TELKOM
NIKA
, Vol.12, No
.3, Septembe
r 2014, pp. 6
83~688
ISSN: 1693-6
930,
accredited
A
by DIKTI, De
cree No: 58/DIK
T
I/Kep/2013
DOI
:
10.12928/TELKOMNIKA.v12i3.108
683
Re
cei
v
ed
Jan
uary 29, 201
4
;
Revi
sed Ma
y 26, 201
4; Accepted
Jun
e
10, 2014
Resear
ch on Keyhole Diameter’s Vision
Measur
e
ment Based on Parallel Technology
Zhang Chua
n-chu
a
n, Ze
ng Zhi-qiang
, Wang Jun
-
y
u
an, Du We
n-hua
North Univ
ersi
t
y
of Chi
na
Schoo
l of Mechan
ical a
nd Po
w
e
r Eng
i
ne
eri
n
g
Xu
e
y
u
an R
d
.3,
T
a
i
y
u
an, 03
00
51 Ch
ina
e-mail: zhcc
22
060
06
0@1
63.c
o
m
A
b
st
r
a
ct
A keyh
ole
di
a
m
eter
of the c
a
rtridg
e
’
s
me
a
s
ure
m
e
n
t bas
e
d
on
mach
ine
vision
is a
par
t of the
cartridg
e
’
s g
e
o
metry measu
r
ement syste
m
, accor
d
in
g
to the system requir
e
me
nts, to complete
the
me
asur
e
m
ent
w
i
thin 5 seco
nds.
A Image
Collecti
on Sy
stem w
a
s
con
s
tructed usin
g
comp
uter, CCD
camera,
LED s
ource,
meanw
hile
a
meas
ure
m
e
n
t syste
m
w
a
s co
mpil
ed
by
C#
on
VS20
1
0 pl
atfor
m
b
a
s
e
d
on
mac
h
in
e vi
sion. Use t
he
Otsu algor
ith
m
to extracts
the keyhol
e
’
s
edg
e an
d ne
ar the
pixels
in or
der
to
reduc
e the c
o
mp
utatio
nal
C
anny
op
er
ator,
and
use p
a
ra
l
l
el co
mputi
ng i
n
the C
anny
o
perator to
i
m
pr
ove
computi
ng sp
e
ed p
u
rpos
es. Use Quer
yPerf
o
rmanc
eCo
unt
er timer for eac
h modu
le ti
mi
n
g
Can
n
y op
era
t
or,
Can
n
y op
erato
r
impr
ove
d
co
mp
utatio
n time
is reduc
ed fr
om th
e ori
g
in
al 6s to n
earl
y
a hun
dre
d
ms
improve
d
. Mee
t
the time req
u
ire
m
e
n
ts of c
a
rtri
dg
e ge
o
m
etry me
asur
e
m
e
n
t syste
m
, and
other
mac
h
in
e
vision i
n
w
h
ich
the proj
ect can
be w
i
dely use
d
.
Ke
y
w
ords
:
ma
chin
e visio
n
; trigger key
hol
e d
e
tection; e
dge
detectio
n
; sub-
pixel; thres
hol
d
calcul
ation
1.
Introduc
tion
The ann
ual output of so
me type cart
ridge
rea
c
he
d more than
one million, but the
diamete
r
of t
h
is type
ca
rtridge
keyhol
e
is o
n
ly
1m
m, Produ
ctio
n processe
s
can’t
com
p
let
e
ly
guarantee
th
e quality of t
he
keyhole
machi
n
ing.
when the
keyh
ole tole
ran
c
e
,
likely to
ca
use
cart
ridge launch fail
ure, so the military require
the
production units to
accurately measure the
keyhol
e dia
m
eter
for ea
ch
ca
rtridg
e to
ensure
the
o
v
erall q
uality of the cartri
d
ge. Acco
rdin
g to
the ch
ara
c
te
ristics of the
measurement
obje
c
t,
the propo
se
d met
hod u
s
ing th
e machine vi
sion
measurement
. Firstly sh
oo
ting Partial i
m
age of th
e
keyhol
e an
d
pro
c
e
ssi
ng th
is imag
e, the
n
Cal
c
ulating
t
he di
amete
r
of the
keyhol
e, finally
co
mpari
ng
with
the tole
ran
c
e of the
dia
m
eter
values a
nd d
e
termini
ng whether
com
p
li
ance with the
requi
reme
nts.
