (d
) is CQACA’s
re
sult. It is
sho
w
e
d
that gray part
s
has b
een
ch
ecked a
nd
h
a
ve a relative accuracy.
Becau
s
e of t
he
diversity of th
e CQA
C
A’s
searchin
g spa
c
e, a lot
of d
e
tails h
a
ve b
een
remai
n
e
d
. What’
s
mo
re,
with comp
ari
ng the ite
r
atio
n times
of two algo
rith
m
s
,
the iteratio
n times
of ACA i
s
46, it
s runni
ng
time is 98.6s,
while CQA
C
A’s iteration times is 1
8
, its runni
ng time is 31.6. Obvio
u
sly, CQACA
is
much b
e
tter than ACA.
(a)
(b)
Figure 3. The Di
stribut
ion Dia
g
ra
m of Chaoti
c
Qu
antum Ants
(a)
(b)
(c
)
(d)
Figure 4. The
Compa
r
i
s
on
of Lena Edge
Extraction
(a) T
he MRI
brain im
age
s
(b) ACA
(c) CQA
C
A
Figure 5. The
Compa
r
i
s
on
of Two Kind
s of Brain MRI Image Segm
e
n
tation
Figure 5 is th
e MRI brai
n image
s, Figure (b) a
nd Fig
u
re (c) i
s
the
result of co
mpari
ng
the ant
colo
n
y
algorithm
segm
entatio
n and quantu
m
ant colony al
gorithm. Pa
ra
meter
sel
e
cti
on:
Evaluation Warning : The document was created with Spire.PDF for Python.
ISSN: 23
02-4
046
TELKOM
NI
KA
Vol. 12, No. 7, July 201
4: 5438 – 54
47
5446
Colo
ny si
ze
is 40, vol
a
tile coeffici
e
n
t is
0.7, p
hero
m
on
e u
pdate in
dex
is 0.85,
he
uristi
c
informatio
n u
pdating the i
ndex is 0.8
5
, the maxi
mu
m numbe
r of
iteration
s
of
is 60, mutat
i
on
prob
ability is 0.04, pherom
one maximu
m value is
0.7, the disturb
ance factor i
s
1.6.
Differen
c
e
s
i
n
the segme
n
tation of whi
t
e matte
r an
d
gray matte
r i
n
the brain i
s
not big,
but in the ce
rebro
s
pi
nal flu
i
d of segm
ent
ation,
the sp
a
t
ial diversity of quantum
search al
gorith
m
and th
e
seg
m
entation
re
sults a
r
e
retai
ned
more d
e
tailed i
n
form
ation. Compa
r
i
s
on
of n
u
mb
er
of
iteration
s
and
the converge
nce time of
form 2 to two k
i
nds
of algorit
h
ms
.
Table 1. the
Comp
ari
s
o
n
of Three Algo
rithms’ Iteration Time
s and
Runni
ng Tim
e
s
Iterations times
Running time
ACA 38
78.3
s
CQACA
22
36.8
s
Figure 6 is the comp
arin
g result
s of hum
an ce
ll
s se
gm
ented by two kind
s of algo
rithms,
form segm
e
n
tation effect
, chaoti
c
qu
antum a
n
t colony alg
o
rit
h
m segme
n
tation lo
w g
r
ay
informatio
n which
ha
s m
o
re information,
the tar
get i
s
compl
e
ted, a
nd t
he co
nvergen
ce spe
ed is
much lo
we
r than the ant colony algo
rith
m.
Figure 6. The
Compa
r
i
s
on
of the Two Kind of Cell
s Image Segm
ent
ation
6. Conclusio
n
Image segme
n
tation is a
difficult probl
e
m
in image p
r
ocessin
g
. At pre
s
ent, the
r
e is no
a
comm
on met
hod for different image se
gmentation.
Combi
n
ing th
e theory of chao
s optimization
and refe
rri
ng
the quantu
m
evolutiona
ry algorithm
and ant colo
ny algorithm,
this pape
r u
s
e
s
quantum
ant
sea
r
ch
sp
ace
diversity and th
e a
d
vantage
of conve
r
g
e
n
c
e sp
eed. Af
ter
segm
enting
three
ki
nd
s of
image
s, it i
s
sh
owed th
at this
method
is
effective t
o
solve the
a
n
t
colo
ny algorit
hm and e
a
sy
to fall into local minim
a
a
nd slo
w
conv
erge
nce sp
e
ed problem,
and
on the se
gme
n
tation sp
eed
and pre
c
i
s
io
n has b
een i
m
prove
d
gre
a
tly by experiments.
