TELKOM
NIKA Indonesia
n
Journal of
Electrical En
gineering
Vol.12, No.4, April 201
4, pp. 2776 ~ 2
7
8
3
DOI: http://dx.doi.org/10.11591/telkomni
ka.v12i4.5000
2776
Re
cei
v
ed Se
ptem
ber 10, 2013; Revi
se
d No
vem
ber
1, 2013; Acce
pted No
vem
b
er 16, 201
3
Detection Algorithm of Airport Runwa
y
in Remote
Sensing Images
ZhuZhon
g Yang*
1
, JiLiu Zhou
2
, Fang
Nian Lang
3
1
Chen
gdu U
n
iv
ersit
y
, Co
ll
ege
of Electronics
and Informati
o
n
Engi
ne
erin
g, Che
ngd
u, Chi
n
a
1
Shenzh
en Ke
y L
abor
ator
y
of
High Perform
a
nce Data Mi
nin
g
, Shenzh
en,
Chin
a
2
Colle
ge of Co
mputer Scie
nc
e (Soft
w
a
r
e)
, Sichu
an Un
ivers
i
t
y
, Ch
eng
du,
Chin
a
1,2,
3
Laborat
or
y
of Pattern Rec
ogn
ition
and In
tellig
ent
Inform
ation Proc
essi
ng, Sichu
an, C
h
in
a
*Corres
p
o
ndi
n
g
author, e-ma
i
l
:
y
zz8
919
88
5
@
16
3.com
1
, zhou
jl@sc
u.ed
u
.
cn
2
,
fnlang@
1
63.com
3
A
b
st
r
a
ct
T
h
is article d
e
scribes a ra
pid
detection
met
hod of
the a
i
rp
ort runw
ay for remote sens
in
g imag
es.
T
he specific
al
gorith
m
i
m
ple
m
e
n
tatio
n
step
s are: F
i
rs
tly, we use the Otsu'
s meth
od to se
parate th
e run
w
ay
from the i
m
a
g
e
s. T
hen w
e
use the fractio
nal d
i
ffe
renti
a
l
gradi
ent op
er
ator, w
h
ich ha
s good a
n
ti-n
o
i
se
perfor
m
a
n
ce a
nd effective e
dge extracti
on
ability,
to extract the edge
informatio
n
o
f
imag
e, and
us
e
auto
m
atic thr
e
shol
d to distin
guis
h
the e
d
g
e
of runw
ay a
nd oth
e
r
obj
ec
ts. F
i
nally, w
e
use the Ho
u
g
h
alg
o
rith
m to c
a
lculat
e the r
u
n
w
ay. T
h
is dete
c
tion
meth
od
o
f
the air
port ru
nw
ay, w
h
ich gr
eatly re
duc
es th
e
data op
eratio
n
an
d greatly
e
nha
nces
the c
o
mputi
ng s
p
e
e
d
, has
a
d
vant
a
ge
of g
ood
tes
t
results
an
d f
a
st
spee
d. T
h
is
pa
per s
how
s that
, as the
st
udy
of fraction
al
dif
f
erentia
l, the
a
pplic
atio
n
of fraction
al c
a
lc
ul
us i
n
a w
i
der area w
i
ll be succ
essful
.
Ke
y
w
ords
: ed
ge detecti
on, OT
SU, fractional
differentia
l op
erator, Hou
gh t
r
ansfor
m
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
Targ
et dete
c
t
i
on in
re
mote
se
nsi
ng im
a
ge, the
airp
ort is a
typical
l
i
near target.
Airport
s
are im
portant
military targets,
in order
to paralyze
airport, t
he runway needs
to be damaged
contin
uou
s. T
here
are som
e
literatu
r
e
s
a
bout remote
-sen
sin
g
imag
e dete
c
tion a
nd ide
n
tificati
on
of airpo
r
t run
w
ay in d
o
me
stic and fo
rei
g
n [1-3]. In
mai
n
features
of the
runway, the mo
st obvio
us
feature i
s
a
straight lin
e, so
t
he run
w
ay t
a
rget d
e
tectio
n pro
b
lem tu
rns into h
o
w t
o
dete
c
t strai
ght
lines i
n
the i
m
age. G
ene
rally Hou
gh transfo
rm
wa
s use
d
to d
e
tect ai
rpo
r
t ru
nway. Th
e m
a
in
advantag
e of
the
Hou
gh
transfo
rm
is
not sen
s
itive to noi
se,
b
e
tter a
b
le to
han
dle
parti
al
occlu
s
ion in t
he image a
n
d
coveri
ng ot
her issu
es.
