TELKOM
NIKA
, Vol.12, No
.4, Dece
mbe
r
2014, pp. 91
1~9
2
0
ISSN: 1693-6
930,
accredited
A
by DIKTI, De
cree No: 58/DIK
T
I/Kep/2013
DOI
:
10.12928/TELKOMNIKA.v12i4.304
911
Re
cei
v
ed Au
gust 27, 20
14
; Revi
sed O
c
t
ober 1
4
, 201
4; Acce
pted
No
vem
ber 5,
2014
High Re
cognition Ratio Image Processing Algorithm of
Micro Electrical Components in Optical Microscope
Wu Jie
1,2
,
Feng Zuren
1
,
Wang L
e
i
3
1
State Ke
y
La
borator
y for Ma
nuf
acturi
ng S
ystems Engin
e
e
r
ing,
Xi'
an Ji
aot
ong U
n
ivers
i
t
y
,
Xi'
a
n 71
00
49, PR Chi
n
a
2
School of Ele
c
tronic Informa
tion Eng
i
n
eeri
n
g, Xi'
a
n T
e
chnolo
g
ica
l
Univ
er
sit
y
, Xi'
a
n
710
0
32, PR Chi
n
a
3
Department o
f
Basis, Xi'
an R
a
il
w
a
y
Vocati
o
nal
& T
e
chnica
l Institute, Xi’
a
n
7100
16, PR C
h
in
a
e-mail: xait
_bs
@16
3
.com
A
b
st
r
a
ct
In order t
o
pro
c
ess smal
l co
mp
on
ents of o
r
igin
al
i
m
ag
e u
nder th
e
micr
o
scope, firstly, this p
a
p
e
r
ado
pts medi
an
filterin
g al
gori
t
hm
to
enh
anc
e targets; a
n
d
the target
s
a
r
e shar
pen
ed
by usi
ng l
a
ter
a
l
inhi
biti
on a
l
g
o
ri
thm, the
edg
e
of targets is o
u
t
line
d
. In
ord
e
r to get rel
i
ab
le t
a
rget re
gio
n
, a
daptiv
e thres
h
old
seg
m
e
n
tatio
n
alg
o
rith
m is
u
s
ed to
extract nee
d targ
et
r
egi
on, a
nd c
h
aracteristics
of target is
use
d
to
distin
guis
h
mul
t
iple
targ
ets. Based
o
n
th
e c
h
ip
resi
stor,
o
ne ti
ny c
o
mpo
nent, i
n
th
e c
a
ptured
i
m
a
ge,
w
e
jud
ge if the c
h
i
p
resistor is
qu
alif
ie
d by c
a
lcu
l
atin
g the p
i
xel
s
area va
lu
es. T
he exp
e
ri
me
n
t
al results sh
o
w
that, the i
m
ag
e proc
essin
g
a
l
gorit
hm
an
d q
ualifi
ed
det
ecti
on a
l
gor
ith
m
is
reaso
nab
le, w
h
ich
provi
des t
h
e
theoretic
al b
a
si
s and i
m
p
l
e
m
e
n
tation
meth
od
of effe
ctive target extraction
and further
qua
lified test.
Ke
y
w
ords
:
ele
c
trical co
mp
on
ents, microsc
o
pe, imag
e proc
essin
g
alg
o
rith
m
1. Introduc
tion
With the developme
n
t of manufa
c
turi
n
g
indus
t
r
y an
d electroni
c tech
nolo
g
y, integration
has b
e
come
a mode
rn
manufa
c
turi
n
g
indu
stry
trend [1]. In orde
r to me
et the need
s of
integratio
n,
small
pa
rts a
nd small chi
p
s and
other com
pon
ents are
eme
r
ge
d a
s
the tim
e
s
requi
re. T
h
e
s
e compo
nent
s h
a
ve be
en
made i
n
min
u
te si
ze i
n
th
e process, a
nd ju
st be
ca
use
these
com
p
o
nents
ca
n be
made m
o
re
a
nd mo
re tiny, integratio
n on
a larg
e scal
e
can
be u
s
ed
in
more fields [
2
]. Those
produces
which
imbark i
n
tegration on a
large scale will
be m
o
re perf
ect
and the
fun
c
tion will
be
mo
re p
o
werful
a
nd have
mo
re
co
ping
strate
gy for differen
t
situation. T
h
e
usa
b
ility of th
ose p
r
od
uces will be stre
ng
thened.
Because these
com
ponents
size is too
sma
ll, tiny misoperati
on in the process
will
emerge qu
est
i
ons in
comp
onent. If these uncertain
compon
ents b
e
use
d
in inte
grated p
r
o
d
u
c
e
s
,
there
will
be
hidde
n d
ang
er in
p
r
od
uce
.In pra
c
tical
engin
eeri
ng
context, qualifi
ed te
st of e
a
c
h
comp
one
nt should b
e
don
e. Based o
n
the fact
that comp
otents
h
a
ve been m
a
de in tiny size,
image analy
s
is
of com
pon
ents need
to have
the aid
of micro
sco
p
e
[3]-[4], microscop
e
is u
s
ed
as an a
u
xiliary tool to acco
mplish a
m
plifi
c
ation of sma
ll compo
nent
s.
