TELKOM
NIKA Indonesia
n
Journal of
Electrical En
gineering
Vol. 12, No. 8, August 201
4, pp. 6036 ~ 6046
DOI: 10.115
9
1
/telkomni
ka.
v
12i8.519
1
6036
Re
cei
v
ed
No
vem
ber 2
2
, 2013; Re
vi
sed
April 3, 2014;
Accept
ed Ap
ril 20, 2014
Nonuniform Defect Detection of Cell Phone TFT-LCD
Display
Jahan
gir Ala
m
S.M.*, Hu Guoqing
Dep
a
rtment of Mecha
n
ica
l
& Electric
al E
ngi
neer
ing,
Xi
ame
n
Univ
ersit
y
,
Room 2
28, Sci
ence Bu
ild
in
g, 361
00
5, Simin
g
Distric
t, Xiam
en, F
u
jia
n, Chi
na, telp/fa
x: +
86-59
2-21
86
39
3
*Corres
p
o
ndi
n
g
author, e-ma
i
l
: jaha
ngir
_
u
i
ts@
y
ah
oo.com
A
b
st
r
a
ct
Unev
en and Nonunifor
m
ity
(
M
ur
a) of Thin
Film
Transistor
Liqu
id Crystal Display (TFT-LCD) is
a
ma
jor pro
b
l
e
m of cell pho
ne
displ
a
y. T
he di
fferent ty
pes of uneve
n
an
d n
onu
nifor
m
ity ar
e decre
ase
d
the
p
e
r
fo
rm
an
ce
of TFT-L
CD. To
e
c
o
nom
i
z
e an
d
in
crea
se
its p
e
r
fo
rm
an
ce
,
it is ne
ce
ssa
ry to
d
e
t
e
c
t the
s
e
kin
d
s
o
f
de
fe
ct
s. Th
e
cau
s
e
o
f
th
e
s
e typ
e
s
o
f
no
i
sy
d
e
f
ects ca
n
b
e
stim
ula
t
ed
b
y
the
m
a
te
ria
l
o
f
TFT,
intens
ity of ba
ck light, total
i
n
terna
l
reflecti
on, mirror for
m
of others
mat
e
rials, i
n
tern
al
light, a
nd exter
n
a
l
light. T
he
en
er
gy loss
an
d ga
in i
n
LC
D dis
p
l
a
y is a
noth
e
r i
ssue to
mak
e
t
hese
un
even
a
nd n
o
n
unifor
m
i
t
y.
T
he obj
ective
of this study is to investigate
and d
e
te
ct the defects of cell pho
ne
dis
p
l
a
y consi
deri
ng so
me
para
m
eters w
i
th i
m
age
a
naly
s
is. T
he
back
side
an
d th
e fr
ont si
de
of th
e
defects
hav
e
bee
n
observ
e
d
to
find the u
n
iq
ue
ness of that
de
fects and its mode
l.
Ke
y
w
ords
: un
even a
nd n
o
n
u
n
ifor
mity, defec
t detection, dev
iatio
n
, ener
gy
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
No
w a day, the use of TF
T-L
CD is very popular. Th
e cell pho
ne displ
a
y is on
e of the
wide
part
s
of
TFT-L
C
D di
splay. If the perfo
rman
ce
s of mobile p
hone
display
s
a
r
e not
hig
her
then it will n
o
t
be e
c
ono
mi
zed. T
o
stim
u
l
ate
the cell p
hone
TFT-LCD di
splay m
a
rket
and
attra
c
t
the custo
m
er,
it is
ne
ce
ssa
r
y to p
r
od
uce
the n
on
defe
c
t TFT
-L
CD d
i
splay. To
p
r
o
duce the
go
o
d
quality and
hi
gher pe
rform
ance TFT
-L
CD, it is n
eed
ed to d
e
tect
the defe
c
ts
such
as Mu
ra
or
uneven d
e
fe
cts o
r
non
un
iformity. Thus very
sm
all
scal
e
un
eve
n
and n
onun
iformity can
be
detecte
d by using Fa
st Fo
urie
r Tran
sfo
r
m (
FFT). The
FFT can det
ect different types of unev
en
and no
nunifo
rmity on cell
phon
e TFT-L
CD di
spl
a
y.
The sm
all si
zes of non
unif
o
rmity, uneven
impact of col
o
r, light inten
s
ities a
nd en
ergy di
ssip
ate, reflected n
o
ise a
r
e defe
c
t on cell p
h
o
n
e
TFT-L
C
D di
splay. The
r
e a
r
e two
came
ra sy
stems
th
at ca
n dete
c
t
the all type
s of uneve
n
a
nd
nonu
niformity
whi
c
h
dete
c
t
by u
s
ing
FF
T after th
at
it differentiate
the qu
ality of the TF
T-L
C
D
displ
a
y
.
The
cha
r
a
c
t
e
ri
st
ic
s si
ze
of
no
n
-
inf
o
rm
al d
e
fects and
un
e
v
en impa
ct can be
mea
s
u
r
ed.