In orde
r to improve the
accuracy an
d pr
e
c
isi
on
of the measurem
ent, usi
ng ca
nny
operator
edg
e dete
c
tion
algorith
m
to detect
cart
rid
ge keyhole
edge
point a
r
e u
s
ed
wh
e
n
pro
c
e
ssi
ng t
h
is i
m
age
[1
]. Canny
op
erato
r
at
th
e
sa
me time
improve
the
accuracy
of
the
comp
uter
ha
s in
cre
a
sed
the amou
nt of comp
ut
ation. Un
der th
e co
ndition
s of the exist
i
ng
laboratory e
q
u
ipment, run
n
ing time
of
Gau
ssi
an f
ilter i
s
67.2
8
ms, ed
ge
co
ntour
extracti
on is
5719.2
3
ms i
n
Ca
nny o
p
e
r
ator. So
the
whol
e time
of Can
n
y op
erato
r
is 57
8
6
.61ms,
ca
rtridge
measurement
syste
m
can’t meet
t
he tim
e
req
u
irement
s. Therefore,
in orde
r to im
prove
dete
c
tion
efficien
cy, a
method
of u
s
i
ng Ot
su
op
erator
edg
e co
arse po
sitioni
ng
to
re
du
ce the
comp
utation
[2] and
parall
e
l processin
g
tech
nolo
g
y i
s
p
r
op
osed
t
o
ma
ke su
re
the system
perfo
rman
ce
to
meet the actu
al requi
rem
e
n
t
s.
2. The main parame
ters
of the
CCD c
a
mera and c
o
mputer
The m
a
in fl
ows of m
e
asu
r
ing
sy
stem a
r
e im
a
ge a
c
q
u
isitio
n, image
a
nalysi
s
,
measurement
, and output
s the result. To achi
eve
high-sp
eed
measurement
of the keyh
ole
diamete
r
m
u
st be
in
full
knowl
edge
of
the si
ze
of th
e CCD
cam
e
ra to
captu
r
e
imag
es of t
he
situation fully hard
w
a
r
e p
e
rforman
c
e of your comp
uter.
The hardware consist
s
of an optical
illu
mination system, CCD cam
e
ra, computer
hardware and related
auxiliary equi
pment
el
em
ents. A
c
cording to the
keyhol
e imaging
requi
rem
ents and th
e de
si
gn of the
me
asu
r
ing
sy
ste
m
, Adjust the
colle
cted
im
age, the im
a
ge
pixel si
ze
of
512
× 512,
while the
cam
e
ra'
s
S
N
R i
s
40dB.
Comp
uter i
s
usi
ng
AMD A8
-45
0
0
M
Evaluation Warning : The document was created with Spire.PDF for Python.
ISSN: 16
93-6
9
30
TELKOM
NIKA
Vol. 12, No. 3, September 20
14: 68
3 – 688
684
quad
-core p
r
oce
s
sor,
Wi
n
dows
7
o
p
e
r
ating
system.
Image
Mea
s
urem
ent Syst
em p
r
og
ram
is
written
i
n
C #
un
der
the VS2010 platform.
Imag
e acq
u
isitio
n
sy
stem schem
a
t
ics and
keyh
ole
origin
al imag
e are sho
w
n i
n
Figure 1 an
d Figure 2.
Figure 1. image a
c
qui
sitio
n
system
sch
e
matics
Figu
re
2. keyhole o
r
i
g
inal imag
e
3.
C# Parallel Computing
Emerge
nce o
f
parallel
com
puting i
s
th
e i
nevit
able
re
sult of the
dev
elopme
n
t of
compute
r
s
c
i
e
n
c
e
, is
th
e
s
a
me pr
oc
es
s us
in
g a va
r
i
e
t
y
of
computing
resource
s to
so
lve com
putin
g
probl
em
s.
Un
der
the pre
m
i
s
e of simulta
neou
s,
the
p
r
oce
s
s
can
be
cal
c
ul
ated i
n
to sm
all p
a
rts, to
solve comp
utational p
r
obl
ems in
Con
c
urrent way.
