Internati
o
nal
Journal of Ele
c
trical
and Computer
Engineering
(IJE
CE)
V
o
l.
6, N
o
. 1
,
Febr
u
a
r
y
201
6,
pp
. 16
0
~
16
6
I
S
SN
: 208
8-8
7
0
8
,
D
O
I
:
10.115
91
/ij
ece.v6
i
1.9
343
1
60
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
Def
e
ct Detecti
o
n in Ceramic Im
ages Using Sigm
a Edge
Information and Cont
our Trackin
g
M
e
th
od
Kwa
n
g
-
B
a
ek Kim*
,
Yo
ung Woo
n
Woo*
*
* Department of
Computer Engin
e
e
r
i
n
g,
Si
ll
a Unive
r
si
ty
, Kore
a
** Departm
e
n
t
o
f
Multim
edi
a
En
gineer
ing, Dong-Eui Univ
ersit
y
,
Korea
Article Info
A
B
STRAC
T
Article histo
r
y:
Received Aug 18, 2015
Rev
i
sed
No
v 9, 201
5
Accepted Nov 27, 2015
In this paper, w
e
suggest a method of
detecting
defects b
y
apply
i
ng Hough
transform and least squares on
ceramic
images ob
tain
ed from non-
destructive
testing
.
In th
e
ceramic
images
obtain
e
d from non-destructive testing
,
the
background
area, where
the d
e
fect does
no
t exist, commonly
sho
w
s gradual
change of lumin
o
sity
in ver
tical
dire
ction. In ord
e
r to extract the
background
area which is go
ing to be used in
the
d
e
tection
of
defects, Hough transform is
perform
ed to rot
a
te th
e cer
am
ic i
m
a
ge in a wa
y
t
h
at the dir
e
c
tion
of overall
lum
i
nos
it
y
chan
ge li
es
in th
e v
e
rti
cal
dire
ction
as
m
u
ch as
po
s
s
i
ble. L
eas
t
square is then ap
plied on the ro
ta
ted im
age to app
r
oxim
a
te the
con
t
rast valu
e
of the b
ackgrou
nd are
a
.
The
ext
r
act
ed ba
ckgrou
nd are
a
is
us
ed
f
o
r extr
act
ing
defects from the ceramic images
. In th
is p
a
per
we applied this
method on
ceram
ic
im
ages
acquir
e
d from
n
on-des
t
ruct
ive
te
s
ting. I
t
was
con
f
irm
e
d tha
t
extracted b
ackg
r
ound area could be ef
f
ectively applied for searching
th
e
s
ection
where
th
e def
e
c
t
exis
ts and detecting
the
defect.
Keyword:
Ceramic im
age
C
ont
ou
r t
r
ac
ki
ng
m
e
t
hod
Defect detection
Non
-
d
e
st
ru
ctive testin
g
Si
gm
a edge i
n
f
o
rm
at
i
o
n
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
:
Y
oun
g W
o
o
n
W
o
o
,
Depa
rt
m
e
nt
of
M
u
l
t
i
m
e
di
a En
gi
nee
r
i
n
g,
Dong
-Eu
i
Un
iverity,
1
7
6
Eo
m
g
w
a
ng
no
B
u
san
j
i
n
-G
u, Busan 614-
714
,
K
o
r
ea.
Em
a
il: ywwo
o@d
e
u
.
ac.kr
1.
INTRODUCTION
No
n-
dest
ruct
i
v
e t
e
st
i
ng i
s
a
speci
al
m
e
t
hod by
w
h
i
c
h t
h
e st
at
us, chara
c
t
e
ri
st
i
c
s, i
n
t
e
rnal
st
ruct
u
r
e
and e
x
istence
of fla
w
can be
analyzed
withou
t affecting
th
e o
r
ig
i
n
al sh
ap
e an
d
fu
n
c
ti
o
n
of th
e
m
a
terial
o
r
th
e
pr
o
duct
.