Then it
can
be defin
ed th
e mod
e
l of the dete
c
t
ed
defect
s
. Cu
rrently, there a
r
e ma
ny defe
c
t
s
detectio
n
alg
o
rithm
s
ha
s b
een e
s
tabli
s
h
ed [1-8] for
T
FT-L
CD
su
ch
as
Wavelet
Tran
sfo
r
m [1, 2]
and level-set
method [3]. It is noted that the reg
r
e
s
sion di
agn
ost
i
cs al
go
rithm
[4] and usin
g
adaptive thre
shol
d algo
rithm to detect
defect in
T
FT-L
CD
are
prop
osed [5,
6] to detect the
defect
s
. It has bee
n devel
oped the Mini
mum Error T
h
re
shol
ding A
l
gorithm by J.
Kittler [7].
Image seg
m
entation, thre
sho
u
ldin
g, FFT an
al
ysis, an
gle
adju
s
tment
, energy
perfo
rman
ce
analysi
s
a
r
e
helpe
d to d
e
tect the
uneve
n
an
d no
nunif
o
rmity on
TF
T-L
CD. F
o
r hi
gh
perfo
rman
ce
and bette
r efficien
cy, it needs to in
sp
ect the image segmentat
io
n
with its prope
rly
wind
ow
re
sol
u
tion an
d si
zes. It is n
e
ce
ssary to di
sti
ngui
sh the
d
e
fect an
d no
n defe
c
t ima
ges
then it m
u
st
be
se
gment
ed the
un
even a
nd/or
n
onunifo
rmity
to dete
c
t the
defe
c
ts. T
h
ese
segm
ented i
m
age
s can b
e
pro
d
u
c
ed
b
y
FFT analyzing [9]. Finall
y
all kind
s of
cha
r
a
c
teri
sti
c
s
can b
e
dete
r
mine
d from
the model. After su
cce
ssfully defe
c
t detection
on TFT-LCD, i
t
differentiate
s the perfo
rma
n
ce of the p
r
o
ductio
n
.
2. Res
earc
h
Method
High
re
soluti
on Allied
GIGE 490
0 an
d Keyen
c
e h
i
gh spee
d H200
C mod
e
l
came
ra
s
have b
een
u
s
ed to t
a
ke
pi
cture i
n
Fig
u
re
1, in
spe
c
t, a
nd o
b
serve
th
e un
even
and
non
uniformity
Evaluation Warning : The document was created with Spire.PDF for Python.
TELKOM
NIKA
ISSN:
2302-4
046
Non
uniform
Defe
ct Dete
ction of Cell Ph
one TFT
-L
CD Displ
a
y (Ja
h
angir Alam
S.M.)
6037
on TFT-LCD.
To inspe
c
tio
n
of uneven and nou
nifo
rmity on mobile phone TF
T-LCD, it has been
use
d
the
s
e t
w
o
cam
e
ra
s
whi
c
h
can
ca
pture
high
re
solution im
age
s. Two
came
ra ha
s b
een
u
s
ed
for disting
u
ished the differences. To
ch
eck def
e
c
ts
on TFT-LCD,
it has been
use
d
ba
ck lig
ht
whi
c
h can co
ntrolled by A
R
M sy
stems.
The ba
ck
lig
ht make
s the
intensity of back si
de of the
TFT-L
C
D. In the front up
side, it inspe
c
ts
the uneven a
nd non
uniformity or any defects.
Figure 1. Une
v
en and Non
uniformity Inspectio
n
Equip
m
ents
3.
Theore
t
ical Analy
s
is
3.1.
Fast Fourie
r
Transform
The frequ
en
cy transfo
rm
of the Fast Four
ie
r Tra
n
sform is the most impo
rtant for
freque
ncy
re
spo
n
ses of
any si
gnal
[10]. The
FF
T an
alysi
s
can d
e
tect th
e un
even
a
n
d
nonu
niformity
on TFT-LCD whi
c
h
ca
n
faster an
d
save the tim
e
[1]. To economi
z
e a
n
d
to
prod
uce a la
rge amo
unt of
prod
uct by th
is metho
d
is
more
efficient
and in th
e ro
botic visi
on, it is
necessa
ry to
dete
c
t the
defect
s
mo
re
vastly. T
he
defect
s
cha
r
acteri
stics, p
o
sition, d
ept
h of
defect, mo
del
of defe
c
ts,
si
ze
s of va
riati
ons,
and
ba
ck
to front
sid
e
s’ i
n
sp
ectio
n
,
it is ne
ce
ssary
to introdu
ce the 3D FF
T. The FFT can b
e
m
odified int
o
3D FFT to
detect the def
ect as:
,
,
∑∑
∑
,
,
(
1
)
Whe
r
e,
,
,
is the gray level of the world co
ordin
a
tes
,
,
of
3D imag
e sig
nal in
the Equation
(1). Here
,
,
are
the coo
r
dina
tes in freq
ue
ncy domai
n a
nd
,
,
are the
size
of image axis, resp
ectively. The re
al ima
ges a
r
e
in the
time domain
whi
c
h can be
transl
a
ted int
o
freque
ncy re
spo
n
se [9]. Then if there
is any
defe
c
t the freq
ue
ncy re
sp
on
se can
sh
ow
the
uneven a
nd n
onunifo
rmity on the re
spo
n
s
e
s
by solvin
g the followin
g
equatio
n as:
,
,
300
,
1
,
,
9
(
2
)
If
a, b
and
c
are sm
all, the value
,
,
is larger. The
small
a, b, c
means lo
w
freque
ncy re
gion. So, the more lo
w freque
ncy va
l
ue is multipl
i
ed by the large
r
value [9].