This i
s
an ef
fective mean
s to improve
the
comp
uter sy
stem calculate
s
the
spe
ed
and
processi
ng p
o
wer.
Th
e ba
si
c id
ea
i
s
to
u
s
e
multi
p
le
pro
c
e
s
sors t
o
solve the
same
proble
m
coll
abo
rati
ve, that the
probl
em i
s
b
r
oken
do
wn
into
several pa
rts,
each p
a
rt by a sep
a
rate p
r
oce
s
sor to pa
rallel comp
uting.
With the ra
pi
d developm
e
n
t of comput
er ha
rd
ware
, image pro
c
essing o
n
m
u
lticore
pro
c
e
s
sors h
a
s al
rea
d
y become a f
a
ct of
existi
ng, but the
traditional im
age p
r
o
c
e
ssi
ng
prog
ram
m
ing
mode m
u
st
b
e
co
mpatible
with the n
e
w
hard
w
a
r
e
env
ironm
ent to
make
the ima
g
e
pro
c
e
ssi
ng
speed to
achie
v
e the be
st a
pplication results [3]. C# i
s
integrated T
P
L (Ta
s
k Parallel
Library) and PLINQ (Paral
lel
LINQ
), parallelized appli
c
ations
can
be
achieved, whi
c
h will
greatly
enha
nce the spe
ed of the appli
c
ation is
runni
ng.
Static cla
s
s
named
syste
m
.
Threading.
Parallel
provides thre
e i
m
porta
nt wa
ys
Fo
r
,
Fore
ach
and
I
n
vo
ke
lo
cated two n
a
m
esp
a
ces
which a
r
e n
a
m
ed
System
.Threa
ding
an
d
System
.Thre
ading.Ta
sks
.
3.1 Guas
s filter par
a
llel computing
Gau
ssi
an filter is a linea
r filtering, the value
of each pixel, both by
itself and the other
pixel values
throu
gh the n
e
ighb
orh
ood
to get a
weig
hted avera
g
e
[4]. The main feature is t
he
Gaussi
an function i
s
still a Gaussian function
after F
ourier transform, so th
e
application of fast
Fouri
e
r
t
r
an
sf
orm ca
n
p
u
t convol
ution o
f
airspa
ce
t
r
a
n
sformed
int
o
the
pro
d
u
c
t operation
s
o
f
freque
ncy do
main,
which greatly red
u
ces
th
e comp
u
t
ation time. In
pra
c
tical a
p
p
lication, th
e t
w
o-
dimen
s
ion
a
l
Gau
ssi
an fu
n
c
tion
G(x, y)
is d
e
compo
s
ed into
on
e-d
i
mensi
onal
G
aussia
n
fun
c
t
i
on
G(x)a
nd G
(
y)i
n
the x directi
on and y dire
ction for ima
g
e
filtering:
)
2
ex
p(
2
1
)
,
(
2
2
2
y
x
y
x
G
(1)
22
()
1
2
e
x
p
(
2
)
Gx
x
(2)
22
(
)
1
2
e
xp(
2
)
Gy
y
(3)
Image acqui
sition
H
o
od
Tested part
s
Bas
e
Adjustabl
e ba
ckli
ght
Evaluation Warning : The document was created with Spire.PDF for Python.
TELKOM
NIKA
ISSN:
1693-6
9
30
Re
sea
r
ch on
Keyhol
e Dia
m
ater's Visio
n
Measurem
ent Based on
…. (Zhan
g
Chuan
-
Chua
n)
685
In actual prog
rammin
g
, method of
Forea
c
h
an
d metho
d
of
Partitioner.Create
whi
c
h are in stati
c
cla
ss n
a
med
System
.Thre
ading.Pa
rallel
is com
b
ined
for parallel co
mputing of al
gorithm
[5].
Whe
n
the
ho
rizo
ntal filteri
ng, two
-
dim
e
nsio
nal
array
of ima
g
e
s
i
s
autom
aticall
y
divided
into several
small two
-
dime
nsi
onal
array
by Parallel.ForEach (Pa
r
titioner.
C
reate
(0,
BmpData.
