TELKOM
NIKA
, Vol. 11, No. 10, Octobe
r 2013, pp. 5
571 ~ 5
578
ISSN: 2302-4
046
5571
Re
cei
v
ed Ap
ril 19, 2013; Revi
sed
Jun
e
22, 2013; Accepted July 1
0
,
2013
Resear
ch on Bottom Detection in Intelligent Empty
Bottle Inspection System
Bin Hua
ng, Sile Ma*, Yufeng Lv
, Hualong Zhang,
Chunming Li
u and Huajie
Wang
Schoo
l of contr
o
l scie
n
ce a
nd
eng
ine
e
ri
ng, Shan
do
ng Un
ive
r
sit
y
179
23, Jin
g
shi
Roa
d
, Ji’na
n
, Chin
a,
Ph./F
ax: +
86-0531
88
3
929
59/0
531
82
964
68
2
*Corres
p
o
ndi
n
g
author, e-ma
i
l
: hb-sdu
@
12
6.com,
silema
01
02@
gmai
l.com
*
, lv
y
u
f
eng
12@
ya
ho
o.cn
A
b
st
ra
ct
Intelli
gent e
m
p
t
y bottle ins
p
e
c
tion syste
m
i
s
an i
m
p
o
rtant
inspecti
on
eq
uip
m
ent of e
m
pty bottle
before fi
lli
ng
b
eer, an
d it is
a
ble
nd of
machi
ne vis
i
on,
preci
s
ion
machi
ne
a
nd re
al-ti
m
e c
o
ntrol. T
hey n
e
e
d
to cooperat
e perfectly to
achieve
the desired effect. In
the design of the
empty
bottle inspection system,
one of the k
e
y
technol
og
ies i
s
the bottle b
o
ttom detec
ti
on
w
h
ich affects the spe
ed a
n
d
accuracy of the
system. It incl
u
des p
o
sitio
n
i
n
g
and
defect rec
ogn
ition
of
bott
l
e b
o
ttom. For
the pro
b
le
ms s
u
ch as th
e sl
o
w
detectio
n
spe
e
d
and
low
dete
c
tion prec
isio
n
of bottle
botto
m d
e
tection, s
o
me new
meth
ods are
prop
os
ed
in this
p
aper.
T
he p
o
sitio
n
i
n
g al
gor
ith
m
of
the b
o
ttle b
o
ttom in
i
m
a
ges
i
s
studi
ed
after pre
p
rocess
ing
t
h
e
obtai
ne
d i
m
ag
es, an
d th
e acc
u
rate
positi
o
n
i
n
g
is
ach
i
e
v
ed
by im
p
r
o
v
i
n
g
th
e R
a
nd
omi
z
ed
H
o
ug
h
tran
sfo
r
m
.
In the
defect r
e
cog
n
itio
n
of b
o
ttle b
o
ttom,
a
metho
d
of
c
a
l
c
ulati
ng
opti
m
u
m
r
adi
us i
n
F
o
urier s
pectru
m
i
s
used to solv
e the pro
b
le
m of the det
ecti
on ac
curacy be
ing i
n
fluenc
ed by the
antiskid ve
ins
of bottle botto
m.
It can i
m
prov
e
the rec
o
g
n
itio
n
accur
a
cy
effectively. Ex
p
e
ri
ments sh
ow
the
meth
ods
pr
opo
sed
in
this
pa
p
e
r
can effectively
improve th
e pr
ecisio
n an
d sp
eed of the b
o
ttle botto
m detec
tion.
Ke
y
w
ords
:
int
e
lligent empty bottle ins
pec
tion system
,
m
a
c
h
ine vision, posi
tioning of bottle bottom
,
def
ect
re
co
gn
i
t
i
on
Copy
right
©
2013 Un
ive
r
sita
s Ah
mad
Dah
l
an
. All rig
h
t
s r
ese
rved
.
1. Introduc
tion
Empty bottle inspection bef
ore filling beer is
one of the important
processes of t
he beer
prod
uctio
n
project. It is difficult to guarantee
the rel
i
ability and adapt to the requireme
nts of
mode
rn
high
-spe
ed
produ
ction li
ne
usi
ng the
tra
d
it
ional
way
of
manual
d
e
tection. The
r
efo
r
e, it
results in so
me sub
s
tan
d
a
rd be
er influ
x
into
the market and bri
ngs da
mag
e
to the corp
orate
image [1]. Intelligent e
m
pt
y bottle inspe
c
tion
system
based o
n
ma
chin
e visio
n
tech
nolo
g
y ca
n
not only overcome the d
e
fects of the traditional
ma
n
ual insp
ectio
n
, but also can achi
eve the
automatic co
ntrol of a
pro
ductio
n
line t
h
rou
gh
com
p
uter p
r
o
c
e
ssi
ng, wh
i
c
h
ca
n greatly imp
r
ove
the degree of
automation o
f
the beer pro
ductio
n
line
s
[2-4].
Intelligent e
m
pty bottle i
n
sp
ectio
n
system i
s
a hi
gh-spe
ed
onl
ine te
sting
e
quipme
n
t,
whi
c
h gathe
r
machi
ne visio
n
, preci
s
io
n machi
n
e
r
y and real
-time co
ntrol in one
system. It mainly
con
s
i
s
t of p
r
e-in
spe
c
tion
unit, re
sidu
e
insp
ectio
n
unit, bottle
mouth in
sp
e
c
tion u
n
it, b
o
tto
m
insp
ectio
n
un
it, bottle wall
s in
spe
c
tion
unit, control
unit and m
a
n
-
ma
chin
e inte
rface
unit, as is
sho
w
n in Fig
u
re 1. The m
a
in function
s
inclu
de bottle
mouth brea
kage inspe
c
tio
n
, the dirt an
d
foreign
body
insp
ectio
n
of the bottle mo
uth, bo
ttom a
nd the wall, insp
ectin
g
re
sidue liqui
d in
a
bottle and
rej
e
cting th
e b
o
ttles unq
ualifi
ed in time. B
o
ttom dete
c
tion is impo
rta
n
t in the
who
l
e
detectio
n
system, and the detection sp
eed and a
ccu
ra
cy still ca
n’t meet the
deman
d of high-
spe
ed
pro
d
u
c
tion li
ne [5].