In orde
r to improve the p
r
o
duct reli
ability of
small com
pone
nts etc, this pa
per u
s
e
s
chi
p
resi
sto
r
s
of electri
c
al
co
mpone
nts a
s
mea
s
ured
o
b
ject. With t
he aid
of mi
cro
s
copy im
a
g
ing
prin
ciple[5], a
numb
e
r of
chip resi
sto
r
s
are
ca
ptur
ed.
Firstly, the o
r
iginal
imag
e
is p
r
e
-
processed
and targ
ets are en
han
ce
d.The wh
ole
processin
g
pro
c
ed
ure is finished by
using ad
apt
ive
threshold
se
gmentation
a
l
gorithm, etc.
We ad
opt reasona
ble re
cog
n
ition alg
o
rithm ba
se
d
on
cha
r
a
c
teri
stic of chip re
sist
ance to acco
mplish q
ualifi
ed test.
2. The optica
l
imaging principle on mi
crosco
pe an
d image acq
uisition method
In the p
r
od
uct
i
on p
r
o
c
e
ss,
becau
se th
e t
i
ny si
ze
of chi
p
re
si
stors
are do
ne, the
n
t
he
chip
resi
sto
r
s im
a
ge is a
c
q
u
ire
d
with the he
lp of mi
croscope to amplif
y chip re
si
sto
r
s ima
ge, whi
c
h
will facilitate t
he next a
nalysis
of the im
a
ge, pr
ocessin
g
and
ultimat
e
qualifie
d te
st. The im
agi
ng
schemati
c
is
con
s
tituted b
y
area a
rray
came
ra
a
nd
microsco
pe i
s
sh
own a
s
F
i
gure
1. The
tin
y
measured
obj
ects can
be a
m
plified th
rou
gh the
mi
cro
s
cop
e
len
s
, th
en the
amplifi
ed mi
crosco
p
i
c
image
of me
asu
r
ed
obje
c
t
s
can
be o
b
tained. T
he
a
m
plified mi
croscopi
c im
ag
e is
co
ndu
cive to
process the t
a
rget
s.The
mi
croscopi
c im
age is
capt
ured by optical l
ens, the i
m
age will be form
ed
on the imagin
g
surfa
c
e.
Evaluation Warning : The document was created with Spire.PDF for Python.
ISSN: 16
93-6
930
TELKOM
NIKA
Vol. 12, No. 4, Dece
mb
er 201
4: 911
– 920
912
Figure 1. Microscopy imagi
ng prin
cipl
e
As shown in
Figure 2, th
e mea
s
u
r
ed
obje
c
ts a
r
e l
o
cate
d in o
p
t
ical effective
regio
n
,
minute obje
c
ts are amplif
ied by micro
s
copi
c
optica
l
system. Area array ca
mera
captu
r
es
microsco
pic i
m
age by optical len
s
of ca
mera.T
he ca
ptured ima
g
e
enters the co
mputer by image
acq
u
isitio
n ca
rd [6]. In the
pro
c
e
ssi
ng of
the com
pute
r
, usi
ng ima
g
e processing
algorith
m
in t
h
is
pape
r extra
c
t
s
target regio
n
and
u
s
ing
h
i
gh recogni
tio
n
ratio
algo
rit
h
m recogni
ze
s target regio
n
.
The me
asure
d
obje
c
ts
qu
ality will be t
e
sted,
the un
qualitied
o
b
je
cts will
be re
jected and
th
e
qualitied
obj
ects will be
persi
s
ted Image acquisition diagram
under microscope i
s
showed as
Figure 2.
Figure 2. Image acqui
sition
princi
ple un
d
e
r the micro
s
cop
e
3. The high precision image te
st method for
tiny
componen
t
s
under optic
al microsco
pe
3.1 Image processin
g
algorithms for
tin
y
components un
der
microscope
It is
nec
ess
a
ry to abs
trac
t target effec
t
ively
from imag
e duri
ng te
sting the chip resi
stors.
Ho
wever, du
e to these factors effect
which
a
r
e
uneven lighti
ng, diffractio
n
effect, camera
perfo
rman
ce
and
i
n
tern
al and external
noise,
the
fe
ature of com
puter ca
pture
d
imag
e
i
s
lo
w
contrast a
nd
blurs ed
ge [7]
.
These
fe
atu
r
es
are not co
ndu
cive to the
target final t
e
st; therefo
r
e
,
it
is nee
d to research a set of effective pro
c
e
ssi
ng alg
o
ri
thm.
In orde
r to ef
fectively abst
r
act
ea
ch target in
a b
a
tch of chi
p
resi
stors, multipl
e
target
s
image p
r
o
c
e
s
sing al
gorith
m
is studie
d
, the flow ch
art
is sho
w
n a
s
Figure 3.
In view of a
captu
r
ed
ori
g
inal imag
e, d
ue to the
ch
ange
s of b
a
ckgroun
d is
sl
ow a
nd
irre
gula
r
, whi
c
h
will
ob
struct ta
rget te
sting,
so
we
use b
a
ckg
r
o
und
su
ppression
alg
o
rith
m to
achi
eve enh
a
n
cem
ent of target, which i
s
co
ndu
cive to test target; The ed
ge co
ntains imp
o
rt
ant
informatio
n of
the target, a
c
cordi
ng to th
e lateral inhi
b
i
tion pri
n
ci
ple
of huma
n
visual to
sha
r
pe
n
target [8]. Becau
s
e the n
o
i
s
e an
d targ
et edge
sho
w
same characte
ristics, so
target is sharpe
ned
and
noise
si
gnal i
s
al
so
enha
nced at
the
same
time; In o
r
de
r to get
relia
ble target re
gion
Evaluation Warning : The document was created with Spire.PDF for Python.