Hei
ght), (H) => {
}
).The lo
op of
each
sma
ll t
w
o-dimensional array in
h
e
ight directio
n is
for (j = H.Item1; j <H.Item2;
j + +), in width direc
t
ion is
for (i
= 0; i
<
B
mpData.
W
idt
h
; i + +).
When
the vertical filtering, two-di
mensi
onal
array of
image
s
is auto
m
atica
lly divided int
o
seve
ral
sm
all
two-di
men
s
io
nal array by
Parallel.Fo
rE
ach (Pa
r
ti
tion
er.Create (0, BmpData. Wi
dth), (H) => {}).
The loop of eac
h
s
m
all two-dimens
ional array in
widt
h direc
t
ion is
for (i
=
W.Item1; i<
W.Item2;
i++), in
heig
h
t
dire
ction
is for
(j= 0; j
<
B
m
pData.
Hei
ght; j++).
The
filtering
state
m
ent of
a p
o
i
n
t
on the image
is temp +
= grayValues
[i
* length1
+ rem
]
* filter [k + radiu
s
].
3.2 Parallel computing of
edge con
t
ou
r extra
ction
The e
dge
co
n
t
our extra
c
tio
n
procedu
re i
s
a
se
rial o
p
e
r
ation
before
multi-core
co
mputers
appe
are
d
. Wi
th the advent
of multi-core
techn
o
logy, the metho
d
of
Parallel. Invoke
i
n
C #
pu
t a
picture into
fo
ur p
a
rt
s to th
e ed
ge
conto
u
r extractio
n
[6]. Method
of
P
a
rall
el.
I
n
vo
ke
i
s
th
e ea
si
est
way to parall
e
lize the
seri
al cod
e
.
Parallel.Invoke
pu
t a cartrid
ge
keyhol
e ca
pture o
r
igin
al imag
e
is divided into
four se
ction
s
for paral
l
e
l proce
s
sing a
s
shown in Figu
re 3.
Figure 3. parallel pro
c
e
s
si
ng image
seg
m
entation
4. Threshold
segmentati
on and edg
e
detec
t
ion
For
a
pixel
si
ze
of 5
12
×
5
12 im
age
fire
-hole
of th
e
cartrid
ge,
whi
c
h is only
a fe
w p
o
ints
arou
nd the
e
dge p
o
int an
d the ed
ge contour
of edg
e point
s
nee
d to be p
r
o
c
e
s
sed. Setting
two
con
c
e
n
tric ci
rcle
s with
the
outline,
on
e slightly large
r
t
han the
cont
our
edg
es,
sli
g
htly more
th
an
one
contou
r
edge. As
sh
o
w
n in Fi
gure
4, only the pi
xels within th
e red
circle
o
n
ly have the
edge
conto
u
r featu
r
e is the obj
e
c
t sho
u
ld be
extracted.
Figure 4. position of the edge to be pro
c
essed
Evaluation Warning : The document was created with Spire.PDF for Python.
ISSN: 16
93-6
9
30
TELKOM
NIKA
Vol. 12, No. 3, September 20
14: 68
3 – 688
686
4.1 Impro
v
e
d Otsu thres
hold calculation
Adding Ot
su
algorith
m
after Ga
ussian
filter
in Cann
y operato
r
ca
n achi
eve au
tomatic
threshold
cal
c
ulatio
n, this method is
rel
a
tively simple
and fast[7,8].
The
ba
sic ca
lculatio
n met
hod i
s
: the
g
r
ay rang
e [
L
m
in, Lm
ax
] o
f
the ima
g
e,
the g
r
ay
value of the pixel is divid
ed into two categori
e
s
C
1
and
C2
in accordan
ce
with the thre
sh
old
value
T
,
C1
i
s
con
s
i
s
ted
b
y
the pixel
which
the
gray
ran
g
e i
s
[
Lm
in, T
],
C2
i
s
con
s
i
s
ted by the
pixel which th
e gray rang
e
is [
T +1, Lm
ax
], by the fo
rmula
(4) cl
asse
s
squ
a
re
e
r
ror bet
wee
n
t
w
o
types of cal
c
u
l
ations.