No
n-
dest
r
u
ct
i
v
e t
e
st
i
ng i
n
cl
u
d
es m
a
gnet
i
c
pa
r
ticle testin
g
,
liq
u
i
d
p
e
n
e
tran
t testin
g
,
electro
m
a
g
n
e
tic
in
du
ctio
n
testin
g, rad
i
o
g
rap
h
ic testin
g
,
u
ltrason
ic tes
tin
g
,
leak
testin
g
,
v
i
su
al in
sp
ectio
n, acou
s
tic emissio
n
t
e
st
i
ng, i
n
fra
re
d t
e
st
i
ng,
def
o
r
m
at
i
on t
e
st
i
ng and m
o
re.
F
o
r
no
n
-
dest
ruct
i
v
e t
e
st
i
ng of ce
r
a
m
i
c
m
a
t
e
ri
al
li
qui
d
penet
r
ant testing is use
d
[1]. In liqui
d
pene
trant testin
g,
penetra
n
t sol
u
tion is firs
tly spread
on the
surface
of
t
h
e speci
m
e
n. Ove
rfl
owi
ng
p
e
net
r
a
n
t
sol
u
t
i
on i
s
rem
oved
and de
vel
o
per
i
s
appl
i
e
d. Th
e exi
s
t
e
nce of
a fl
aw
on t
h
e su
rface
and i
t
s
l
o
cat
i
on ca
n
be co
n
f
i
r
m
e
d by
det
ect
i
ng t
h
e e
x
u
d
i
n
g pe
net
r
a
n
t
sol
u
t
i
o
n w
h
i
c
h wa
s
abs
o
rbed into the
flaw i
n
th
e
s
u
rface
of the s
p
ecim
e
n [2].
Non
-
d
e
st
ru
ctive testin
g
is
p
e
rform
e
d
to
en
h
a
n
ce th
e reliab
ility
o
f
th
e p
r
o
d
u
c
t, i
m
p
r
o
v
e
man
u
f
act
u
r
i
n
g
tech
no
log
y
and
redu
ce m
a
n
u
factu
r
i
n
g cost
s. Thoroug
h quality co
n
t
ro
l enh
a
n
c
es th
e reliab
ility
o
f
t
h
e produ
ct, wh
ich
allo
ws
th
e u
s
ers to
t
r
ust an
d
u
s
e
t
h
e
m
with ease. Also appl
y
i
ng n
o
n
-
dest
r
u
ct
i
v
e t
e
st
i
ng
du
ri
n
g
t
h
e m
a
nu
fact
u
r
i
n
g st
age w
oul
d
en
able d
e
fectiv
e pro
d
u
c
ts to
go
u
nde
r control at appropriate mom
e
nt,
savi
n
g
t
i
m
e and
res
o
urces
.
Ho
we
ver
,
fi
na
l
con
f
i
r
m
a
t
i
on of
p
r
od
uct
s
or m
a
t
e
ri
al
s t
h
at
u
n
d
er
we
nt
n
o
n
-
d
e
stru
ctiv
e testin
g
is do
n
e
b
y
v
i
su
al in
sp
ectio
n. Sin
ce
v
i
sual in
sp
ectio
n
is a
m
a
n
u
a
l p
r
o
c
ess, it req
u
i
res a lo
t
of t
i
m
e and m
a
np
o
w
er
. I
n
a
d
d
i
t
i
on, t
h
e
r
e
wo
ul
d
be
di
ffe
re
n
ces am
ong t
h
e
t
e
st
resul
t
s
d
u
ri
ng
vi
s
u
al
i
n
s
p
e
c
t
i
o
n
d
u
e
to
i
n
terv
entio
n
o
f
in
sp
ect
o
r
s’ sub
j
ectiv
it
y. Thu
s
, pro
duct b
eco
m
e
s less reliab
l
e.
Evaluation Warning : The document was created with Spire.PDF for Python.
I
S
SN
:
2
088
-87
08
IJEC
E V
o
l
.