Ke
y
ence
ca
m
e
ra
1
Ke
y
ence
ca
m
e
ra
2
Ca
p
tured Im
a
g
e
ARM S
y
ste
m
s
Cell Phone T
F
T-
LCD Displa
y
GigE
4900C
Evaluation Warning : The document was created with Spire.PDF for Python.
ISSN: 23
02-4
046
TELKOM
NI
KA
Vol. 12, No. 8, August 2014: 603
6 –
6046
6038
Thro
ugh thi
s
pro
c
e
ss, the
more lo
w fre
q
uen
cy co
m
p
o
nent is to put
a high p
r
op
ortion and for t
h
is
purp
o
se the Equation (2)
can b
e
rea
r
ra
nged a
s
follo
ws:
,
,
1,
1,
1
,
,
,
1
,
,
9
(3)
,
,
1,
129
,
1
,
,
,
1
,
,
9
(4)
∑
∑
∑
|
,
,
|
∑∑
|
,
,
|
(
5
)
The value
is obtaine
d from
Equation (5).
If this value large
r
than sp
ecific value
, i
t
is
determi
ned t
hat the segm
ented ima
ge
has
defe
c
t. In
the op
po
site ca
se, the
segmente
d
im
age
doe
s not
hav
e defe
c
t. The
lowe
r frequ
e
n
cy
comp
one
nt of segme
n
ted ima
ge
whi
c
h
ha
s defe
c
t
is
large
r
than th
at of segm
en
ted image
wh
ich h
a
s
not defect. The
specifi
c
value
is defined as
follow as
:
√
(
6
)
Whe
r
e,
0
.
0
1
,
is gray level me
an of the seg
m
ented ima
g
e
and
is win
dow
size
and it i
s
a
s
sumed th
at
128
. The
con
s
tant
value
is o
b
tained
expe
rimentally by
usi
n
g
Equation (6).
3.2. 3D Discr
ete Cosine T
r
ansform
The
3D Di
screte
Co
sin
e
Tran
sform
(DCT
) stra
ightforwardly
de
scrib
ed the
3
D
measurement
s from discre
t
e resp
on
se
s [9]. T
he une
ven and non
uniformity also unified by 2D
image pla
ne
and re
st of the dimen
s
io
n
can hel
p to
detect for ch
ara
c
teri
stics
of defects m
odel
and source of
defects, de
pth of defects.
Thus
h
e
re the 3D DCT
ca
n be define
d
as follo
ws:
,
,
,,
1
2
1
2
1
2
i.
e.
,
,
,
,
1
2
(
7
)
Her
e
,
,
,
are
3D DCT len
g
th of the axis;
,
,
is the step from left to right
side
an
d fron
t to ba
ck o
r
b
a
ck to
front
resp
on
se
s
whi
c
h i
s
in
cre
a
sed the
fre
que
ncy by
1/2
cy
cle
in the Equatio
n (7). The val
ue of
,
,
are the scala
r
fact
or;
,
,
is the am
plitude scalin
g
of the DCT [
1
]. The angle
of cosin
e
ca
n indicate
the angle of de
fects by usi
n
g different an
gle
and different
positio
n of the cam
e
ra
s
shown in Figu
re 1. It can b
e
increa
se
d the efficien
cy
to
detect the un
even and n
o
n
uniformity of cell ph
one TF
T-L
CD di
spl
a
y.
The Mo
dified
DCT
(M
DCT
)
as th
e lap
p
ed tran
sfo
r
m
ed is
com
p
a
r
ed to Fou
r
ie
r-rel
a
ted
transfo
rm
s [1
1] with Equat
ion (7
). Usua
lly it
has a h
a
lf of many outputs a
s
in
puts. It can
be
defined
as
a l
i
near fun
c
tion
as,
:
→
; where
is real n
u
mbe
r
sets. Th
e
2N
re
a
l
nu
mb
ers
,…,
can
be t
r
an
sform
ed into
the
N
real
nu
mbers
,…,
acco
rding to th
e formul
a
as:
∑
(
8
)
The inve
rse
MDCT i
s
kno
w
n
as the IM
DCT
an
d thi
s
tech
niqu
e i
s
kno
w
n
a
s
tim
e
-do
m
ain
aliasi
ng
can
c
ellation (TDA
C). Th
e IMDCT tra
n
sfo
r
m
s
N
re
al nu
mbers
,…
,
into
2N
real
numbe
rs
,…,
accordin
g to the formula fro
m
above Equati
on (8
) ca
n be
transfo
rmed
into
IMDCT a
s
:
Evaluation Warning : The document was created with Spire.PDF for Python.