)]
(
)
(
)[
(
)
(
2
1
2
1
2
t
t
t
t
(4)
In the formula
(4),
1
(
t
) is the
numbe
r of pixels of
C
1
,
2
(
t
) is the num
b
er of pixels of
C
1
,
1
(
t
)
is the averag
e gray va
lue
of the pixels
of
C
1
，
2
(
t
)
is the avera
ge
gray value of
the pixels of
C
2
. Cal
c
ulati
ng the
opti
m
al thre
sh
ol
d at the
sa
me time o
b
taining
an im
age of th
e
same
maximum pixel value.
Achieving Ot
su algo
rithm
in C #, fin
d
ing the mi
nimum pixel
values
L
mi
n
and the
maximum pix
e
l values
L
ma
x
by cycle
com
pari
s
on
, a
nd
then
cal
c
ulati
ng the
optim
a
l thre
shol
d
T
of
the image. th
e gray’
s
ra
ng
e of the edg
e
point and
ne
ar is i
n
the in
terval [
T, L
ma
x
], at the sam
e
time recordi
n
g positio
n of these pixels.
Extracting the
image e
dge
points
and te
n point
s gr
adi
e
nt dire
ction
near th
e pixel
values
of the pixel
s
t
h
rou
gh
a p
r
o
g
ram. A
s
sh
o
w
n i
n
Ta
ble
1
,
the g
r
adie
n
t is th
e bi
gge
s
t whe
n
the
g
r
ay
value of the pixel is 204,
so this p
o
int is dete
r
mine
d
as an ed
ge
point. Optima
l threshold of
the
image is 1
29.
As sho
w
n in
Figure 5, the
intermediat
e
circula
r
part
is extracte
d e
d
ge point
s and
near.
Table 1. Gray and gra
d
ient
near the e
d
g
e
of
a ten-poi
n
t and point-p
ixel gradie
n
t dire
ction
Gra
y
55
64
73
91
109
154 204 252
254 254
Gradient
4
9
9
18 18
45 50 48
2
0
Figure 5. improved Ot
su o
perato
r
Figu
re
4.2 Edge Extraction
Origin
al im
ag
e contain
s
a t
o
tal of 2
62,14
4 pi
xel
s
, the i
m
prove
d
Ot
su op
erato
r
extracting
the numbe
r
of pixels and
near a
r
e ab
out 2560.
Used by the edge contou
r extraction p
o
i
nt
decrea
s
e
d
from 262,14
4 to 2,560. Thi
s
can g
r
eatly
increa
se the e
dge contou
r
extraction
sp
eed.
Therefore,
when u
s
ing th
e
C # lan
gua
g
e
, don’t ne
ed
to traverse e
a
ch
pixel of the whole im
a
ge,
only need to
cal
c
ulate
Otsu o
perator
extracted
pi
xels. Figu
re
6
is imp
r
oved
arou
nd the
e
dge
conto
u
r extra
c
tion compa
r
i
s
on
cha
r
t, it can be
seen from Figure extractio
n
is no
different.
Evaluation Warning : The document was created with Spire.PDF for Python.
TELKOM
NIKA
ISSN:
1693-6
9
30
Re
sea
r
ch on
Keyhol
e Dia
m
ater's Visio
n
Measurem
ent Based on
…. (Zhan
g
Chuan
-
Chua
n)
687
(a) T
he effect
of no improv
ed Ca
nny
(b) The effect of improved Ca
nny
Figure 6. improved Canny
operator b
e
fo
re and afte
r the treatme
nt effect com
p
arison
cha
r
t
5. Measurem
e
nt res
u
lts a
nd analy
s
is
Usi
ng a
timer
whi
c
h
the lo
west ide
n
tified is 1m
s in
C#
name
d
QueryPe
r
form
anceCount
er
re
co
rd the
time of improved Can
n
y operator befo
r
e and after
each
module. And
for the evalua
tion of the performa
n
ce
of parall
e
l com
p
uting, includi
n
g
spe
e
dup a
n
d
effic
i
enc
y
.
Speedu
p is
defined a
s
: if the executio
n time
of an algorithm e
x
ecution time
of th
e
optimal serial
algorithm i
s
T
s
, parallel al
gorithm
s is
T
p
, the ratio of parallel
sp
e
edup al
gorith
m
S
=
T
s
/
T
p
[9].