TELKOM
NIKA
ISSN:
1693-6
930
High
Recogni
tion Ratio Image Processi
ng Algorithm
of Micro Elect
r
ical
Com
ponents .... (Wu Jie)
913
informatio
n, we a
dopt th
e
method
whi
c
h i
s
a
dapt
iv
e thre
sh
old
segmentatio
n
to extract ta
rget
regio
n
. Thi
s
method
ca
n remove mo
st
of the noi
se,
but a
small a
m
ount of n
o
ise and ta
rg
et will
be extra
c
ted;
In orde
r to remove the i
n
fluen
ce
of
noise and di
stingui
sh m
u
ltiple mea
s
ured
targets,
and
we u
s
e th
e m
a
rk cl
uste
ring
algo
rithm,
according
to the
cha
r
a
c
teri
sti
cs
of the ta
rg
et,
to distingui
sh
multiple targ
e
t
s, and fina
lly we test the m
a
rked target quality.
Figure 3. Image pro
c
e
s
sin
g
algorith
m
bl
ock diag
ram
3.2 The algo
rithm of image prepro
ce
ssing and ta
rget en
hanc
ement und
er
the microsc
ope
Usi
ng th
e a
v
erage
value
of a
pixel-d
o
main
as th
e filtering
re
sults is the
simple
st
smoothi
ng filtering
method
, all the coeffi
cient
of filteri
ng template
are 1. F
o
r th
e
3
3
template,
the value
of R ne
ed to
b
e
divided
overall
co
effi
cie
n
t 9. The
sm
oothing filte
r
i
ng ha
s i
nhibi
ting
effect on noi
se, but as t
he re
sult of smoothi
ng filtering, ima
g
e
becom
es fu
zzi
er, it can
be
deeme
d
that
the detail
s
ha
s a dimi
nutio
n. Due to
the
i
r sm
all si
ze,
targets
are e
a
sy to drown
in
backg
rou
nd.
So the smo
o
thing filterin
g is un
suitabl
e. This p
ape
r u
s
e
s
the medi
an filtering to
do
image p
r
ep
ro
ce
ssi
ng.
For a imag
e
y
x
g
,
, the output of 2-D m
edia
n
filtering can be
written:
y
x
g
median
y
x
S
y
x
N
y
x
median
,
,
,
,
(1)
For the me
di
an filtering
which u
s
e
s
a
n
n
template, its o
u
tput sho
u
ld
be greater th
an
or e
qual to th
e
2
1
2
n
piexl value
of the templ
a
te[7]. Norm
a
lly, if the domain of ima
ge i
s
to
o
bright o
r
too
dark an
d
size
of the dom
ai
n is
smal
l
e
r t
han h
a
lf of the template
si
ze, the d
o
ma
in
woul
d be elim
inated.
The me
dian
filtering is
a nonlin
ear
sig
nal
processin
g
tech
nology
based on th
e ord
e
r
s
t
atis
tical theory; it c
an effec
t
ively s
u
ppres
s
th
e no
ise, which i
s
a typical no
nlinea
r spati
a
l
filtering tech
n
o
logy. The m
edian filter
can prot
e
c
t well the sign
al
details while
removing n
o
i
se
[9]. Moreover, the median
filter is
ea
sy to adaptive, so it can furt
h
e
r imp
r
ove filter pe
rform
a
n
c
e.
It can
remov
e
the
sing
ula
r
ity-gray
s in i
m
age, a
nd d
ue to thi
s
fea
t
ure, medi
an
filtering al
gori
t
hm
is usually use
d
for ba
ckg
ro
und supp
re
ss
ion and ima
g
e
depo
sin
g
for dim target.
The imag
e borde
r in
clude
s impo
rtant informat
io
n a
bout image
s,
borde
r is sit
uated the
area
whi
c
h has obvio
us energy differen
c
e [10]. There are some differen
c
e
s
in target
an
d
background
energi
es after
removing noi
s
e, but th
e energy
difference is sm
all, whi
c
h will causes
that sha
r
pe
ni
ng effect is n
o
t good. Thi
s
pape
r us
es
sharp
enin
g
alg
o
rithm ba
se
d
on the pri
n
ci
p
l
e
Evaluation Warning : The document was created with Spire.PDF for Python.
ISSN: 16
93-6
930
TELKOM
NIKA
Vol. 12, No. 4, Dece
mb
er 201
4: 911
– 920
914
of lateral inhi
bition [11], this sh
arp
enin
g
algor
ithm wi
ll be propitio
u
s to se
parated target fro
m
background and ma
kes target cont
our cl
early visibl
e,
whi
c
h
will
enhance the
difference between
the target an
d backg
rou
n
d
,
image qualit
y will be improved.
Loo
king eve
r
y pixel of gray image as a sensor
in
lateral inhibi
tion netwo
rk,
all light
sen
s
e
unit
can be
supp
ressed
by nei
ghbo
ring
nu
mero
us units, so late
ral
inhibition
bet
wee
n
nume
r
ou
s uni
ts can b
e
sh
o
w
n a
s
:
oi
i
n
c
i
i
ic
c
c
y
y
f
y
1
n
c
,...