Referen
ces
[1]
Yang
LC, Z
h
a
o
LN, W
u
XQ. Medic
a
l ima
g
e
segme
n
tatio
n
of fuzz
y
C-m
e
ans cl
usterin
g
base
d
on t
h
e
ant col
o
n
y
alg
o
rithm. Jour
nal
of SHANDON
G
Univ
ersit
y
(
E
ngi
neer
in
g S
c
ienc
e) (in
Chi
nese). 2
0
0
7
;
6(3): 51-5
4
.
[2]
Bai Y. App
licat
ion
of ant col
o
n
y
al
gorithm
in
the segm
enta
t
ion of MRI.
C
h
in
ese Jo
urn
a
l
of Medic
a
l
Imag
in
g T
e
chn
o
lo
gy (in Ch
in
e
s
e)
. 2007; 2
3
(9
): 1402-1
4
0
4
.
[3]
Yang K, Ji
ang
HW
. Researc
h
of improv
ed
gen
et
ic al
gorit
h
m
for image s
e
gmentati
on b
a
s
ed o
n
fuzz
y
C-means cl
ust
e
rin
g
.
Co
mput
er Engi
ne
erin
g and Ap
plic
atio
ns (in Ch
ines
e)
.
2009; 45(
33): 179-
183.
[4]
Z
hang
LB. F
u
z
z
y
C-M
e
a
n
Cl
u
s
tering B
a
se
d
on Partic
le S
w
arm Optimizati
on.
Jour
na
l
of Jilin Un
iversity
(Scienc
e Editio
n) (in Chi
nes
e)
. 2006; 44(
2): 217-2
21.
[5]
Che
n
Z
Y
. F
u
zzy Cl
usterin
g
Al
gorithm Bas
e
d
on Particle S
w
a
rm.
Comput
er Engin
eer
ing
(in Chin
ese).
200
7; 33(2): 19
8-19
9.
[6]
Zuo JL, Yu G
L
. QoS Multicast Routing
w
i
th Re
strai
n
Ba
sed o
n
Qua
n
tum Ant Co
lo
n
y
Al
gor
ithm.
Co
mp
uter Engi
neer
ing (i
n Chi
nese).
20
12; 3
8
(2): 172-
17
4.
(
a) The cells
image
(b) ACA
(c) CQACA
Evaluation Warning : The document was created with Spire.PDF for Python.
TELKOM
NIKA
ISSN:
2302-4
046
Im
age segm
entation Base
d
on Cha
o
tic Q
uantum
Ant Colon
y
Algorith
m
(Li Jiying
)
5447
[7]
W
ang L, W
a
ng
XT
, Yu JS. Naphth
a
crack
i
n
g
fur
nac
e fau
l
t
dia
gnos
is b
a
se
d on
ad
aptiv
e
qua
ntum a
n
t
co
l
ony
a
l
go
ri
thm.
CIESC Journal.
20
09; 60(
2
)
: 401-40
7.
[8]
Color
n
i A, D
o
ri
go M, Man
i
ezz
o
V, et al, Ant
s
y
stem for
jo
b sho
p
sch
ed
u
ling.
B
e
lg
ia
n o
f
operati
o
n
s
Rese
arch stati
s
tics and Co
mputer Scie
nce
.
199
4; 34(1): 39
-53.
[9]
A Nara
ya
na
n, M Moore.
Qua
n
tum-
ant Al
gor
ithms.
Proc
eedings of IEEE International Conference o
n
Evoluti
onar
y
C
o
mputati
on.
IEEE press. 199
6; 61-66.
[10]
Han
neke D, H
o
me JP, Jost
JD,et al. Reali
z
at
ion of a pro
g
ramma
ble t
w
o-qu
bit qua
ntu
m
processor
.
Nature Phys
ics
. 2010; 6: 13-1
6
.
[11]
Li BB, W
ang L
.
A h
y
bri
d
qu
a
n
tum-ins
pire
d gen
etic
al
gorit
hm for multi-ob
jective flo
w
s
h
op sche
d
u
lin
g.
IEEE Transactions on System
s, m
an and cybemetics
. 200
7; 37(3): 57
6-5
9
1
.
[12]
W
ang L, W
u
QD.
Ant System algorit
hm for
opti
m
i
z
at
ion i
n
contin
uo
us spa
c
e.
Proc of the 2001 T
EEE
Internatio
na
l C
onfere
n
ce o
n
Contro
l Appl
ic
a
t
ions, Me
x
i
co: IEEE Press. 2001; 1: 395-
400.
[13]
Ja
yaram
an VK
, Kulkar
ni B
D
, Sachi
n
K, et
al
. Ant col
o
n
y
fra
m
e
w
ork for
o
p
timal
desi
g
n
sn
d sch
edu
li
n
g
of batch pl
ants.
Computer a
n
d
Che
m
ic
al Eng
i
neer
ing.
2
000;
24(8): 19
01-
19
12.
Evaluation Warning : The document was created with Spire.PDF for Python.