Parallel effici
ency is defin
ed as:
If a speed
up of parallel al
gorit
hm is
S
, parallel
executio
n thread
s of node
s is
N
, the parallel efficiency is
S
p
=
S/N
[10].
(a)
Re
cord the
ti
me of t
w
o
kin
d
s
of G
a
u
ssi
an filt
er op
erations which
use
technol
o
g
y of pa
rall
el
comp
uting an
d don’t use techn
o
logy of p
a
rallel
co
m
p
u
t
ing, the resul
t
s are
sho
w
n
in Table 2.
Table 2 the time of Gau
ssi
an f
ilter impro
v
ed and unim
p
roved
(ms)
Gaussian filter
unimproved
improved
time 67.28
16.45
(b)
Re
cord th
e ti
me of th
re
e
kinds of
edge
conto
u
r extra
c
tion
algo
rith
m, incl
uding
unimp
r
oved
algorith
m
, th
reshold
seg
m
entation
al
gorithm
only
and
the
al
gorithm
s th
a
t
thre
shol
d
segm
entation
and p
a
rallel
comp
uting a
r
e usi
ng at th
e sa
me time.
The
re
sults
are
sh
own
in
Table 3.
Table 3 the time of edge e
x
traction imp
r
oved and u
n
i
m
prove
d
(m
s)
edge extraction
unimprove
d
threshold
segme
ntation
threshold segme
ntation
and parallel computing
time 5719.23
645.26
145.47
By experimen
tal data in Table 2 and Ta
b
l
e 3 Comp
ari
s
on follo
wing
con
c
lu
sio
n
s:
(a)
The time of Gau
ssi
an filter red
u
ce from
67.28ms to
16.45m
s
by improve
d
. By
analyzi
ng the
time of techn
o
logy whi
c
h
only take th
re
shol
d se
gme
n
tation and t
he time of techn
o
logy bot
h
take threshol
d se
gmentati
on and
parall
e
l com
puti
ng,
the time of parallel te
ch
no
logy dro
ppe
d
from 645.2
6
ms to14
5.47
msTh
e speed
up of Gau
s
si
an filtering a
n
d
edge
extraction is 4, and
then the paral
lel efficien
cy of Gaussia
n
filtering an
d ed
ge extractio
n
is 1.
Evaluation Warning : The document was created with Spire.PDF for Python.
ISSN: 16
93-6
930
TELKOM
NIKA
Vol. 12, No. 3, September 20
14: 68
3 – 688
688
(b)
The time of edge contou
r e
x
traction, whi
c
h th
re
sh
old
segm
entation
firstly, decre
ase fro
m
the
origin
al 57
19.
23ms to 64
5.26ms. F
a
ll ti
me of this
m
e
thod d
epe
n
d
s o
n
the m
a
gnitude
of the
numbe
r of poi
nts nea
r the e
dge of the ed
ge and thresh
old se
gmenta
t
ion extracted
.
(c)
Thro
ugh im
proving Ca
nny
algorith
m
, co
mputing
time
from the o
r
i
g
inal 57
86.61
ms redu
ced
to 161.92m
s.
6. Conclusio
n
Usi
ng imp
r
ov
ed Otsu
algo
rithm extra
c
t ke
yhole’
s e
dge poi
nts a
nd nea
r, ca
n
greatly
redu
ce
the a
m
ount
of co
m
putation Can
n
y
ope
rato
r.
On the
ba
si
s
of multi-core t
e
ch
nolo
g
y, using
parall
e
l te
chn
o
logy on
Ga
ussian
filterin
g an
d c
ontou
r extra
c
tion
a
l
gorithm
in
Canny o
perato
r
,
signifi
cantly redu
ced
the ti
me
Canny
op
erato
r
ima
g
e
pro
c
e
ssi
ng to
improve the
operating
sp
e
ed.
Make full
use of Can
n
y o
perato
r
ed
ge
extrac
tion
a
c
cura
cy adva
n
tage
s whil
e
overcoming
the
disa
dvantag
e
s
of
Ca
nny o
perato
r
co
nsumes a l
ong
time. The
im
proved
Cann
y operator m
eet
time requi
rem
ents of the m
easure
m
ent system.