2
,
1
(2)
For formula
(2),
c
f
is emi
s
si
on freq
uen
cy
of pluse
s
when a lig
ht sense unit is
sho
n
e
alone;
i
y
is th
e emission f
r
equ
en
cy of pluses a
fter
a light sen
s
e
unit has go
t the lateral
inhibition;
ic
is the coefficie
n
t of lateral inhibition;
oi
y
is the threshol
d to generat
e lateral
inhibition [12]
.
Usi
ng the formula (2
) a
s
a
template, we
figure out th
e gray value
s
after every p
i
xel has
been
sup
p
re
ssed by am
bi
ent pixels, th
e cal
c
ulate
d
gray value i
s
lateral in
hibition value of t
h
is
pixel, lateral i
nhibition val
u
es from all
pi
xels of
im
age
co
nstitute
a
new
gray value mat
r
ix, wh
ich
is
calle
d late
ral in
hibition
matrix of ima
ge [13]. Th
e
re
sult of l
a
teral i
nhibitio
n
is
not o
b
vio
u
s
becau
se
of th
e overmu
ch
pixels, in
hibitory a
c
tion i
s
wea
k
wh
en t
he templ
a
te i
s
ove
r
7
7
a
nd
the comp
utational
process
will ta
ke
too l
ong tim
e
.
So we ch
oo
se
a suitabl
e cal
c
u
l
ation
tem
p
lat
e
w
h
ic
h
is
5
5
, that mean
s
we
con
s
id
er th
e
cent
re late
ral inhibitio
n
effect whi
c
h
is from
the
ambient 24 pi
xels. Form
ula
(3) is the
co
mputing form
ula.
2
2
2
2
,
05
.
0
,
,
ab
b
y
a
x
K
D
y
x
P
y
x
H
(3)
Her
e
,
b
y
a
x
P
b
y
a
x
P
K
,
,
0
.
For fo
rmul
a
(3
),
0
,
y
x
D
,
2
-
,
2
-
y
x
P
,
1
-
,
2
-
y
x
P
,…,
2
,
2
y
x
P
are
gray va
lues of
origin
al imag
e.
2
-
,
2
-
0
y
x
P
,
1
-
,
2
-
0
y
x
P
,…,
2
,
2
0
y
x
P
are th
reshold
s
of l
a
teral in
hibiti
on for
every pixel,
y
x
H
,
is the point
y
x
P
,
gray value af
ter the point
y
x
P
,
has got the
lateral
inhibition fro
m
the ambie
n
t 24 pixels,
2
-
,
2
-
y
x
D
,
1
-
,
2
-
y
x
D
,…,
2
,
2
y
x
D
are th
e differen
c
e
of lateral
inhibition
coef
ficient betwee
n
every point and center p
o
i
nt.
3.3 The tar
g
et region de
tection un
der
the microsc
ope
Whe
n
the co
ntrast
i
s
different everywhere
in im
ag
e, if we
only
use a fixed
globa
l
threshold to
segme
n
t whole imag
e, segm
entat
ion
result
s will
be affected
beca
u
se global
threshold
ca
nnot give
co
nsid
eratio
n t
o
contrast
of everywhe
re
in im
age. S
o
we u
s
e
lo
cal
threshold
whi
c
h
correlat
es
coo
r
din
a
te to
segm
ent ima
ge. This
relat
ed with
coo
r
d
i
nate thre
sh
ol
d
is also
called
adaptive thresh
old. Firstl
y, the
image
is decomp
o
sed into a seri
es of subim
a
ges,
these
subim
age
s
can
ov
erlap
o
r
ju
st
adjoin
wi
th
o
t
hers.
If su
bimage i
s
sm
a
ll, the proble
m
s
cau
s
e
d
by shado
w o
r
sp
atial variation
of cont
rast
will be le
ss, and the
n
we
can
cal
c
ulat
e a
threshold
for
one
subi
mag
e
. Segme
n
tation i
s
impl
em
ented
by thre
shol
d
comp
a
r
iso
n
b
e
twe
e
n
every pixel and co
rrespon
ding subima
g
e
[14].
In ord
e
r to
segment ta
rge
t
regio
n
, this
pape
r a
dopt
s adaptive th
resh
old
seg
m
entation
algorith
m
. Firstly, the mean value
and varian
ce
of the image a
r
e
calculated. T
he image i
s
divided i
n
to
subim
age
s o
f
sam
e
size, and
the
pix
e
l me
an val
ues of
every
su
bima
ge
a
r
e
cal
c
ulate
d
re
spe
c
tively. The cal
c
ulatio
n formul
a for m
ean value a
n
d
varian
ce a
r
e sho
w
n:
Evaluation Warning : The document was created with Spire.PDF for Python.