Howeve
r, beca
u
se it is a type of exhau
stive
sea
r
ch, so its computatio
n
a
l compl
e
xity and spa
c
e
complexity is very high [4, 5], which can not
meet the req
u
irem
ents of
real-t
ime sy
stem
s. Over the years a
l
a
rge n
u
mbe
r
of resea
r
ch
ers
con
d
u
c
ted a lot of study [6, 7], proposed variou
s al
gorithm
s to significa
ntly incre
a
se Ho
ug
h
operation spe
ed. Ho
wever,
these alg
o
ri
t
h
ms a
r
e mo
st
ly application
-
spe
c
ific.
There are th
ree m
a
in issues
of run
w
ay det
ec
tion: Firs
tly, different remote
s
e
ns
ing
image
s with different
b
r
ig
htness and
contra
st, ru
nway is
ea
sy to
be confu
s
e
d
with
oth
e
r
targets,
whi
c
h ma
ke
s more difficul
t
to sepa
rate
of t
he run
w
ay from the
backg
rou
nd.
Secon
d
ly, no
is
e
intensity of the airp
ort ru
nway is relat
i
vely
large in the remote sen
s
in
g imag
es, maki
ng the
accurate extraction
of the
edge
s
of obj
e
c
ts i
n
the
ima
ge b
e
come
s
more
difficult.
Finally,
Hou
gh
transfo
rm
cal
c
ulatio
n, de
ci
ded by it
s ch
ara
c
teri
stic
s,
is too
slo
w
. F
o
r la
rge
-
scale
remote
sen
s
i
n
g
image
s is th
e comp
uting
pro
c
ess ta
kes a lon
g
time
, it is difficult to meet
real-time
re
mote
sen
s
in
g ima
g
e
targ
et dete
c
tion
req
u
ire
m
ents. T
o
so
lve these p
r
o
b
lems, thi
s
p
aper p
r
e
s
ent
s a
n
improve
d
airp
ort run
w
ay de
tection alg
o
rit
h
m. Firs
tly, we use Ot
su'
s
method (OTS
U) to sepa
rat
e
airpo
r
t run
w
a
y
area
from
t
he ima
g
e
s
. T
hen
we
u
s
e t
he fra
c
tion
al
differential
gradient
ope
rat
o
r to
extract imag
e
edge inform
ation, and use automatic
t
h
re
shol
d to extract edge,
whi
c
h sig
n
ificantly
redu
ce
s the
need
to
deal
with im
age
in
formation.
Fi
nally, we
u
s
e
Ho
ugh
calcu
l
ation to
extract
the run
w
ay.
Evaluation Warning : The document was created with Spire.PDF for Python.
TELKOM
NIKA
ISSN:
2302-4
046
Dete
ction Alg
o
rithm
of Airport Run
w
a
y
in
Rem
o
te Sensing Im
ages
(ZhuZh
ong Ya
ng)
2777
2. Algorithm and Implementa
tion
2.1. OSTU T
h
reshold Se
gmenta
tion
Algorithm
Airport
ru
nway dete
c
tion
in the
remot
e
sen
s
ing i
m
age,
we first
deal
with th
e imag
e
segm
entation
.
Typically, different rem
o
te
sen
s
ing
ima
ges h
a
ve different b
r
ightn
e
ss, contrast
and
resolution, a
nd al
so a
ccompani
ed b
y
high noi
se, whi
c
h m
a
ke
s it difficult to autom
atic
segm
entation
of the airpo
r
t run
w
ay. Th
e
s
e ofte
n
ma
ke the a
r
ea
co
ntaining th
e runway ca
n n
o
t
be
su
ccessfu
lly extracted,
or
to
gethe
r
with oth
e
r
ob
jectives. Alth
ough
there a
r
e m
any ima
g
e
segm
entation
method
s [8],
but they
are
po
werl
es
s
fo
r
r
u
nw
a
y
r
e
c
o
gn
itio
n
in re
mo
te
s
e
ns
ing
image
s. Fo
r exampl
e, iterative
threshol
d
segmentatio
n, mathemati
c
al m
o
rp
hol
ogy
segm
entation
.
Therefore, we prop
ose algor
ithm
s for remote
sen
s
ing im
age usi
ng O
T
SU
threshold
se
gmentation a
l
gorithm, whi
c
h ma
ke
s
the brightn
e
ss and cont
ra
st of the runway
relative to the backgroun
d has
b
een
sig
n
ificantly enh
anced.