Ackn
o
w
l
e
dg
ment
This p
ape
r
suppo
rted by
the Gra
dua
te
Innovation
Program of
Shanxi (2
0
1330
95),
Shanxi Provi
n
ce S
c
ien
c
e
Found
ation
（
2013
0110
25
-1
）
, th
e
Sp
ecia
liz
ed
Re
s
ear
c
h
F
u
nd
fo
r
th
e
Do
ctoral Prog
ram of Hig
her E
ducation of
Chin
a (20
131
4201
2000
2)
Referen
ces
[1]
John C
a
n
n
y
. A
computatio
n a
ppro
a
ch to ed
ge detecti
on.
IEEE Trans Patten Analysis
a
nd Machin
e
Intelli
genc
e.
19
86; 8(11): 6
79-
697.
[2]
Xi
an
g
y
an
g X, Enmin
S, Lia
n
gha
i
J.
Ch
aract
e
rist
ic a
nal
ys
is
of thresh
old
b
a
sed
on Otsu c
r
iterion.
Acta
Electron
ica Sin
i
ca.
200
9; 37(1
2
): 2716-
27
19.
[3]
Bele
an B, Bor
da M, LeGa
l
B, et al.
F
P
GA techno
logy
and
para
lle
l c
o
mputi
ng tow
a
rds auto
m
atic
micr
oarray i
m
age pr
ocess
i
n
g
.
T
e
lecommu
nicati
ons a
n
d
Signa
l Proc
essin
g
(T
SP), 2011
34t
h
Internatio
na
l C
onfere
n
ce
o
n
IEEE. 2011: 60
7-61
0.
[4]
W
ang W
en-
yu
an. Sel
e
cting t
he Optimal Ga
ussi
a
n
F
ilterin
g Scale v
i
a th
e SNR of Ima
ge.
Jour
nal o
f
Electronics & Informa
tio
n
T
e
chno
logy.
2
009;
10: 248
3-2
487.
[5]
Bing
jin
g M, Bo
X,
Xi
aome
i
C,
et al.
Par
a
ll
el
Co
mp
uting
Re
nder
ing
In Sp
e
c
ific Re
mote S
ensi
ng I
m
a
g
e
Processi
ng
. Proc. of SPIE. 7850: 785
02
8-1.
[6]
Praja
pati HB,
Vij SK.
Analytic
al study of p
a
r
a
lle
l
a
nd d
i
strib
u
ted i
m
age
pro
c
essin
g
. Image Information
Processi
ng (IC
IIP), 2011 Internatio
nal
C
onfer
ence o
n
IEEE. 201
1: 1-6.
[7]
Kumar S, Pant M, Ray
A.
Differenti
a
l ev
oluti
on e
m
bed
ded Otsu'
s method for o
p
ti
mi
z
e
d i
m
ag
e
thresho
l
di
ng
. Informatio
n
an
d
Commun
i
cati
o
n
T
e
chnol
o
g
i
e
s (WIC
T
)
, 2011 World C
o
n
g
r
e
ss on IEEE.
201
1: 325-
329.
[8]
XU
Xi
an
g-
ya
n
g
, Song En-m
i
n
, Jin Lia
ng-
h
a
i. Char
acteris
t
ic Anal
ysis of
T
h
reshol
d Ba
sed on Otsu
Criterio
n.
Acta Electron
ica Sin
i
ca
. 200
9;12: 2
716-
271
9.
[9]
T
ang S, Lee B S, He B.
Spee
du
p for Multi-L
e
vel Par
a
llel C
o
mp
uting
.
Paralle
l and
Distribute
d
Processi
ng S
y
mposi
u
m Workshops &
PhD
Forum (IPDPS
W), 2012 IEE
E
26th Inter
n
a
t
iona
l IEEE.
201
2: 537-
546.
[10]
Hao
Xia
o
-
y
u
n
, F
an Yu-mei. Ho
w
to De
monstrate Par
a
lle
l Efficienc
y
w
i
th Pictur
e.
Comput
e
r
Engi
neer
in
g an
d Appl
icatio
ns
. 200
3;19: 11
8-1
19+
23
2.
Evaluation Warning : The document was created with Spire.PDF for Python.