6, No
. 1, Feb
r
uar
y
20
1
6
:
16
0 – 16
6
16
1
C
e
ram
i
c
m
a
t
e
ri
al
s t
h
at
have
hi
g
h
t
h
e
r
m
a
l
resi
st
ance, l
o
w
den
s
i
t
y
, an
d
hi
gh
de
g
r
ee
of
h
a
rd
ness
are
a
traditional NDT
application
a
r
ea. Howe
ver, cracks
,
s
p
iracl
es, an
d
ot
her
f
o
rei
gn s
u
bst
a
n
ces t
h
at
fo
rm
vari
o
u
s
defects on
the
surface ha
ve negative
in
flue
nce on its reliabilit
y and
ha
rdness
[3]. Thus,
we
need an effective
m
e
t
hod t
o
det
ect
t
h
em
from
ceram
i
c im
age. There
f
ore,
t
h
i
s
resear
ch
o
n
a
u
t
o
m
a
t
i
call
y
ext
r
act
i
n
g t
h
e
d
e
fect
s
fr
om
t
h
e im
ages acq
ui
re
d
fr
o
m
non-
dest
r
u
ct
i
v
e t
e
st
i
n
g
w
o
ul
d s
a
ve
t
i
m
e
and
m
a
npo
wer
,
as wel
l
as e
n
ha
nci
n
g
th
e reliab
ility of th
e test, even
t
u
ally i
m
p
r
ov
ing
th
e qu
ality o
f
th
e test.
2.
THE PR
OPO
S
ED
METHO
D
F
O
R
E
X
TR
ACTI
N
G
DE
FECTS F
R
O
M
CERI
MI
C
IMA
G
ES
Co
nv
en
tio
n
a
l
meth
o
d
to
ex
tract d
e
fects from
cera
m
ic im
a
g
es[4] incl
ude
s
two
processes
,
classifyi
ng
the im
age into 9 clusters
usi
ng t
h
e contras
t
values
of the
im
age and extracting ca
ndi
date objects showi
n
g
defect
s
usi
n
g e
ach cl
ust
e
r’s c
ont
rast
val
u
es.
Ho
we
ver t
h
i
s
c
o
n
v
e
n
t
i
onal
m
e
t
h
o
d
has a
pr
o
b
l
e
m
t
h
at
duri
ng t
h
e
pr
ocess
o
f
e
x
t
r
act
i
ng
8m
m
im
age,
due
t
o
l
o
w
c
ont
ra
st
va
lue of
de
fect
a
r
ea, noise ar
ea is extracted ins
t
ead of
the de
fect area. In t
h
is pa
per, in or
d
e
r to so
l
v
e th
is p
r
ob
lem
we ex
t
r
act th
e defect
a
r
ea
by
det
ect
i
ng h
o
ri
z
ont
al
an
d ver
tical edg
e
s, th
en
co
rr
ectin
g
t
h
e
b
ound
ar
y
o
f
th
e defect ar
ea
b
y
app
l
yin
g
con
t
ou
r tr
ack
i
n
g m
e
th
o
d
to
the detecte
d
e
d
ges. Process
of analyzing defe
cts
fr
om
ul
t
r
asou
n
d
i
m
ages i
s
sh
ow
n i
n
Fi
g
u
r
e
1.
Fi
gu
re
1.
Pr
oce
ss o
f
e
x
t
r
act
i
ng de
fects from
ceram
ic images
Using the
fact
that defect a
r
ea
show
brighter
cont
rast tha
n
other a
r
eas in ce
ram
i
c
im
ages, ROI a
r
ea is
ex
tracted. Fin
e
n
o
i
se is rem
o
v
e
d
b
y
app
l
yin
g
Blurri
n
g
m
e
th
od
to
ex
tracted
RO
I area. Th
en
Prewitt mask
m
e
t
hod i
s
ap
pl
i
e
d t
o
t
h
e R
O
I
area f
o
r
det
ect
i
n
g
ve
rt
i
cal
an
d
h
o
ri
z
ont
al
e
d
g
e
s. T
h
e
det
ect
ed e
dge
s are
ap
pl
i
e
d
wi
t
h
c
ont
ou
r t
r
acki
n
g m
e
t
hod
t
o
co
rrect
t
h
e
bo
u
nda
ry
o
f
de
fect
area.