TELKOM
NIKA
ISSN:
2302-4
046
Non
uniform
Defe
ct Dete
ction of Cell Ph
one TFT
-L
CD Displ
a
y (Ja
h
angir Alam
S.M.)
6039
∑
(
9
)
In the case
of a wind
owe
d
MDCT
with
the usual
wind
ow normali
zati
on, the
norm
a
lization
coefficie
n
t in
front of the IMDCT
shoul
d be multiplie
d by 2 (i.e., becomi
ng 2/
N). It
can
be n
o
ted
for 3
D
DCT
as 2/3. T
h
e
wind
ow
ha
s con
s
id
ere
d
for ima
ge p
r
o
c
e
ssi
ng a
s
t
he
IMDCT
wind
o
w
and the im
age si
gnal h
a
s
bee
n pro
c
e
s
sed a
s
a DCT signal p
r
o
c
essing [12].
3.2.
Nonunifor
m
Discrete F
ourier Tran
s
f
orm
3D No
nunifo
rm Discrete F
ourie
r Tra
n
sf
orm (NDF
T) [11] of a sequence
,
,
of
siz
e
is
:
,
,
∑∑
∑
,
,
(
1
0
)
Whe
r
e,
0
,
1
,….,
,
,
,
is the 3D z-tra
n
s
form
of
,
,
, a
nd
,
,
are arbitra
r
il
y distinct
points in the
4D
,
,
space. The above
equatio
n (10
)
can be exp
r
e
s
sed a
s
the vector fo
rm a
s
:
(
1
1
)
Whe
r
e, the matrix
is also dep
end
s only on the
choi
ce of those sam
p
ling
points.
Ho
wever,
even if tho
s
e
sampling
poin
t
s are di
stinct,
could
still
be
sing
ular.
No
rule
s fo
r
determi
ning
wheth
e
r th
e
matrix is no
nsin
gula
r
o
r
not have
be
en fou
nd. T
herefo
r
e, fo
r all
impleme
n
tation of 3D
NDFT, it is nee
ded to ju
st check
det
for a specifi
c
set of
sampli
ng
points. Thi
s
3
D
NDFT is h
e
l
pful to detect
for discrete u
neven an
d no
nuniformity.
4.
Unev
en and Nonuni
form Detec
t
ion Process
Size of image
is
4872
3276
in Figure
2. The image
has be
en se
gmented to d
e
tect the
defect
s
an
d i
t
performan
ce. The ima
g
e
se
gmentati
on impo
rtant
for efficie
n
t insp
ectio
n
, e
a
ch
segm
ented
i
m
age i
s
overl
appe
d ne
arby se
gmente
d
i
m
age
at lea
s
t
2 pixel
s
in
th
e ho
rizontal
a
n
d
vertical di
re
ction [13]. If segme
n
ted
im
age is
not overlap
p
ed other
se
gmented im
age,
perfo
rman
ce
of the prop
osed algo
rithm
tries to defe
c
t inspe
c
tion f
o
r ea
ch
seg
m
ented ima
g
e
.
The Figu
re 3
sho
w
s the order of inspe
c
tion of segm
e
n
ted image.
(a)
(b)
Figure 2. Une
v
en and Non
uniform
Defe
cts (a
) white (b) differe
nt types of defe
c
t
s
And then, pe
rform 3
D
FFT
to each se
gm
ented
ima
ge
and
calculate
comp
actn
ess that is
mean of lo
w frequ
en
cy regi
on. In orde
r t
o
cal
c
ul
ate
compa
c
tne
ss,
it has to defin
e low fre
que
n
c
y
regio
n
. The l
o
w fre
que
ncy
regio
n
is d
e
fined a
s
9
-
by-9 uppe
r left side exce
pt first ro
w an
d first
colum
n
an
d 9
-
by-9 l
o
wer le
ft side except
first column.
On the
cell p
hone T
FT-LCD di
splay p
a
n
e
l
it can b
e
lo
cated the
r
e dif
f
erent type
s
of def
ect
s
o
r
uneven
and
nonu
niformity
in Figu
re 2.
To
analyze the i
m
age
here it
has bee
n
con
s
ide
r
ed
the
Fi
gure
2(a) t
hat
have two d
e
fects which
a
r
e
Evaluation Warning : The document was created with Spire.PDF for Python.