TELKOM
NIKA
ISSN:
1693-6
930
High
Recogni
tion Ratio Image Processi
ng Algorithm
of Micro Elect
r
ical
Com
ponents .... (Wu Jie)
915
M
i
N
i
j
i
g
N
M
11
,
1
(4)
M
i
N
i
j
i
g
N
M
1
2
1
2
,
1
(5)
The threshol
ds of
whol
e i
m
age
and
su
bimage
s
a
r
e
cal
c
ulate
d
. T
he calculatio
n formul
as
are sho
w
n:
m
T
whole
(6)
i
i
i
m
T
(7)
For form
ula
(6) a
nd (7
),
m
is weight coeffici
ent whose ran
ge i
s
from 3 to 10. The
sele
ction of segmentatio
n thre
shol
d
T
ne
ed abid
e
the followin
g
form
ula:
whole
i
i
whole
i
whole
T
T
T
T
T
T
T
(8)
If image
ha
s
many differen
t
regi
on
s of
g
r
ey
val
ue, we
can sel
e
ct a seri
es of thre
shol
ds,
and ea
ch pi
xel is divide
d into the a
ppro
p
ri
ate threshold[1
5
]. Selection
of image thresh
old
segm
entation
can be d
e
fin
ed as:
This
will be assigned to t
he appropriate category
to eac
h pixel. The image after
threshold
seg
m
entation ca
n be define
d
as:
T
y
x
g
T
y
x
g
y
x
g
,
0
,
1
,
(9)
For form
ula (9),
y
x
g
,
is original image. T
he formula
(9) ca
n be e
s
timated an
d
applie
d at an
y pixel positio
n.
3.4 High pre
c
ision identi
fication alg
o
rithm of image under the
microscope
The q
uality o
f
target in
im
age i
s
im
pro
v
ed by
a
se
ri
es
of processing
which
a
r
e targ
et
enha
ncement
and
ed
ge
d
e
tection
etc,
the processin
g
re
sult i
s
co
ndu
cive to
extract ta
rget
a
n
d
sub
s
e
que
nt p
r
ocessin
g
. Capture
d
ima
g
e
co
ntain
s
m
u
ltiple targ
ets, these ta
rget
s a
r
e di
strib
u
ted
in image
with
out rule. In order to corre
c
tly extr
act each target, mult
iple mea
s
u
r
e
d
target
s nee
d
to be
distin
gu
ishe
d. We a
d
opt ma
rk cl
ustering
algo
rithm [16]. Multiple targets
,
acc
o
rding to their
own
cha
r
a
c
te
ristics, are disti
nguished. And finally, these ma
rked
targets are dete
c
ted on q
ualit
y.
Becau
s
e the
magnification
and ca
pture
d
angle of mi
cro
s
cop
e
are uncertain, wh
en chi
p
resi
sto
r
is det
ected o
n
qua
lity, if we onl
y just
calcula
t
e the area v
a
lue of sol
d
e
r
regi
on on
chip
resi
sto
r
to d
e
termin
e wh
ether o
r
not
the chip
re
sisto
r
is q
ual
ified, t
he testing re
sult is not
accurate [17]
-[18]. Thi
s
p
a
per
ado
pts t
h
is m
e
thod,
whe
n
the
pixels
are
a
valu
es
A
1
and
A
2
of
sold
er region
which are lo
cated o
n
ea
ch side of
chip
resi
stor a
r
e
equal, we de
em that this chip
resi
sto
r
is qu
alified. As sh
own in Fig
u
re
4.
Evaluation Warning : The document was created with Spire.PDF for Python.
ISSN: 16
93-6
930
TELKOM
NIKA
Vol. 12, No. 4, Dece
mb
er 201
4: 911
– 920
916
chip resistor
soldering region
1
A
2
A
Figure 4. Chi
p
resi
sto
r
ske
t
ch map
Whe
n
image
is captu
r
e
d
, the captu
r
e
d
cente
r
is
not locate
d in the center of ea
ch chi
p
resi
sto
r
, so t
here
i
s
certai
n calculation
error when
the pixel
s
are
a
value
s
A
1
and
A
2
of sol
der
regio
n
a
r
e
calcul
ated, the
permi
ssible
cal
c
ulatio
n error
ra
nge i
s
%
10
. So wh
en th
e pixel
s
area
value
s
of sold
er re
gi
on is satisfie
d with fo
rmul
a (1
0),
we
d
eem that th
e
chip
resi
stor is
qualified.
2
2
1
%
10
A
A
A
(10)
4. The experi
m
ent and an
aly
s
is
In the p
r
od
uction process,
the tiny chip
re
si
stors h
a
ve
bee
n d
one,
we
nee
d to h
a
ve the
aid of
amplifi
ed fun
c
tion
o
f
microsco
pe
to com
p
lete t
he ima
ge
acquisitio
n
. The
ca
pture
d
im
age
is processse
d
by filtering
and e
dge d
e
tection
etc. T
he pu
rpo
s
e i
s
to elimin
ate noi
se
which i
s
gene
rated
in
pro
c
e
s
s of a
c
quisitio
n
an
d
transmissio
n, to ma
ke th
e
edge
cle
a
r an
d tidy. A se
ri
es
of image processing i
s
essential found
at
ion for the ne
xt target extraction.
One
ori
g
inal
amplified i
m
a
ge i
s
captu
r
e
d
in thi
s
pap
e
r
whi
c
h i
s
sh
o
w
n
as Fig
u
re
5. Fro
m
the amplified
image by mi
cro
s
cop
e
, we
can
see
th
at the iamge
h
a
s lo
w cl
arity
and noi
se, t
he
origin
al imag
e need
s to be
further p
r
o
c
e
s
sed.
Figure 5. Orig
inal image u
n
der mi
cro
s
co
pe
If we can effectively supp
ress backg
ro
u
nd
duri
ng ba
ckgroun
d su
pp
ressio
n stag
e
,
which
will red
u
ce the pro
c
e
s
sing
burd
en. In the backg
ro
u
n
d
supp
re
ssion
stage, not onl
y the interesti
n
g
area
of i
m
ag
e shoul
d b
e
h
i
ghlighted,
bu
t also
the
po
ssible
target a
r
ea
sho
u
le
be
found
out. Th
is
pape
r
co
ntra
sts sm
ooth
filtering
an
d
median
filt
eri
ng,
both
of
t
hem are u
s
e
d
for backg
round
Evaluation Warning : The document was created with Spire.PDF for Python.