B
l
ob
Label
i
n
g m
e
t
hod
i
s
ap
pl
i
e
d t
o
t
h
e
corrected im
age to e
x
tract
the
final
defe
ct area.
Defect
area s
h
ows
bri
ght
e
r
cont
rast
t
h
an ot
her ar
ea
s in ceram
ic
images as
sho
w
n i
n
Fi
gu
re 2
(
a)
.
Using
th
is fact, areas
with
lower con
t
rast valu
e th
an
de
fe
ct
area was c
o
nsi
d
e
r
ed a
s
n
o
i
se and
rem
oved. T
h
e
resu
lt is sho
w
n in
Fi
g
u
re
2
(
b).
Fi
gu
re
2.
Ext
r
a
c
t
e
d R
O
I a
r
ea
Evaluation Warning : The document was created with Spire.PDF for Python.
IJECE
ISS
N
:
2088-8708
Def
ect
Det
ect
i
o
n
i
n
C
e
r
a
mi
c
Im
ag
es
Usi
n
g
Si
g
m
a
E
dge
I
n
f
o
rm
at
i
o
n
an
d
C
ont
our
Tr
ack
i
ng …
(
Y
.
W
. W
oo)
16
2
B
l
urri
ng m
e
t
hod
i
s
a m
e
t
hod
use
d
t
o
rem
o
v
e
fi
ne
n
o
i
s
e are
a
fr
om
ext
r
act
ed R
O
I a
r
ea
by
bl
u
rri
n
g
t
h
e
b
ackgr
oun
d
o
r
th
e fo
cu
s. Figur
e 3(
a)
is ex
tr
acted
RO
I
a
r
ea
whe
r
eas Fi
gure 3(b) is e
x
trac
ted ROI a
r
ea a
pplied
wi
t
h
B
l
u
rri
ng
m
e
t
hod.
Fi
gu
re
3.
R
e
sul
t
of
ap
pl
y
i
ng
bl
ur
ri
n
g
m
e
t
hod
Fi
gu
re 4 s
h
o
w
s a 3x3 m
a
sk i
n
w
h
i
c
h ne
w v
a
l
u
e fo
r a spec
i
f
i
c
pi
xel
i
s
det
e
rm
i
n
ed as t
h
e avera
g
e of
sur
r
o
u
ndi
n
g
pi
xel
s
’
val
u
e
s
. T
h
e
val
u
es i
n
t
h
e 3
x
3
m
a
sk i
n
Fi
gu
re
4 i
s
set
up
t
o
m
a
ke t
h
e
wei
g
ht
of t
h
e
m
a
sk
t
o
be
1
f
o
r
bl
ur
ri
n
g
.
Fi
gu
re 4.
W
e
i
g
ht
m
a
sk
fo
r bl
u
r
ri
n
g
Whi
l
e
c
onse
r
v
i
ng e
dge
s i
n
f
i
ne n
o
i
s
e rem
ove
d R
O
I a
r
e
a
, im
pul
se n
o
i
s
e i
s
rem
oved
by
ap
pl
y
i
ng
Median filtering
m
e
thod which
is used
to rem
ove
rem
a
ining noise in the ROI a
r
ea.
Fi
gure
5(a
)
s
h
ows the
result of a
pplying Blurring m
e
thod. Fi
gure
5(b) shows the resu
lt of applying Media
n
filtering
m
e
thod on fine
noi
s
rem
oved
R
O
I a
r
ea.
Fig
u
re 5
.
Med
i
an
filterin
g
In
noi
se rem
o
v
e
d R
O
I a
r
ea, p
a
rt
whe
r
e t
h
e b
r
i
g
ht
ness
sh
ows d
r
am
atic ch
a
n
g
e
is con
s
id
ered
to
b
e
the
sl
ope
. The ed
g
e
i
s
det
ect
ed by
usi
ng
th
e fact th
at first d
e
riv
a
tiv
e v
a
l
u
e o
f
th
e slop
e is v
e
ry larg
e or
sm
a
ll.