OTSU
algo
rithm was p
r
op
ose
d
by
Ja
pa
nese
schola
r
s Ot
su i
n
1
9
7
9
[9]. We
u
s
e
OTSU
algorith
m
to extract the inte
restin
g re
gion
, whic
h
contai
ns the runwa
y
. The method is define
d
a
s
:
s
(
x
,
y
)
i
f
s(
x,
y
)
(,
)
0
O
t
her
w
i
s
e
T
fx
y
(1)
Whe
r
e
s(x, y
)
is the o
r
igi
nal ima
ge, f(x, y)
is the
segmente
d
im
age, T i
s
the
thre
shol
d which
divides
sou
r
ce image pixe
ls into two
categori
e
s, a
n
d
makes the
betwee
n
-cla
ss va
rian
ce t
o
maximum. It is determine
d by:
()
m
a
x
{
(
)
}
,
,
0
,
1
,
2...255
|
BB
tk
tk
k
Tt
(2)
Whe
r
e
()
B
k
is the betwe
en-cla
s
s varian
ce, gi
ven by:
22
00
1
1
(
)
()
(
(
)
(
)
)
()
(
(
)
(
)
)
Br
r
kk
k
k
k
k
k
(3)
00
1
1
1
()
(
)
(
)
()
()
(
)
L
r
i
ki
P
i
k
k
k
k
(4)
0
1
0
1
1
1
()
()
()
()
k
i
L
ik
iP
i
k
iP
i
k
(5)
0
1
1
1
()
(
)
()
(
)
k
i
L
ik
kP
i
kP
i
(6)
()
()
ni
pi
N
(7)
Whe
r
e
L i
s
t
he n
u
mbe
r
o
f
gray level
s
;
n(i
))
i
s
th
e
sum
of i-th
p
i
xel gray l
e
vel; N i
s
th
e t
o
tal
numbe
r of pix
e
ls the im
age
;
p(i)
i
s
the p
r
obability of gray level i;
ω
0
and
ω
1
a
r
e th
e pro
bability
of
the first cla
s
s (for exam
p
l
e: region of
interest
ROI) and the se
con
d
categ
o
ry (for examp
l
e:
b
a
c
k
g
r
o
un
d)
;
μ
0
,
μ
1
,
μ
r
are the mea
n
s
of first, se
co
nd
and im
age
resp
ectively;
δ
r
i
s
th
e i
m
ag
e
varian
ce. OT
SU advanta
g
e
is that it is an optim
al al
gorithm to di
stingui
sh b
e
twee
n two types,
whi
c
h can ea
sily obtain th
e regio
n
s of i
n
tere
sti
ng (ROI) from the i
m
age. But it
doe
s not appl
y to
many differe
n
t targets, be
cau
s
e in th
e final analysi
s
it is only a single threshol
d seg
m
entati
on
method.
Evaluation Warning : The document was created with Spire.PDF for Python.
ISSN: 23
02-4
046
TELKOM
NI
KA
Vol. 12, No. 4, April 2014: 2776 – 2
783
2778
2.2. Using Fr
action
a
l Differential Gr
a
d
ient Ope
r
ator to Extr
act Edge
Edges
of the
object a
r
e t
he impo
rtant
featur
e
s
of target d
e
tecti
on and i
denti
f
ication,
whi
c
h i
s
an
i
m
porta
nt clu
e
to visual
pe
rceptio
n. The
edge
of the t
a
rget i
s
g
r
ay-scale
cha
nge
s in
the image, g
r
ay-scale ch
ange
s have
many form
s,
and the mo
st ba
sic fo
rm
is the ide
a
li
zed
model. The id
eal edge i
s
a set of conn
ected pixe
ls, wh
ich is ge
nerated on ideali
z
ed model, ea
ch
pixel is on a
gray-scale vertical jum
p
e
d
step.
In practice, the multiplicative n
o
ise blu
r
s the
contrast
of ad
jace
nt are
a
s in rem
o
te sen
s
ing
im
age
s,
that maki
ng
chang
es i
n
adj
ace
n
t areas to
flatten, there
appe
ars a
tra
n
sition
zone
at the e
dge
a
r
ea
s, the
r
e n
o
long
er is a
singl
e pixel
e
dge
of ideal co
nd
ition, it is difficult to determi
ne the exa
c
t locatio
n
of the edge. Remote se
nsi
n
g
image ba
ckg
r
ound i
s
more uniform a
nd contain
s
high
e
r
noise, the cl
assical gra
d
ie
nt-ba
s
ed e
d
g
e
detectio
n
alg
o
rithm is ve
ry sen
s
itive to noise,
so that
classi
cal ed
ge dete
c
tion
method
s are not
suitabl
e for the rem
o
te sensi
ng imag
e
s
. To solve t
h
is proble
m
, we u
s
e fra
c
ti
onal differe
ntial
gradi
ent op
erator to extra
c
t edge of im
a
ge. Fra
c
ti
on
al
differential fil
t
er can b
e
de
duced fro
m
the
integer-o
rde
r
differentiation
filter [10, 11]. Fraction
al di
fferential finite impulse (FI
R
) filter tra
n
sf
er
function a
s
fo
llows:
1
1
()
v
v
z
Dz
T
(8)
Referen
c
ing
the bin
o
mial
seri
es expa
n
s
ion
2
(1
)
(
1
)
(1
)
1
!
vk
k
vv
v
k
x
vx
x
k
,
and u
s
ing
1
z
instead of
x
, the
Equation (8)
can b
e
writte
n as:
11
20
1(
1
)
(
1
)
1
(
1
)
()
1
(
)
(
1
)
!(
1
)
(
1
)
vi
i
i
vv
ii
vv
v
i
v
Dz
v
z
z
z
ii
v
i
TT
(9)
Whe
r
e: T
is
sampli
ng
peri
od, z i
s
th
e d
i
spla
cem
ent
operator and
()
is th
e G
a
m
m
a fun
c
tion.