ISSN: 23
02-4
046
TELKOM
NI
KA
Vol. 12, No. 8, August 2014: 603
6 –
6046
6040
white poi
nts (white un
even
n and no
nuni
formity). Afte
r FFT pro
c
e
s
sing, it can loo
k
in view of 3
D
plane in Fig
u
r
e 3. The two
peaks value
in the
Figure
3(a)
sho
w
s the 3D mo
del
of the uneven
and n
onu
nifo
rmity on the
surfa
c
e
which ha
s b
een i
ndicated by
arrow
whi
c
h
i
s
not
able. In
the
Figure 3(b
)
, there a
r
e hi
dd
en layers re
prese
n
tation of uneven a
nd n
onunifo
rmity.
(a)
(b)
Figure 3. Une
v
en and Non
uniformity Ch
eck by
3D Co
ntours and
Hi
dden L
a
yers Analysis
5.
Resul
t
s and
Discus
s
ions
The pe
rform
ance of defe
c
t detectio
n
of
the uneve
n
and no
nu
niformity have been
cal
c
ulate
d
by
differe
nt met
hod
s. Abo
u
t
100
cell
ph
o
ne TF
T-L
C
D
displ
a
ys
we
re che
c
ked
a
s
a
sampl
e
to de
tect uneven
and no
nunifo
rmity by us
in
g prop
osed
algorith
m
. Th
e statistics al
so
defined to
d
e
tect the
def
ect such a
s
Mura
or
une
ven and
non
uniformity [1
2]. The stati
s
tical
results
are in
spe
c
ted
by compa
r
ing i
n
t
he d
e
fecte
d
and
non
-defe
c
ted im
age
o
r
cell p
hon
e
TFT-
LCD p
anel. T
he d
e
fect
ca
n
be i
ndentifie
d by o
b
serv
in
g the m
ean
a
nd d
e
viation
statistics
of the
displ
a
y panel
. The mean,
varian
ce, an
d deviation were o
b
served
on the display panel in th
e
different po
sit
i
ons
of the ROI. If there is
any
defe
c
t exist on
ROI of
TFT-L
C
D the
n
the mea
n
a
nd
varian
ce val
ue is hi
ghe
r than non
d
e
fect ROI. Mean a
nd d
e
viation ca
n
find by followin
g
equatio
n:
∑
,
∑
(
1
2
)
Whe
r
e,
is th
e standa
rd deviation,
is the mean,
is
sampl
e
spa
c
e on ROI and
is
the numb
e
r o
f
sample val
u
e on
ROI. Th
us the
defe
c
ts on th
e cell phon
e TFT
-L
CD
display can
be define
d
as:
,
Def
ect
;
σ
,
μ
much
gr
eat
e
r than no
def
e
c
t
R
OI
No d
ef
ect ;
σ
,
μ
much less
than d
e
f
e
c
t
ed R
OI
(
1
3
)
Whe
r
e,
,
is the image pl
an
e. The ca
ptured imag
e ha
s been
segm
e
n
ted for p
r
ep
aring
the ROI an
d
che
c
ks the
b
o
x sp
aces.
T
he filterin
g
proce
s
s i
s
h
e
lpf
u
l for elimin
ating the
noi
se
and
more
con
c
ise
l
y. For b
e
tter analy
s
is
and
getting
bette
r result this p
r
ocess i
s
i
m
p
o
rtant to fini
sh
the statistical
pro
c
e
ss by
usin
g Equati
on (1
3)
. The
all pro
c
e
s
se
d has b
een
expre
s
sed in
the
followin
g
pro
posed alg
o
rit
h
m in Figure 4.
The mea
n
a
nd deviation
has b
een
co
mpared by
intensity mea
s
ureme
n
t [6]. On cell
phon
e TFT
-L
CD di
splay, t
here
a
r
e
som
e
sample
sp
ace
ha
s
bee
n create
d
o
n
cell
ph
one
T
FT-
LCD ima
ge
p
anel
by ‘Halcon’
softwa
r
e i
n
Figu
re
5(a).
The
sample
spa
c
e
ha
s
be
en in
dicated
by
red
sq
uare b
o
x. The
size
s of the
squ
a
re
box
e
s
a
r
e sa
me. Th
e
r
e a
r
e t
w
o
square b
o
xes
are
indicated
by arrow whi
c
h has an
u
nev
en
an
d
no
nu
niformity vast
ly i.e. there a
r
e defe
c
ted
ROI
whi
c
h ha
s be
en sh
own as
a magnified i
m
age in Fig
u
re 5(b
)
.
Evaluation Warning : The document was created with Spire.PDF for Python.
TELKOM
NIKA
ISSN:
2302-4
046
Non
uniform
Defe
ct Dete
ction of Cell Ph
one TFT
-L
CD Displ
a
y (Ja
h
angir Alam
S.M.)