TELKOM
NIKA
ISSN:
1693-6
930
High
Recogni
tion Ratio Image Processi
ng Algorithm
of Micro Elect
r
ical
Com
ponents .... (Wu Jie)
917
sup
p
re
ssion,
the re
spe
c
tive filt
ering re
sults are sh
own as Fig
u
re
6.
From the im
age after
smo
o
th
filtering, It is clea
r that image be
com
e
s fuzzie
r,
the visual detail
s
of image a
r
e re
du
ced.
And
from the ima
ge after m
edi
an filtering, T
he contra
st o
f
the image h
a
s imp
r
ove
d
, the media
n
filte
r
has
goo
d abil
i
ty to remove
noise an
d will
not ma
ke th
e detail
s
too f
u
zzy, whi
c
h i
m
prove
s
ima
g
e
visual effect.
(a) Smo
o
thin
g filtered processing ima
g
e
(b) Median filtere
d
pro
c
e
ssi
ng
image
Figure 6. The
filtered pro
c
e
ssi
ng imag
e
After median
filterting ima
ge preprocessing, the
n
the ta
rget
s are
sha
r
pe
ned
by usin
g
lateral in
hibiti
on alg
o
rithm.
Lateral
inhib
i
tion sh
arp
eni
ng effect i
s
shown a
s
Fig
u
re 7
(
a
)
, at the
same time, t
h
is pa
pe
r also use
S
obel
operator to e
x
tract edg
e o
f
the target, and the
resul
t
is
sho
w
n a
s
Fig
u
re 7(b). The
result
s of edge extracti
o
n
show that la
teral inhibitio
n
algorithm
can
extract the ed
ge and the e
x
tracted cont
our is not
def
ormatio
n
, and the extracti
on effect is b
e
tter
than conve
n
tional
So
bel
operator. A
d
j
a
ce
nt targets ca
n b
e
effectively distin
g
u
ish
ed
by ed
ge
extraction.
(a) p
r
o
c
e
ssi
n
g
image of lateral inhi
bition
algorithm (b) Pro
c
e
ssi
ng
image of
Sobel
ope
rato
r
Figure 7. The
edge extra
c
tion pro
c
e
s
sin
g
image
After the ima
ge sharpeni
n
g
, the image
is th
re
sh
old
segmente
d
. T
h
is p
ape
r ad
opts a
n
adaptive
th
resh
old se
gm
entation alg
o
rithm,
th
e
mean
and
v
a
rian
ce
s of
the ima
ge
a
r
e
cal
c
ulate
d
. The imag
e is
divided into
subimag
e
s
of same
si
ze, a
nd the pixel
mean valu
es of
Evaluation Warning : The document was created with Spire.PDF for Python.
ISSN: 16
93-6
930
TELKOM
NIKA
Vol. 12, No. 4, Dece
mb
er 201
4: 911
– 920
918
every subima
ge a
r
e
cal
c
ul
ated respe
c
tively. The thre
shol
ds
of wh
ole ima
ge a
n
d subima
ge
s
are
obtaine
d by
calcul
ation. An
d then
a
c
cording to
the th
reshold
sele
ction rul
e
s,
we
accom
p
lish t
he
whol
e image
threshold
segmentatio
n, and the ta
rget regi
on
s are
segm
ent
ed obviou
s
ly
from
image. The e
x
perime
n
tal result is
sho
w
n as Figu
re 8.
Figure 8. Adaptive thresh
ol
d segm
ent
ati
on algo
rithm
pro
c
e
ssi
ng i
m
age
Figure 9. Qua
lified detectio
n
algorith
m
proce
s
sing ima
g
e
The cle
a
r
i
m
age
i
s
obtain
ed
by above
a
seri
es
of p
r
ocessin
g
. According
to th
eir
own
cha
r
a
c
teri
stics, multipl
e
ta
rgets a
r
e
dist
ingui
s
hed.
We ad
opt im
ag
e cl
uste
ring
method
to la
bel
adja
c
ent
re
gions. Eve
r
y t
a
rget
is extracted.
We
j
udge
if the
chip
resi
stor is qu
alified
by
cal
c
ulatin
g the pixels a
r
e
a
values
A
1
an
d
A
2
.
Finally, some un
qualifi
ed chip resi
st
ors
are rej
e
ct
ed
whi
c
h is
sho
w
n a
s
Figu
re
9.
The ori
g
inal i
m
age have b
een proc
esse
d by above a
serie
s
of co
mpari
s
o
n
alg
o
rithm. It
is
clea
r th
at t
he effe
ct of
median
filtere
d
is b
e
tter th
an
smo
o
thing
filtered
from
pro
c
e
ssi
ng
re
sult.
The
contrast
of the ima
g
e
ha
s imp
r
ov
ed afrer m
e
d
i
an filtering,
At the sam
e
time to re
m
o
ve
noise, the details will no
t be made too fuzzy, which imp
r
ove
s
image visual effect. The
comp
ari
s
o
n
result
s of
edg
e extra
c
tion
show that
late
ral inhi
bition
al
gorithm
can
e
x
tract the
ed
g
e
and the extracted
conto
u
r
is not
defo
r
mation, an
d
the extractio
n
effect is b
e
tter than
So
bel
operator. A
d
j
a
ce
nt targets ca
n
be
effectively dist
ingu
ishe
d by
edg
e extra
c
tion.