Acco
rdi
ng to the limited impact of fra
c
tio
nal differ
entia
l (FIR) filter transfe
r fun
c
tio
n
, sele
cting the
suitabl
e
N
, obtained a
pproximate first-ord
e
r ba
ck
wa
rd f
i
nite differen
c
e formula.
1
0
11
(
1
)
()
(1
)
(1
)
(
1
)
v
N
vi
i
v
i
zv
Dz
z
Ti
v
i
T
(10)
It can get a si
gnal differe
n
tial equatio
n:
()
(
1
)
(
1
)
()
(
)
(
1
)
(
2
)
(
1
)
(
)
2!
(
1
)
v
n
v
df
t
v
v
v
ft
v
f
t
f
t
f
t
n
nv
n
dt
(11)
For di
gital i
m
age
s, ba
se
d on the
si
gnal diffe
r
en
ce e
quatio
n, fractio
nal di
fferential
gradi
ent form
ula ca
n be ob
tained in diffe
r
ent directio
n
s
.
Hori
zo
ntal direction:
11
(
1
,
)
(
1
,)
(
,
)
(
,)
vv
v
XL
XR
XL
XR
n
n
D
D
D
a
Ix
y
a
Ix
y
a
Ix
n
y
a
I
x
n
y
(12)
Vertical di
re
ction:
11
(,
1
)
(,
1
)
(,
)
(
,
)
vv
v
YU
Y
D
YU
YD
n
n
D
D
D
a
Ix
y
a
Ix
y
a
I
x
y
n
a
I
x
y
n
(13
)
135° di
re
ction
:
11
(1
,
1
)
(
1
,
1
)
(
,
)
(
,
)
vv
v
LU
R
D
LU
R
D
n
n
D
D
D
a
I
x
y
a
Ix
y
a
Ix
n
y
n
a
I
x
n
y
n
(14
)
Evaluation Warning : The document was created with Spire.PDF for Python.
TELKOM
NIKA
ISSN:
2302-4
046
Dete
ction Alg
o
rithm
of Airport Run
w
a
y
in
Rem
o
te Sensing Im
ages
(ZhuZh
ong Ya
ng)
2779
45° directio
n:
11
(1
,
1
)
(
1
,
1
)
(
,
)
(
,
)
vv
v
RU
L
D
RU
L
D
n
n
D
D
D
a
Ix
y
a
Ix
y
a
Ix
n
y
n
a
Ix
n
y
n
(15
)
Whe
r
e
1
av
,
2
(1
)
2
vv
a
,
3
(1
)
(
2
)
6
vv
v
a
,
4
(1
)
(
2
)
(
3
)
24
vv
v
v
a
,…,
(1
)
(1
)
!(
1
)
n
n
v
a
nv
n
Selecting th
e previou
s
item
s n, the four d
i
rectio
ns
of the gra
d
ient of fractio
nal diffe
rential
mask can be
achieve
d
by the truncate
d
. To not
make filter erro
rs too
large, we choo
se t
he
previou
s
thre
e items of fraction
al ord
e
r
differen
c
e
definition to con
s
tru
c
t the
following 5
×
5
different directions fra
c
tion
al differential
gradi
ent ma
sk.
Selecting th
e previou
s
item
s n, the four d
i
rectio
ns
of the gra
d
ient of fractio
nal diffe
rential
mask can be
achieve
d
by the truncate
d
. To not
make filter erro
rs too
large, we choo
se t
he
previou
s
thre
e items of fraction
al ord
e
r
differen
c
e
definition to con
s
tru
c
t the
following 5
×
5
different directions fra
c
tion
al differential
gradi
ent ma
sk.