Equ
a
tio
n (1
) sh
ows t
h
e
p
r
imar
y d
i
fferen
tial op
eration
.
∆
,
,
(1
)
Evaluation Warning : The document was created with Spire.PDF for Python.
I
S
SN
:
2
088
-87
08
IJEC
E V
o
l
.
6, No
. 1, Feb
r
uar
y
20
1
6
:
16
0 – 16
6
16
3
In
eq
uat
i
o
n (
1
)
vect
or
(
∆
) is the slope
of im
age
f(x,y)
.
ex
ists for
d
i
fferen
tiatio
n
in
ho
rizo
n
t
al
di
rect
i
o
n an
d t
hus
ve
rt
i
cal
edge co
rres
p
on
di
ng t
o
col
u
m
n
is det
ect
ed.
ex
ists for d
i
fferentiatio
n
in
v
e
rti
cal
di
rect
i
o
n a
n
d t
hus
h
o
ri
zo
nt
al
edge
c
o
r
r
es
po
ndi
ng
t
o
ro
w
i
s
det
ect
ed.
T
h
eref
ore
,
is a co
lu
m
n
d
e
tect
o
r
and
i
s
a ro
w
det
e
c
t
or. B
y
a
ppl
y
i
ng t
w
o
det
ect
ors
si
m
u
l
t
a
neousl
y
wi
t
h
gra
d
i
e
nt
vect
o
r
(
∆
), edges ca
n be
clearly
d
e
tected
[5
]. In
th
is p
a
p
e
r
am
o
n
g
v
a
ri
o
u
s
pr
im
ary d
i
fferen
tial mask
s,
we
u
s
e Prewitt m
a
sk
meth
o
d
wh
ich
is less sen
s
itiv
e th
an
So
b
e
l m
a
sk
in detecti
n
g
ch
ange of
b
r
igh
t
n
e
ss. Figu
re 6 shows Prewitt m
a
sk
.
Fig
u
re 6
.
Prewitt
m
a
sk
Figure 7(a) s
h
ows t
h
e res
u
lt of applying Median
filteri
ng m
e
thod
[6]
and Fi
gure 7(b) s
h
ows the
resu
lt
o
f
app
l
yin
g
Prewitt m
a
s
k
to no
ise
remo
v
e
d
R
O
I area.
Fig
u
re
7
.
Resu
l
t
o
f
app
l
yin
g
Prewitt m
a
sk
to
ROI area
Ob
ject
s ar
e e
x
t
r
act
ed
by
a
ppl
y
i
n
g
c
ont
o
u
r t
r
ac
ki
ng m
e
t
h
o
d
t
o
det
e
c
t
ed ed
ges
[7]
.
Am
ong t
h
e
extracted
obje
cts with bright cont
rast,
objects
with s
m
aller size than
the a
v
era
g
e size is considere
d
un
necessa
ry
an
d rem
ove
d. O
b
ject
s wi
t
h
a
v
erage size is considere
d
candidate
defect area
and its boundary is
co
rrected
.
Figure 8(a) is th
e
resu
lt o
f
app
l
yin
g
Prewitt m
a
sk
an
d
Figu
re
8
(
b
)
sho
w
s t
h
e resu
lt o
f
correcti
n
g
t
h
e
b
oun
d
a
ry o
f
d
e
fect
area
in
Prewitt
m
a
sk
.
Fi
gu
re
8.
A
p
pl
i
cat
i
on
of
co
nt
o
u
r
t
r
acki
n
g
m
e
ho
d a
n
d c
o
r
r
ec
t
i
on
of
de
fect
a
r
ea
bo
u
nda
ry
Evaluation Warning : The document was created with Spire.PDF for Python.