6041
Figure 4. Une
v
en and Non
uniformity De
tection Algo
rithm
(a)
(b
)
Figure 5. Intensity Comp
are on the Cell
P
hone TFT
-L
CD
Display Panel (a
) ori
g
in
al, (b)
magnifying
The im
age
h
a
s
been
an
al
yzed
by ‘Hal
con’ an
d
che
c
ked
the statistical
results a
s
sho
w
n
in Table
1. The arro
w indi
cated t
w
o red
squ
a
re
d box
es that h
a
ve an un
even a
nd no
nunifo
rmity
and thei
r devi
a
tion is hi
ghe
r than oth
e
r red sq
ua
r
ed b
o
x. The other red
squa
re
d box on ROI h
a
s
no defe
c
ts. T
he defe
c
ted two ROI devia
tions are 6.24
and 6.27.
Evaluation Warning : The document was created with Spire.PDF for Python.
ISSN: 23
02-4
046
TELKOM
NI
KA
Vol. 12, No. 8, August 2014: 603
6 –
6046
6042
Table 1. Mea
n
and Deviation on the Cel
l
Phone TFT
-LCD Di
splay
Panel
ROI1
ROI2
ROI3
ROI4
ROI5
ROI6
ROI7
ROI8
ROI9
Mean
196.80
201.03
204.33
207.22
206.30
160.17
209.14
223.98
196.76
Deviation
3.86
6.24
4.15
4.17
6.27
4.49
4.01
4.22
4.49
The Fig
u
re
6
sho
w
s the
region
gro
w
in
g si
mul
a
tion
results fo
r u
neven
situati
on an
d
nonu
niform
d
e
fect dete
c
tio
n
. The
captu
r
ed imag
e ha
s been
pro
c
e
s
sed i
n
to medi
an imag
es
as a
mirro
r im
age.
In the
medi
a
n
imag
e, the
r
e is ge
nerat
e
d
a
ci
rcl
e
in
tolera
nce
whi
c
h can
acce
pt
the
ROI pixel v
a
lue. Th
e m
edian im
age
divides i
n
to
seg
m
ent fo
r re
gion
gro
w
ing [1
4]. The
‘regio
n
g
r
owi
n
g’ metho
d
i
s
used to
fin
d
the
defe
c
ts. Th
e tole
ra
nce
of the
‘region
gro
w
in
g
’
is
arou
nd 2 pixels to 3 pixels for this det
ection. In
the 2 pixel to 3
pixel toleran
c
e, the indicat
ed
uneven a
nd n
onunifo
rmity can
be dete
c
t
ed by usi
ng ‘
H
al
con’
software in Fi
gu
re
6(a
)
and
Figu
re
6(b
)
sh
ows that the indicated circ
le of
defects a
r
e
smalle
r tolera
nce. Out of this toleran
c
e
all
defect
s
also can be dete
c
te
d by this method.
(a)
(b)
Figure 6. Def
e
ct Dete
ction
by Regio
n
Grow
in
g at Tole
ran
c
e (a)
3 level, (b) 2 level
(a)
(b)
(c
)
(d)
Figure 7. FFT
Proce
s
sing o
f
Defected TF
T-L
CD
Di
spla
y (a) filtered i
m
age, (b
) co
nvolution FFT
,
(c) image F
F
T, (d) correlat
ion FFT
Evaluation Warning : The document was created with Spire.PDF for Python.
TELKOM
NIKA
ISSN:
2302-4
046
Non
uniform
Defe
ct Dete
ction of Cell Ph
one TFT
-L
CD Displ
a
y (Ja
h
angir Alam
S.M.)
6043
The Figu
re
8 sho
w
s the
r
e is n
o
def
ect of the T
FT-L
CD
disp
lay panel. Here it is
pro
c
e
s
sed th
e imag
e
with
simila
r
way
by ‘Hal
co
n’.
Since i
n
thi
s
cell
pho
ne T
FT-L
C
D di
spl
a
y
panel h
a
s
no
defect
s
or
u
neven
situation an
d non
u
n
iformity in Fi
gure
8, theref
ore, the
r
e a
r
e
no
uneven a
nd n
onunifo
rmity events on the
panel. The
compa
r
ison of Figure 7 and
Figure 8 sh
o
w
s
that wh
ether there
is any
d
e
fect
o
r
n
o
t.
By using
eq
u
a
tion (11
)
, it
also
ha
s
bee
n dete
r
min
e
d
that
the uneven a
nd non
uniformity is absent
.
Depth of def
ects
can b
e
found fro
m
focu
s t
hat is ca
lled Depth F
r
om Focus (DFF) [10].
This i
s
a m
e
thod th
at en
a
b
les the
re
co
nstru
c
tion
of
3D
su
rface in
formation
of
several im
ag
es
taken
at different fo
cu
s di
stan
ce
s bet
ween
came
ra
and the
defe
c
ts. With
dep
th from focus, it
can be
re
con
s
tru
c
ted the surface of a 3D defe
c
t based on the kn
o
w
led
ge that defect regi
on h
a
s
different dist
ances to the
camera and
the ca
mera
has a limited depth of field in Figure
3.