0 0
0
0
0
0 0
0
0
0
(v
2
-v
)/2
-v
0
v
(v
-v
2
)/2
0 0
0
0
0
0 0
0
0
0
(a) Hori
zo
ntal
directio
n
0 0
(v
2
-v
)/2
0
0
0 0
-v
0
0
0
0 0 0
0
0 0
v
0
0
0 0
(v-v
2
)/2
0
0
(b) Ve
rtical di
rectio
n
(v
2
-v)/2
0
0
0
0
0 -v
0
0
0
0 0
0
0
0
0 0
0
v
0
0 0
0
0
(v-v
2
)/2
(c) 135
° dire
ction
0 0
0
0
(v
2
-v
)/2
0 0
0
-v
0
0 0
0
0
0
0 v
0
0
0
(v
-v
2
)/2
0
0
0
0
(d) 4
5
° dir
e
cti
o
n
Figure 1. 5×5
Different Di
rection
s
Fracti
onal Different
ial Gradi
ent Mask
Referrin
g the
first derivati
v
e Sobel gra
d
ient
ope
rato
r and hi
ghlig
hting the rol
e
of the
central row
a
nd column
m
a
sk by weigh
t
ed, we
can
con
s
tru
c
t the
Tiansi f
r
a
c
tional different
ial
gradi
ent ma
sk as follo
w:
(v
2
-v
)/2
-v
0
v
(v
-v
2
)/2
(v
2
-v
)
-2v
0
2v
(v
-v
2
)
3(v
2
-v
)/2
-3v
0
3v
3(v
-
v
2
)/2
(v
2
-v
)
-2v
0
2v
(v
-v
2
)
(v
2
-v
)/2
-v
0
v
(v
-v
2
)/2
(a) X directio
n fraction
al di
fferential grad
ient
mask
(v
2
-v
)/2
(v
2
-v
)
3(v
2
-v
)/2
(v
2
-v
)
(v
2
-v
)/2
-v
-2v
-3v
-2v
-v
0 0
0
0 0
v
2v 3v 2v
v
(v
-v
2
)/2
(v
-v
2
) 3
(
v-
v
2
)/2
(v
-v
2
)
(v
-v
2
)/2
(b) Y directio
n fraction
al di
fferential grad
ient
mask
Figure 2. Tiansi Fractio
nal
Differential G
r
adie
n
t Mask
Fra
c
tional dif
f
erential op
erator re
alization
and a d
e
tailed de
script
ion of the an
ti-noise
perfo
rman
ce,
see the articl
e [12]. Remote sen
s
i
ng im
age co
ntain
s
the airpo
r
t ru
nway, of course,
whi
c
h al
so
in
clud
es
so
me
other goal
s,
su
ch
as
buil
d
ing
s
, streets, noise
an
d
so on. Plu
s
, the
length, width
and structu
r
e
of airport run
w
ay are
different in remote
sen
s
ing ima
ges. If the linear
matchin
g
, Ho
ugh t
r
an
sform is directly
applie
d to
th
e
image,
ea
ch
pixel ne
ed
s t
o
be
dete
c
ted
the
dire
ction
s
an
d angl
es. A
s
a re
sult, the
required ti
me i
s
too l
ong, a
n
d
so
me n
on-li
near obje
c
tives
have bee
n detected. The
r
efore, we
u
s
e
automatic th
reshold to extract ima
ge e
dge informati
on.
After the aut
omatic th
re
shold op
eratio
n, t
he outpu
t is a bina
ry
image, whi
c
h
can
red
u
c
e
Evaluation Warning : The document was created with Spire.PDF for Python.
ISSN: 23
02-4
046
TELKOM
NI
KA
Vol. 12, No. 4, April 2014: 2776 – 2
783
2780
comp
utation time and in
crease accu
ra
cy of detecti
on of the airp
ort run
w
ay. Comp
ari
ng to
the
traditional
me
thods, th
e di
rection
s
a
nd
angle
s
of th
e
ea
ch pixel
can be
re
du
ce
d effectively i
n
binary image.
Computing time will also be great
ly reduced. Meanwhile,
those non-st
r
aight edg
e
and noi
se
will not be false
detecte
d.
Automatic thresh
old is d
e
fined a
s
:
1i
f
g
(
x
,
y
)
b(
x
,
y
)
0
O
the
r
w
i
s
e
(16)
Whe
r
e
b
(
x,
y)
is the output
image, whi
c
h is after a
u
tomatic thresh
old ope
ration
. Note that, the
definition of
automatic thresh
old
seg
m
entation i
s
u
nder the
assumption th
at
the bri
ghtne
ss of
the target i
s
greate
r
tha
n
the ba
ckgro
und b
r
i
ghtn
e
ss. If on the
oppo
site
ca
se, the o
perator
Symbol
sh
ould chan
ge
to
.T
h
r
es
ho
ld
β
is autom
atically cal
c
ul
ated from th
e sou
r
ce ima
g
e
f(x
,
y)
and lo
w-pa
ss filter
m(x
,
y)
.
2.3.
Using Hough Calc
ulation to Ex
tr
act the Airpo
r
t Run
w
ay
Hou
gh tra
n
sf
orm
wa
s pro
posed by Pa
ul
Ho
ugh in
1962 [1
3]. Hough tran
sform wa
s
origin
ally defined a
s
:
ba
x
y
(17)
Whe
r
e x an
d y is 2-dim
ensi
onal
spa
c
e
coo
r
din
a
tes of a
poin
t
set. For e
x
ample, in two
-
dimen
s
ion
a
l image
s, a a
nd b are re
spe
c
tively
the slop
e and
intercept of a straig
ht line
para
m
eters. Thus, for e
a
c
h poi
nt (x,y), this
set of
param
eters {(a,b)} ca
n be obtain
ed
by
Equation (17
)
. If there is a grou
p of coordi
nate
s
{(x,y)} in the same line, the
y
must have the
same p
a
ram
e
ters
(a,b).