IJECE
ISS
N
:
2088-8708
Def
ect
Det
ect
i
o
n
i
n
C
e
r
a
mi
c
Im
ag
es
Usi
n
g
Si
g
m
a
E
dge
I
n
f
o
rm
at
i
o
n
an
d
C
ont
our
Tr
ack
i
ng …
(
Y
.
W
. W
oo)
16
4
Blo
b
Lab
e
ling is app
lied
to
ROI area, i
n
wh
ich
th
e
bo
un
d
a
r
y
of
th
e def
ect
area is c
o
rrected, t
o
extract the
fina
l defect area
. Figure
9(a)
shows the im
ag
e with c
o
rrected
bounda
ry of
de
fect
area. Figure
9(b)
shows
the
final
de
fect area
ac
qui
re
d fr
o
m
th
e bo
und
ar
y co
rr
ected
im
ag
e by ap
p
l
ying
Blob
Lab
e
ling
m
e
t
h
od
.
Fi
gu
re 9.
Det
e
ct
i
on of
de
fect
3.
E
X
PERI
MEN
T
AN
D
RES
U
LT AN
ALY
S
ES
In t
h
i
s
pa
per
,
p
r
o
p
o
sed m
e
t
hod t
o
det
ect
t
h
e defect
area wa
s im
pl
em
ent
e
d by
Vi
sual
St
u
d
i
o
2
0
10 C
#
o
n
PC equ
i
pp
ed
w
ith
Pen
t
i
u
m
(
R)
D
u
al-
C
or
e CPU
T420
0 2
.
0
0
G
B
RAM. 13
60x
102
4 size i
m
ag
es acq
u
i
r
e
d
fr
om
8
m
m
and
10m
m
obt
ai
ned
fr
om
di
f
f
ere
n
t
n
o
n
-
d
es
t
r
uct
i
v
e t
e
st
s
were
use
d
a
s
speci
m
e
ns f
o
r t
h
e
expe
ri
m
e
nt
. Th
e res
u
l
t
o
f
det
ect
i
ng
defect
s
u
s
i
n
g
t
h
e m
e
t
h
o
d
pr
o
pose
d
i
n
t
h
i
s
pape
r i
s
s
h
ow
n i
n
Fi
gu
re
10
.
Fig
u
r
e
10
. Th
e pr
opo
sed pro
c
ess of
d
e
tecting
d
e
f
ects
Fi
gu
re 1
1
(
a) s
h
o
w
s t
h
e
resul
t
of ext
r
act
i
n
g
candi
dat
e
def
ect
ob
ject
by
appl
y
i
n
g
t
h
e c
o
n
v
e
n
t
i
ona
l
m
e
thod
[4]. Fi
gure
11(b) shows
the
res
u
lt of e
x
tracting
candi
date
defe
ct
object
by a
pplying t
h
e m
e
thod
su
gg
ested
in this p
a
p
e
r.
As s
h
ow
n
i
n
F
i
gu
re
11
, c
o
nv
ent
i
onal
m
e
t
hod
of
det
ect
i
n
g
defect
s
i
n
c
e
ra
m
i
c im
ages [3]
ap
pl
i
e
d
K
-
m
eans al
g
o
ri
t
h
m
t
o
ext
r
act
c
a
ndi
dat
e
defec
t
areas
base
d
on
cl
ust
e
rs’
c
ont
rast
dat
a
.
Ho
we
ver
d
u
ri
ng
t
h
e
clustering
proc
ess de
fect area
was cluste
red with nois
e a
r
ea. Thus t
h
e defect area c
oul
d not be
extra
c
ted.
Ho
we
ver
,
m
e
tho
d
pr
op
ose
d
i
n
t
h
i
s
pa
pe
r de
t
ect
defect
area
by
ap
pl
y
i
ng c
ont
ou
r t
r
ac
ki
n
g
m
e
t
hod
on e
d
ge dat
a
analyzed from ROI area. Ta
ble 1 sh
ows t
h
e num
b
er of
im
ages that we
re success
f
ul
in detecting defects
am
ong
1
2
ce
ra
m
i
c im
ages by
appl
y
i
n
g
pr
o
p
o
s
ed m
e
t
hod
.