Targ
et regio
n
s
have be
en
se
gmented
by
adoptin
g ad
a
p
tive thre
s
hol
d se
gme
n
tation alg
o
rithm.
The
seg
m
en
ted
regio
n
s have
bee
n p
r
o
c
e
s
sed
by hig
h
pre
c
isi
on i
d
e
n
tification
alg
o
rithm,
we
can o
b
tain im
age
whi
c
h only co
ntains q
ualifie
d chip resi
sta
n
ce
s.
Evaluation Warning : The document was created with Spire.PDF for Python.
TELKOM
NIKA
ISSN:
1693-6
930
High
Recogni
tion Ratio Image Processi
ng Algorithm
of Micro Elect
r
ical
Com
ponents .... (Wu Jie)
919
The exp
e
rim
ental result shows th
e e
d
ge of
ta
rget
s have
bee
n
clea
r by
usi
n
g imag
e
pro
c
e
ssi
ng
al
gorithm
from
this p
ape
r
an
d the
imag
e
quality ha
s
b
een i
m
proved
. We
have
m
ade
several exp
e
riments on
10
0 chip
re
sista
n
ce
s
by ad
op
ting imag
e p
r
oce
s
sing
alg
o
rithm
and
hi
gh
pre
c
isi
on i
d
e
n
tification
alg
o
rithm i
n
thi
s
pap
er.
Hom
ogen
eity of e
x
perime
n
tal result i
s
nice
and
inaccu
rate re
cog
n
ition ratio is lo
w. The
s
e alg
o
rith
m
s
in this pap
er effectively improve
s
a
c
curate
recognitio
n
ra
tio, which p
r
o
v
ides reli
able
method for q
ualified test.
5. Conclusio
n
The ca
pture
d
imag
e
from microsco
pe e
x
hibits
lo
w co
ntrast,
poo
r
cl
arity. In thi
s
p
aper,
a
seri
es of ima
ge p
r
o
c
e
ssi
n
g
algo
rithm
s
are
re
se
a
r
ch
ed, the
captu
r
ed im
age
fro
m
micro
s
cop
e
is
pro
c
e
s
sed
by pre
p
ro
ce
ssin
g algo
rithm
a
nd ad
apt
ive
t
h
re
shol
d seg
m
entation alg
o
rithm etc.
T
he
sold
er
regi
on
s o
n
ea
ch
si
de of
chip
re
sisto
r
a
r
e
effectively extra
c
ted, a
nd th
e
target
qualifi
ed
detectio
n
is
a
c
compli
she
d
by usin
g qual
ified dete
c
tio
n
algo
rithm.
The expe
rime
ntal re
sults
show
that the alg
o
r
ithm i
s
rea
s
onabl
e an
d f
easi
b
le.
Go
o
d
expe
riment
al re
sult i
s
achi
eved. Th
e
probl
em existing microscopic im
age
s is solve
d
. The alg
o
rith
m can b
e
a
pplied in im
age
processi
ng and other
simil
a
r probl
em
s, whi
c
h will improve the ac
curacy of simil
a
r products.
Ackn
o
w
l
e
dg
ements
This work was
partially supported
by
Ph.
D. Pro
g
ram
s
F
oun
dation of Mi
nistry of
Educatio
n of
Chin
a (2
0
1002
0111
003
1), Natio
nal
Natural Science Fou
n
d
a
tion of Chi
na
(611
051
26 &
6087
5043
), a
nd Spe
c
ial
F
ound
ation
of
Presi
dent of Xi’an
Te
ch
nol
ogical University
(XGYXJ
J05
2
4
).
Referen
ces
[1]
W
angh
on
gmei.
Researc
h
Progress o
n
Image Match
i
n
g
.
Computer e
n
g
in
eeri
ng a
n
d
Applic
ations
.
200
4; 19(2): 42
-44.
[2]
HO. Lim, W
.
Z
hang,
LM. K
oh. Autom
a
ted
visua
l
i
n
sp
ection f
o
r IC
w
i
r
e
-bo
nd
usin
g
auto-focus
i
n
g
techni
qu
e.
IEEE/CHMT Electronics Manufact
u
ring Technology Sym
p
osium
. 1993; 12(
7): 31-36.
[3]
Hans
han L
i
. Researc
h
on p
h
o
toel
ectronics
prop
erties
of arra
y
emittin
g
dio
de a
nd its light e
ner
g
y
distrib
u
tion i
n
d
e
tection scre
e
n
.
Optik
. 2014; 125(3): 10
96-
11
00.
[4]
Bailif
en,
Xu
yu
xi
an, Y
u
shu
i
.
Rese
arches
o
n
the
microsc
ope
aut
o focu
s metho
d
b
a
s
ed
on Ima
g
e
process
i
ng.
Ch
ines
e Journ
a
l o
f
scientific instr
u
ment
. 199
9; 2
0
(6): 612-
61
4.
[5]
Peters W
H
, Rans
on W
F
. Digita
l
im
agi
ng tech
ni
ques
ine
x
perim
ent
al stress a
n
a
l
y
s
is.
Optical
Engi
neer
in
g.
1982; 21(
3): 427
-432.