Hou
gh tran
sf
orm is a
c
tual
ly created a
n
accumulato
r of these poi
nts
.
Searchin
g local maxima of the accumul
a
tor, If
the ma
ximum value is app
roximat
e
ly equal to the
numbe
r of
po
ints, whi
c
h
in
dicate
s the
s
e
points in
the
same
line, ot
herwise not i
n
the
same
li
ne.
Mappin
g
at th
e local maxim
u
m of the p
a
rameters
(a
,b)
is
the set
of points of
a straight
line slo
p
e
and inte
rcept. Ho
wever, d
u
e
to features
of slop
e an
d i
n
tercept, pa
rameters
(a,b) may be infini
te.
This cre
a
tes accumul
a
tor become
s
imp
o
ssible
. The
r
efore, Ho
ugh
transfo
rm i
s
d
e
fined a
s
:
co
s
s
in
xy
(18)
Whe
r
e
i
s
the straight di
stan
ce which
is fo
rm the
origin p
o
int to the line, an
d
is the a
ngle
whi
c
h is f
r
om
the origin
p
o
int to perpe
ndicular li
ne
of the line. In this definitio
n,
and
are
limited,
is from 0 to the image dia
gon
al length;
from 0 ° to 360 °
In Houg
h tra
n
sform, there
are three va
lues
n
eed to
accumul
a
te, resp
ectively, , x, y,
(or
). O
b
viou
sly, the cal
c
ul
ation is u
s
u
a
ll
y very compli
cated a
nd
slo
w
. To solve this p
r
obl
em, we
use the
Hou
g
h
transf
o
rm
which b
a
sed o
n
dire
ctional
mask. Define
d as:
mm
1
i
f
and
O(
x
,
y
)
O(
x
,
y
)
A(
,
)
xc
o
s
(
)
ys
i
n
(
)
0O
t
h
e
r
w
i
s
e
flag
(19)
Whe
r
e A
(
,
)is t
h
e a
c
cum
u
lat
o
r
of
an
d
, O
m
(x,
y) is th
e di
re
ctional ma
sk.
is th
e a
ngl
e of
an acce
ptabl
e offset,
1 indicates pl
us
one.
Be note
d
that
in th
e tra
d
itional
Ho
ugh
tran
sform,
the
angl
e
ran
ge i
s
0
-360
. H
o
ug
h
transfo
rm
of
an ima
ge, th
e o
r
igin
point
is
set to th
e u
pper left poi
n
t
or l
o
wer l
e
ft point, the
an
gle
rang
e is -9
0
-+180
. But the directio
nal mask an
gle ran
ge is -90
- +
9
0
, therefore, the
dire
ctional
m
a
sk an
gle n
e
eds to
be
co
nverted to
a
Hou
gh tra
n
sf
orm a
ngle. T
here
directio
nal
mask a
ngle
is
O
m
(x, y
)
,, after
co
nve
r
ted
Hou
gh
transfo
rm
an
gle rang
e i
s
O
m
’
(
x,
y)
, t
h
e
conve
r
si
on fo
rmula i
s
as fo
llows:
Evaluation Warning : The document was created with Spire.PDF for Python.
TELKOM
NIKA
ISSN:
2302-4
046
Dete
ction Alg
o
rithm
of Airport Run
w
a
y
in
Rem
o
te Sensing Im
ages
(ZhuZh
ong Ya
ng)
2781
mm
'
m
m
O
(
x
,
y
)
180
i
f
O
(
x
,
y
)
0
and
(
x
,
y
)
0
O(
x
,
y
)
O
(
x
,
y
)
O
t
her
wi
s
e
(20)
mm
xt
a
n
(
O
(
x
,
y
)
9
0
)
y
i
f
O
(
x
,
y
)
0
(x
,y
)
sk
i
p
O
t
h
e
r
w
i
se
(21)
Acco
rdi
ngly, Equation (19) is repla
c
e
d
b
y
:
''
mm
1
i
f
and
O(
x
,
y
)
O(
x
,
y
)
a
n
d
A(
,
)
xc
o
s
(
)
ys
i
n
(
)
0
O
ther
w
i
se
flag
(22)
Once the Ho
ugh tran
sfor
m was
compl
e
t
ed, you ca
n sea
r
ch for accumul
a
tor
A(
,
)
to
find a lo
cal m
a
ximum. Sea
r
ch
can u
s
e
a
n
y sea
r
ch
me
thod. Tra
d
itio
nal s
ear
ch
al
gorithm
s
kno
w
n
as exha
ustive
sear
ch, i
s
de
fined as:
(,
)
m
a
x
{
(
,
)
}
mi
i
AA
(23)
Whe
r
e
i=
-K,
-K+
1
,…
,
,
+1,…
,
+K
;
i=
-K,
-K+
1
,…
,
,
+1,
…
,
+K
.