Evaluation Warning : The document was created with Spire.PDF for Python.
I
S
SN
:
2
088
-87
08
IJEC
E V
o
l
.
6, No
. 1, Feb
r
uar
y
20
1
6
:
16
0 – 16
6
16
5
Fi
gu
re
1
1
. C
o
m
p
ari
s
on
of
de
fect
det
ect
i
o
n
m
e
t
hods
Table 1. Num
b
er of succes
s
a
n
d
failure of
de
fect
detection
Type
Success
Failure
8
m
m
5/9
4/9
10
m
m
3.
3
0/3
Fi
gu
re
12
sh
o
w
s t
w
o e
x
am
pl
es o
f
fai
l
u
re i
n
det
ect
i
n
g
de
fe
ct
whi
l
e
a
ppl
y
i
ng
t
h
e
pr
o
pose
d
m
e
t
hod.
I
n
som
e
8
mm cera
m
ic images, object size
of
defect area a
n
d noise area
was si
m
ilar. In these cases
noi
se areas
were
extracte
d
as de
fect area
s.
Fi
gu
re
1
2
. T
w
o E
x
am
pl
es of
fai
l
u
re i
n
det
e
c
t
i
ng
defect
4.
CO
NCL
USI
O
N
In t
h
i
s
pa
per
,
we pr
op
ose
d
a
m
e
t
hod t
o
det
ect
defect
s i
n
ceram
i
c
im
ages acqui
re
d fr
om
non-
d
e
stru
ctiv
e testin
g
.
Usi
n
g
th
e fact th
at defec
t
area show brighte
r
contrast
than
othe
r areas
in ceramic images,
Evaluation Warning : The document was created with Spire.PDF for Python.
I
J
ECE
I
S
SN
:
208
8-8
7
0
8
Def
ect
Det
ect
i
o
n
i
n
C
e
r
a
mi
c
Im
ag
es
Usi
n
g
Si
g
m
a
E
dge
I
n
f
o
rm
at
i
o
n
an
d
C
ont
our
Tr
ack
i
ng …
(
Y
.
W
. W
oo)
16
6
R
O
I area i
s
ex
t
r
act
ed. Fi
ne
n
o
i
s
e i
s
rem
ove
d by
ap
pl
y
i
ng
B
l
urri
ng m
e
t
hod t
o
e
x
t
r
act
e
d
R
O
I area
. M
e
di
an
filtering m
e
thod is a
p
plied to rem
ove rem
a
ining
noise
in
fine noise
rem
oved ROI a
r
ea.
The
n
P
r
ewitt
m
a
sk
m
e
t
hod i
s
ap
pl
i
e
d t
o
t
h
e R
O
I
area f
o
r
det
ect
i
n
g
ve
rt
i
cal
an
d
h
o
ri
z
ont
al
e
d
g
e
s. T
h
e
det
ect
ed e
dge
s are
ap
pl
i
e
d
wi
t
h
c
ont
ou
r t
r
acki
n
g m
e
t
hod
t
o
co
rrect
t
h
e
bo
u
nda
ry
o
f
de
fect
area.
B
l
ob
Label
i
n
g m
e
t
hod
i
s
ap
pl
i
e
d t
o
t
h
e
corrected im
ag
e to extract th
e fi
nal
defect
ar
ea. Fut
u
re rese
arch w
o
ul
d be
foc
u
se
d o
n
sol
v
i
n
g t
h
e pr
o
b
l
e
m
of
being
unable t
o
e
x
tract
defec
t
areas
from
som
e
8
mm, 11mm, 16mm
,
22mm
cera
m
ic
im
ages. T
o
s
o
l
v
e t
h
is
pr
o
b
l
e
m
,
defect
area ext
r
act
i
on
per
f
o
r
m
a
nce wo
ul
d
be e
nha
nce
d
by
a
p
pl
y
i
ng F
u
zzy
i
n
fe
rence m
e
t
hod t
o
analyze de
fects
’
various
m
o
rphol
ogical
features.