Dep
endin
g
o
n
the
di
stan
ce an
d th
e fo
cus,
the
def
ect
s
regi
on are displ
a
yed mo
re or le
ss sha
r
ply
in the im
age
, i.e., only those
pixel
s
wi
thin the
co
rrect di
stan
ce
to the
came
ra are fo
cu
se
d.
Takin
g
im
age
s
with vari
ou
s p
o
ints on
the d
e
fect
s to
mea
s
u
r
e th
e
dista
n
ces,
e
a
ch
defe
c
t p
o
int
on the d
e
fect regio
n
can
be di
splaye
d sh
arply in
at least
on
e pixel of a
n
image. Su
ch
a
seq
uen
ce
of i
m
age
s i
s
call
ed “fo
c
u
s
sta
c
k” [14]. By d
e
termini
ng in
whi
c
h im
age
an o
b
ject
poi
nt
is in focu
s, i.e., sharply imaged, the d
i
stan
ce
of ea
ch defe
c
t poi
nt on the defect regi
on to the
came
ra
can b
e
cal
c
ulate
d
[14]. This pri
n
ciple i
s
displa
yed in Figure 9.
(a)
(b)
(c
)
(d)
Figure 8. FFT
Proce
s
sing o
f
Non-d
e
fecte
d
TFT-L
C
D Display (a
) filtered ima
ge, (b
) convol
ution
FFT, (c) imag
e FFT, (d)
correlation FF
T
If
the maximum focus pix
e
l length is
∆
and the lengt
h of depth in hidden layer is
∆
,
then the
total 3D defecte
d pixel length is
∆
∆
along wit
h
Z-axis, in th
e
image pla
ne i
f
the defect le
ngth is
∆
to X-axis and the
∆
to the Y-axis
then the defe
c
t depth in
pixel can b
e
defined a
s
:
∆x ∆y d
x
d
y
dz
(
1
4
)
Evaluation Warning : The document was created with Spire.PDF for Python.
ISSN: 23
02-4
046
TELKOM
NI
KA
Vol. 12, No. 8, August 2014: 603
6 –
6046
6044
Figure 9. DF
F on 3D Plan
e of Cell Pho
ne TFT-LCD
Display
The Fig
u
re 1
0
sh
ows the
back
side
of the def
e
c
ts. T
he pe
ak val
u
e of the b
a
ck
side
and
the front
side
are
not same.
The inte
nsity
of ba
ck sid
e
and
front
si
de
are not sam
e
whi
c
h ca
n be
comp
ared wit
h
Figure 5.
Figure 10. 3D Che
cki
ng fro
m
the View of Back Side of
Defect
s
The
defe
c
ts
are
u
s
ually
n
on lin
ear thu
s
the
n
on li
n
ear defe
c
ts i
n
Fig
u
re
3
(
b
)
ca
n
be
expre
s
sed a
s
the autoreg
ressive model
as:
∑
∈
(
1
5
)
Whe
r
e,
is the coeffici
ents param
eters,
is the con
s
ta
nt value,
is the hide
n laye
rs
para
m
eters a
nd
∈
is the white noise of
the linear d
e
fects mo
del
[13, 15]. Th
e back shift
operator can stimulate
the model
by
an
d then the Eq
uation (1
5)
ca
n be written a
s
:
∑
∈
(
1
6
)
The polyno
m
i
a
l rep
r
e
s
entat
ion of the defect model
ca
n be expre
ssed as:
∅
∈
(
1
7
)
The
ba
ck shift ope
rator a
n
d
the
hidd
en
l
a
yer
paramet
ers of th
e
no
n line
a
r eq
uat
ion
can
be sho
c
ked
of the defe
c
ts mod
e
l. The
sou
r
ce
of th
e uneve
n
an
d the non
unif
o
rmity and th
e
depth of defe
c
ts can be fin
a
lize
d
by the followin
g
mod
e
l.
Evaluation Warning : The document was created with Spire.PDF for Python.
TELKOM
NIKA
ISSN:
2302-4
046
Non
uniform
Defe
ct Dete
ction of Cell Ph
one TFT
-L
CD Displ
a
y (Ja
h
angir Alam
S.M.)
6045
∅
∈
(
1
8
)
Finally the po
wer
sp
ectral den
sity of the m
odel is
co
rrelated an
d the varian
ce i
s
relate
d
to the con
s
u
m
ption of the defect
s
. Thu
s
it can be exp
r
esse
d as:
∑
(
1
9
)
,
(
2
0
)
Whe
r
e,
is th
e varian
ce,
is the frequ
en
cy re
sp
on
se,
is the i
ndex t
e
rm. Th
e en
e
r
gy
cal
c
ulatio
n of the uneve
n
a
nd no
nunifo
rmity is sho
w
n
in Table
2. It is note
d
that the en
ergy lev
e
l
of the defects are highe
r than t
he non-d
e
fected ROI in Figure 5. T
here a
r
e two
highe
st two p
eak
energy level for the above non-li
nea
r defect
s
mod
e
l as 0.106
7
and 0.1017
on the defected
regio
n
ROI2
and ROI5 in Figure 5.