(2K
+1)
× (
2
K +1
)
i
s
the nu
cle
a
r
size of
se
ar
ch. K valu
e
wa
s d
e
te
rmi
ned
by the
maximum di
stributio
n. In
ou
r
experience, 1<
K<
3 is appropriate.
The li
ne
w
ill
be det
ected which corres
ponds t
o
the maximum
value of para
m
eter
s
an
d
.
3. Experimental Re
sults
and An
aly
s
is
In orde
r to test the feasibil
ity of
the pr
o
posed metho
d
and re
sult
s, experiment
s are as
follows: Rem
o
te sen
s
ing i
m
age
s used for the test
sh
own in Figu
re
3. T
he figure
shows that the
run
w
ay ha
s d
i
fferent brig
htness,
co
ntra
st, and width in comp
ari
s
o
n
with other r
e
gion
s. Figur
e
4
sho
w
s the re
sults
of typical thre
shol
d OTSU
s
egm
entation alg
o
r
ithm. By fractional diffe
re
ntial
gradi
ent
ope
r
a
tor to
extra
c
t the e
dge
im
age
wa
s
s
h
o
w
n i
n
Fi
gur
e
5. Next,
we
u
s
e th
e
autom
atic
linear th
resh
old algo
rithm
to detect edge. Then,
usi
ng Equation
(22
)
of the Hough tra
n
sfo
r
m
method to co
nstru
c
t a str
a
i
ght line, reco
nstr
u
c
ted
stra
ight line wa
s
sho
w
n in Fig
u
re 7.
Figur
e 7 sh
ows, the ru
n
w
ays
we
re su
ccessfully
detecte
d.
Fr
om the exp
e
rime
ntal
results, th
e
method
pr
op
ose
d
in
this pap
er
bette
r tha
n
the
tradition
al m
e
thod of i
m
a
ge
segm
entation
,
which re
du
ce the amo
unt
comp
ut
ation
of Hou
gh tra
n
sfor
m an
d g
r
eatly impr
ove
the oper
ation
spe
ed.
Figur
e 3. The
Original Ima
g
e
Figur
e 4.
The
Results of Pr
oce
s
sed by O
T
SU
Algorithm
Evaluation Warning : The document was created with Spire.PDF for Python.
ISSN: 23
02-4
046
TELKOM
NI
KA
Vol. 12, No. 4, April 2014: 2776 – 2
783
2782
Figur
e 5. The
Results of O
perate
d
by
Fra
c
tional Differential
G
r
ad
ient
Figur
e 6. The
Results of Detected by
Automatic
Threshold
Figur
e 7. The
Result of Ho
ugh Tr
an
sfor
m
4. Conclusio
n
HO
NG alg
o
rit
h
m is very su
itable for the
detec
tio
n
of straight line
s
i
n
the image,
and ha
s
good a
n
ti-noi
se pe
rfor
man
c
e, but it
s calculatio
n amo
unt is hug
e.
If only pro
c
e
s
sing the edg
es of
image, can g
r
eatly red
u
ce
the
comp
uta
t
ion time, but the tradi
tion
al edge
dete
c
tion alg
o
rith
ms
are
sen
s
itive
to noise. This article
wa
s b
a
se
d
on fr
act
i
onal differ
ent
ial edg
e dete
c
tion o
per
ato
r
whi
c
h can im
prove its n
o
ise immunity.The pr
opo
s
ed
algorith
m
is
not only inse
nsitive to noi
se,
but al
so
gr
ea
tly redu
ce
s t
he
com
putati
on time.
M
e
a
n
whil
e thi
s
al
gorithm
can
meet the
lar
g
e-
scale an
d rea
l
-time rem
o
te sen
s
in
g imag
e pro
c
e
ssi
ng.
Ackn
o
w
l
e
dg
ements
This re
se
ar
ch is sup
porte
d by Shenzh
en Key Labo
ratory for Hig
h
Performa
n
c
e Data
Mining with S
hen
zhe
n
Ne
w Industry Developme
n
t Fun
d
unde
r gr
ant
No.CXB20
1
0
0525
0021A.
Referen
ces
[1]
Huerl
a
s A, Co
le W
,
Nevatia
R. Detectin
g
run
w
a
y
s
in c
o
mpl
e
x air
port
scenes.
C
o
mputer Vis
i
o
n
G
r
aphics a
nd Ima
ge Proc
essi
ng
. 199
0; 51(2)
:107-1
45.
[2]
Liu D, H
e
L, C
a
rin L.