[6]
Pratt W
K
.·Correlati
on T
e
chniq
ues of Image R
egistrati
on .
IEEE Traris AES
. 1974; 1
0
(3): 3
53-3
57.
[7]
Hans
han
Li, Z
h
i
y
o
ng L
e
i. Projectil
e
T
w
o-d
i
m
ensi
ona
l Co
ordi
nates Me
a
s
ureme
n
t Method Bas
ed o
n
Optical F
i
b
e
r C
odi
ng F
i
re
and
its Coord
i
n
a
tes
Distributi
ng Pr
oba
bil
i
t
y
.
Me
a
s
ure
m
e
n
t Scie
nce Rev
i
ew
.
201
3; 10(1): 34
-38.
[8]
Sunfen
g, Sun
y
ao. Image s
h
a
r
pen
ing m
e
tho
d
an
d
its ap
pl
icatio
n bas
ed
on t
he
princ
i
pl
e of later
a
l
inhi
biti
on.
T
e
ch
nol
ogy a
nd ap
plicati
on of a
u
tomatio
n
in He
il
ong
jia
ng.
1
9
9
1
; 10(3): 50-5
3
.
[9]
Che
nbi
ng
qua
n,
Liuh
ong
li. Cu
rrent situatio
n and d
e
vel
opm
ent directi
on o
f
digital ima
g
e
processin
g
technology
.
Jo
urna
l of Jisho
u
Univers
i
ty: Natural Sci
ence E
d
itio
n.
200
9; 30(1): 63-7
0
.
[10]
Hans
han Li,
Z
h
i
y
on
g
Lei. Stud
y
and an
al
ysis
o
n
a n
e
w
optic
al d
e
t
ection des
ig
n
method for
photo
e
l
e
ctric d
e
tection targ
et.
Sensor Rev
i
e
w
. 2013; 33(4):
315-3
22.
[11]
Gaoai
hu
a, Z
huchu
an
gui. Si
mulati
on st
u
d
i
e
s of comp
ou
nd e
y
e
ima
g
i
ng e
x
perim
ent
.
Journa
l o
f
Northw
estern
Univers
i
ty (nat
ural sci
ence e
d
i
tion).
19
98; 28
(2): 117-1
20.
[12]
Shum H Y, Szeliski R. S
y
st
e
m
s and e
x
p
e
ri
ment pap
er co
nstruction of p
anor
amic ima
g
e mosaics
w
i
th
glo
bal a
nd l
o
ca
l alig
nme
n
t.
I. J. of Comp
uter Visio
n
. 200
0; 36(2): 101-
13
0.
[13]
Hans
han
Li, Z
h
i
y
o
ng L
e
i, Z
e
min W
ang, Ju
ncha
i
Gao. Re
search o
n
Obj
e
ction Informat
i
on E
x
tractio
n
Arithmetic i
n
Photo-e
l
ectri
c
Detectio
n
T
a
rget Base on W
a
ve
let
Anal
ys
is Me
thod.
Pr
z
egl
ad
Elektrotech
n
ic
zny
. 2012; 8
8
(9)
:
157-16
1.
[14]
Yong
xin Z
han
g
,
Li Chen, Z
h
ih
ua Z
hao, Jia
n
Jia.
Multi-focus
Image F
u
sion
w
i
t
h
Spars
e
F
eature Base
d
Pulse C
o
u
p
le
d Neur
al N
e
tw
o
r
k.
T
E
LKOMNIKA T
e
lec
o
mmunic
a
tio
n
Computi
ng
Electron
ics an
d
Contro
l.
201
4; 12(2): 35
7-3
6
6
.
Evaluation Warning : The document was created with Spire.PDF for Python.
ISSN: 16
93-6
930
TELKOM
NIKA
Vol. 12, No. 4, Dece
mb
er 201
4: 911
– 920
920
[15]
Hans
han
Li, Z
h
i
y
o
ng L
e
i, Z
e
min W
a
n
g
, Ju
ncha
i G
ao. R
e
search Ob
ject
Photoe
lectric
Char
acteristic
and Fire
Co
ord
i
nate D
i
strib
u
ti
ng Pro
bab
ilit
y i
n
Across Scre
en S
y
stem.
Jo
urna
l of Na
noe
lectron
i
cs an
d
Optoelectro
n
ic
s
. 2012; 7(2): 1
99-2
03.
[16]
W
angfen
g, Li
yanj
un. A n
e
w
a
nd Better
W
a
y
of Extra
e
tin
g
D
i
gital
Imag
e Ed
ge.
Jour
na
l of
Northw
ester
n
Polytech
nica
l Univers
i
-Ty.
19
98; 16(4): 5
36-
539.
[17]
Hans
han L
i
, Junch
a
i Gao. Rese
arch on t
a
rget ph
otoe
le
ctricit
y
track method a
nd i
m
prove
d
imag
e
process
i
ng ar
ithmetic in d
y
n
a
m
ic targets det
ection s
y
stem.
Optik
. 2014; 12
5(14): 35
90- 3
595.
[18]
Z
hang,
Xi
up
in
g. T
he re
sear
ch of gr
anu
la
r computi
ng
a
ppli
ed
in
ima
ge mos
a
ic.
TEL
K
OMNIKA
T
e
leco
mmunic
a
tion C
o
mputi
n
g Electron
ics a
nd Co
ntrol.
20
13; 11(3): 5
37-
546.
Evaluation Warning : The document was created with Spire.PDF for Python.