Table 2. Energy Dissipate
by the Defect
s
ROI1
ROI2
ROI3
ROI4
ROI5
ROI6
ROI7
ROI8
ROI9
Energ
y
(mV)
0.0343
0.1067
0.0486
0.0487
0.1017
0.0795
0.0869
0.0653
0.0301
6.
Conclu
sion
This
wo
rk re
pre
s
ent
s u
n
e
v
en and
non
uniformity (M
ura
)
dete
c
tio
n
algo
rithm
of cel
l
phon
e TFT
-L
CD di
spl
a
y b
y
using different techni
que
s such
as
3D
FFT, 3D
DCT
, 3D MDCT, a
n
d
NDCT. It is p
r
opo
se
d a no
vel detection
algorith
m
for
the cell p
hon
e TFT-LCD di
splay con
s
isti
ng
image segm
e
n
tation, filtering, and 3
D
chara
c
te
ri
sti
c
s. Image seg
m
entation
an
d filtering are
the
effic
i
ent process
to complete this
work
.
In t
he 3D F
FT processin
g
it can
able
to determi
ne
the
low fre
que
ncy information
after that it compute
the
d
epth an
d cha
r
acte
ri
stics of
defecte
d mo
del
easily. Thi
s
i
n
formatio
n can be i
m
po
rtant for u
nev
en an
d no
nu
niformity det
ection
on m
o
bile
phon
e TFT-L
CD di
spl
a
y for highe
r pe
rfo
r
man
c
e a
nd b
e
tter efficien
cy.
Ackn
o
w
l
e
dg
ements
Wa
rm exp
r
e
s
sion
and
sin
c
ere th
an
ks to
Chin
a
United
Tech. Co.
Ltd., Shenzhen
, Chin
a
esp
e
ci
ally to Mr. David
Hu
ang, Li Mi
ng,
Ou Ji
He
ng
a
nd Gu
o Bo fo
r their
su
ppo
rt to accompli
sh
the experim
e
n
t.
Referen
ces
[1]
Chen SL,
Chou ST
.
T
F
T-L
CD Mura
d
e
fect
detecti
on
usi
n
g
w
a
vel
e
t a
n
d
cosi
ne tr
ansfo
rms.
J. Adv.
Mech. Des. System
. 20
08; 2: 441-
453.
[2]
Song
T
C
, Cho
i
DH, P
a
rk KH
. W
a
velet
bas
ed
imag
e e
n
h
ancem
ent for
defect d
e
tectio
n i
n
thi
n
fi
l
m
transistor li
qui
d
cr
y
s
tal p
ane
l.
Jpn. J. Appl. Physics.
200
6; 4
5
(6A): 501
6–
5
072.
[3]
X
i
n B, Chungang Z, H
an
D. A ne
w
m
u
ra defect inspection
w
a
y
for tft-lcd
using lev
e
l set
method.
IEEE
Sign
al Process
.
Letter.
2009; 16(4): 31
1–
314
.
[4]
F
an SK, Chu
ang YC. Auto
matic detectio
n
of
Mura d
e
fect in T
F
T
-
LCD b
a
sed
o
n
regress
i
o
n
dia
gnostics.
Pattern Recognition Letters.
20
1
0
; 31: 239
7-24
04.
[5]
Noh
CH, L
ee S
L
, Kim D
H
, Ch
ung
CW
, Kim
SH. An effectiv
e an
d
efficient
defect i
n
specti
on s
y
stem f
o
r
T
F
T
-
LCD po
lar
i
sed
films
usi
n
g a
d
a
p
tive t
h
r
e
sho
l
ds
an
d s
hap
e-bas
ed
im
age
a
nal
ys
es.
Internatio
na
l
Journ
a
l of Prod
uction R
e
se
arch.
2010; 4
8
(17)
: 5115-5
1
3
5
.
[6]
Kim SY, Song
YC, Jung
CD, Park KH. Effective defect
d
e
tection
in T
h
in
F
ilm T
r
ansistor Liq
u
id
Cr
y
s
t
a
l
Displ
a
y
imag
e
s
using a
dapti
v
e multi-lev
e
l
defec
t detecti
o
n
and pr
oba
bil
i
t
y
de
nsit
y
fu
n
c
tion.
Optical
Review.
201
1; 18(2): 19
1-1
9
6
.
[7]
Kittler J, Illing
w
orth J.
Minimu
m error thresh
oldi
ng.
Pattern Reco
gniti
on.
1
986; 19: 4
1
-47.
[8]
Yun JP, C
hoi
S, Seo B, Ki
m SW
, Real-ti
m
e vi
si
on-b
a
s
ed d
e
fect ins
p
ection for
hi
gh
-spee
d stee
l
prod
ucts.
Optical Eng
i
ne
eri
ng.
2008; 4
7
(7).
Evaluation Warning : The document was created with Spire.PDF for Python.