Air
port
detectio
n
in
lar
ge a
e
ria
l
optic
al i
m
a
gery
. Ac
oustics, Spe
e
c
h
, and Si
gn
al
Processi
ng, Proc. Int. Conf. 2
004; 5: 76
1-76
4.
[3]
Bao F
u
m
i
n, L
i
Aig
uo, Q
i
n
Z
hen
g. Autom
a
tic Rec
o
g
n
itio
n
of Airf i
e
l
d
R
u
n
w
a
y
i
n
S
y
nt
hetic A
pertur
e
Rad
a
r Images.
Journa
l of Xia
n
Jiaoto
ng U
n
i
v
ersity
. 2004; 3
8
(2): 124
3-1
2
4
6
.
[4]
W
ang C
h
e
ng,
W
ang R
un-s
h
e
n
. Lin
e
E
x
tracti
on for SA
R Image.
Acta E
l
ec
tronica S
i
nic
a
.
200
3;
31(
6):
816-
820.
[5]
Z
hang Y, Z
h
a
ng C. Segm
en
ting ch
aracters
of
licens
e pl
a
t
e b
y
ho
ugh tr
ansformati
on a
nd the
prio
r
kno
w
l
e
d
ge.
Ch
ines
e Journ
a
l o
f
Comp
uters
. 2
004; 27(
1): 130
-135.
Evaluation Warning : The document was created with Spire.PDF for Python.
TELKOM
NIKA
ISSN:
2302-4
046
Dete
ction Alg
o
rithm
of Airport Run
w
a
y
in
Rem
o
te Sensing Im
ages
(ZhuZh
ong Ya
ng)
2783
[6]
L Xu, E Oja, P Kultan
an. A ne
w
c
u
rve d
e
te
cti
on meth
od: Ra
ndom
ized
Hou
gh transform (
RHT
).
Pattern
Reco
gniti
on L
e
tters
. 1990; 11(
5): 331–
33
8.
[7]
Sun F
,
Liu J.
F
a
st houg
h tra
n
sform al
gorith
m
.
Chines
e Jo
urna
l of Co
mp
uters
. 200
1; 2
4
(10): 1
102-
110
9.
[8]
R Caves, S Queg
an, R W
h
it
e.
Quantitative
comparis
on o
f
the performa
nce of SAR s
egme
n
tatio
n
algorithms.
IEEE Trans. Im
age Processing
. 199
8; 7(11): 15
34–
15
46.
[9]
Jin
L, T
i
an L,
W
ang
R, Guo
L, Ch
u J.
An
i
m
pr
ove
d
Otsu
imag
e se
g
m
e
n
t
ation
alg
o
rith
m for
p
a
th
mar
k
detectio
n
un
de
r variabl
e il
lu
mi
natio
n
. Proc. Intellig
ent Ve
hicl
es S
y
mp
osi
u
m
.
2005; 84
0-84
4.
[10]
T
s
eng CC. Imp
r
oved D
e
sig
n
o
f
Digital F
r
actio
nal-O
rd
er Differentiators Usi
n
g F
r
actiona
l Sa
mple De
la
y.
IEEE Trans. On Circuits and
System
s I: Regular Papers
. 20
06; 5(1):19
3
-2
03.
[11]
T
s
eng CC. D
e
sign
of F
I
R a
nd IIR fracti
on
al order
Simp
son dig
i
tal inte
grators.
Si
gna
l
Process
i
ng
.
200
6; 87(5): 10
45-1
057.
[12]
Yang Z
h
u-zh
o
ng, La
ng F
a
n
g
-ni
an. T
he Constructio
n
of
F
r
actiona
l Diff
e
renti
a
l Grad
i
ent Operator.
Journ
a
l of Co
mputatio
na
l Informati
on Syste
m
s
. 2011; 7(1
2
): 432
8-43
42.
[13] Hou
gh
P.
Meth
od an
d Mea
n
s for Recog
n
i
z
i
n
g Co
mp
lex Pat
t
erns
. U.S. Patent 306
96
54, 1
962.
[14]
Ibaa Jam
a
l, M
Usman Akra
m,
Anam T
a
riq. Retin
a
l Ima
ge Pre
p
rocess
ing: Back
grou
nd a
nd N
o
is
e
Segmentation.
T
E
LKOMNIKA Indon
esi
an Jou
r
nal of Electric
al Eng
i
ne
eri
n
g
.
2012; 1
0
(3): 5
37-5
44.
[15]
Xi
nso
ng W
a
n
g
, Guofen
g Qi
n. Pavem
ent
Image Se
gme
n
tation B
a
se
d
on F
C
M Alg
o
rithm Usi
n
g
Neig
hb
orh
ood Information.
T
E
LKOMNIKA Indo
nesi
an Jo
u
r
nal
of Electric
al Eng
i
n
eeri
ng.
2012; 1
0
(7);
161
0-16
14.
Evaluation Warning : The document was created with Spire.PDF for Python.