REFERE
NC
ES
[1]
http://www.
kandt.
or.
k
r
[2]
S. Vasilic
and Z. Hocenski, “The
Edge Det
ect
i
ng
Methods in Ceram
i
c Til
e
s Defect
s Detect
ion”,
IEEE Internationa
l
Symposium on I
EEE
, vol. 1
,
pp
.
469-472, 2006
.
[3]
Y.W. Woo and
K.B.
Kim
,
“
D
et
ect
ion of F
l
aws
in Cer
a
m
i
cs
M
a
ter
i
als
Us
ing N
on-des
t
ruct
ive
T
e
s
t
”,
Journal o
f
Korean Information Communications Society
, vo
l. 5
,
no
. 3
,
pp
. 32
1-326, 2010
.
[4]
S.W. Hwang, K.B.
Kim, Y.W. Woo, "Fault De
tection of C
e
ramic Imaging Us
ing Nondestructive Testing",
Korea
Electronic
and Communication Scien
ces
pap
ers
, vol. 7, no. 2, pp. 409, 2013.
[5]
Z. Hocenski
,
S. Vasilic and V.
Hocen
ski, “
I
m
p
roved Cann
y
Ed
ge Detec
t
or
in Ceram
i
c Ti
les Defect Det
e
c
tio
n”
,
IEEE Industrial
Electronics, IEC
O
N 2006-32nd
Annual Con
f
erence on
IEEE
, pp
. 3328-3331, 200
6.
[6]
J.W
.
Lim
,
E.K.
Kim
,
"Noise
Reduction b
y
Filter Im
prove
m
e
nt in Mixed
Noise Im
age",
Electronics an
d
Engineering of
Korea papers
, v
o
l. 50
, no
. 5
,
pp
.
1251, 2013
.
[7]
K.B. Kim, D.H. Song
and W.J.
Lee, “Flaw
Detection
in Cer
a
mics using Sigma
Fuzzy
Binarization and Gaussian
Filtering
Metho
d
”,
In
ternat
iona
l Journal
of
Mu
ltimedia
and Ub
iquitous Eng
i
neering
, vo
l. 9, no
. 1, pp. 403-414
,
2014.
BIOGRAP
HI
ES OF
AUTH
ORS
Kwang-Baek Ki
m
receiv
e
d his
M
.
S
.
and
the
P
h
.D. deg
r
ees
in D
e
partm
e
nt
of Co
m
puter S
c
ien
c
e
from Pusan National Univ
ersity
,
Busan,
Korea, in
1993 and
1999,
respectively
.
From 1997
to present, h
e
is a professor, De
partment of Computer
Engineering, and Silla
Univers
i
t
y
in
K
o
rea.
He is
curr
entl
y an as
s
o
cia
t
e
editor
for J
o
urnal of
Th
e K
o
rea S
o
c
i
et
y
of
Com
puter and
In
form
ation,
and
T
h
e Open
Artifi
c
i
a
l In
tel
ligen
ce
Journal (USA).
His
res
earch
int
e
res
t
s
inc
l
ude fu
zz
y neur
al n
e
tw
ork and app
lication, bio
i
nformatics and imag
e
processing.
Young W
oon
W
oo received t
h
e B.S. degre
e
, M.S.
degree a
nd Ph.D. degre
e
in ele
c
troni
c
engineering
fro
m Yonsei University
, Seou
l, Korea
in 1989
, 1991
and 1997
, r
e
spectiv
ely
.
From Septem
ber, 1997 to present, he has been
a professor in
Departm
e
nt of Multim
edia Eng
.,
Dong-Eui University
, Busan
,
Ko
rea.
His
res
earch in
teres
t
s
ar
e in t
h
e are
a
of art
i
f
ici
a
l int
e
l
ligen
ce,
im
age proc
es
s
i
ng, pat
t
ern
recognition and medical
information.
Evaluation Warning : The document was created with Spire.PDF for Python.