It inclu
d
e
s
p
o
sitioni
ng a
n
d
defe
c
t reco
gnition
of bot
tle bottom.
With
the imp
r
ove
m
ent of
pro
d
u
ction
line
a
u
tomation, it
puts forwa
r
d mo
re
nee
d
s
o
n
spe
ed
and
pre
c
isi
on of
bottle insp
ect
i
on. Bottom image po
siti
o
n
ing is o
ne o
f
the importa
nt factors affect
detectio
n
sp
e
ed and i
s
the
basi
s
of pre
c
ise dete
c
tio
n
of bottom d
e
fect. Moreo
v
er, in the bo
ttle
bottom d
e
fect detectio
n
, t
he a
c
cu
ra
cy
of the
def
e
c
t
detectio
n
i
s
i
n
fluen
ced
greatly be
cau
s
e of
the existen
c
e
of the bottom antiski
d vein
s.
In orde
r to i
m
prove th
e d
e
tection
sp
ee
d
and a
c
cu
ra
cy, some
met
hod
s are pro
posed in
this pa
pe
r. A bottom po
sit
i
oning
algo
rithm is
propo
sed
after re
proce
s
sing
th
e image obtain
ed.
Becau
s
e the
bottom image doe
s not
have a clea
r edge a
nd there a
r
e ma
ny interfere
n
c
e
s
comin
g
from
antiskid vein
s, we
process
the image
wit
h
edg
e dete
c
t
i
on an
d chain
-
co
de traci
ng
to
Evaluation Warning : The document was created with Spire.PDF for Python.
ISSN: 23
02-4
046
TELKOM
NIKA
Vol. 11, No
. 10, Octobe
r 2013 : 557
1 –
5578
5572
cal
c
ulate
the
chai
n-cod
e
’s
circumfe
ren
c
e. The
n
we
fi
lter
off some
small clutte
r edge
by setting
threshold
s
. Finally we can
locate the center
of the bottom by using an impro
v
ed rand
omi
z
ed
Hou
gh tra
n
sf
orm. The
r
e a
r
e ma
ny su
rface
defe
c
t
de
tection alg
o
rit
h
ms [6]-[9], a
nd the alg
o
rit
h
m
gene
rally u
s
e
d
is Bl
ob a
nal
ysis [8,9].
Ho
wever, th
ese
algorith
m
s
ca
n’t be u
s
e
d
di
rectly b
e
cau
s
e
of the antiski
d veins on b
o
ttle bottom. Therefo
r
e,
in the defect recognitio
n
o
f
bottle bottom,
becau
se the
antiskid vein
s of b
o
ttle b
o
ttom influen
ce the
dete
c
t
i
on a
c
curacy,
we
solve
d
this
probl
em u
s
in
g a method o
f
calcul
ating
optimum ra
di
us in Fo
urie
r
spe
c
tru
m
. It
can imp
r
ove
the
recognitio
n
a
c
cura
cy effe
ctively. Experimental
re
su
lts sh
ow t
h
a
t
this metho
d
ca
n not
o
n
ly
improve the i
n
sp
ectio
n
sp
eed, but ca
n al
so imp
r
ove
the insp
ectio
n
accuracy.
Figure 1. Structure di
ag
ram
of t
he empty
bottle insp
ecti
on syste
m
2. Image Preproces
sing
In the intelligent empty bottles inspection sy
stem, the original images obtained by CCD
came
ra have
a certain de
gree of noi
se
becau
se
of
being subje
c
t
ed to various noise sources
durin
g thei
r
gene
ration
a
nd tra
n
smissi
on. At the
same time, th
e slig
ht sl
oshing of
bottle
s
in
transmissio
n
and the no
n-unifo
rm illu
mination w
ill
lead to the gray of the obtained im
age
incr
ea
se o
r
d
e
cr
ea
se
sud
d
enly
.
This
wil
l
giv
e
birth to
the edg
e formed by false
obje
c
ts, whi
c
h
cau
s
e
s
ima
g
e
blurrin
g
, an
d then b
r
ing
difficulties to
the image
an
alysis. So it i
s
ne
ce
ssary
to
take mea
s
u
r
es of image
pre
-
processin
g
method
s.
For exampl
e, usin
g morph
o
logy method
s to
remove the n
o
ise a
nd co
rrect the uneve
n
illuminatio
n
to highlight the intere
sting
characte
ri
stics
of the images.
First
a medi
a
n
filter sho
w
n
in Equ.1 i
s
u
s
ed
to
re
mov
e
the n
o
ise of
the imag
e o
b
tained,
and th
en i
m
a
ge
cont
ra
st e
nhan
cem
ent i
s
a
c
hi
eved
b
y
a g
r
ay tran
sform
a
tion fu
nction
sho
w
n
in
Figure 2.
(,
)
(
,
)
(
,
)
g
i
j
me
di
an
f
m
k
n
l
k
l
w
(1)
W
h
er
e (
m, n
) i
s
the
coord
i
nate of
ce
ntral pixel,
and
(
i, j
) i
s
the
co
ordin
a
te of
proce
s
sing
pixe
l.
K, l =
-1,0,1
.
Figure 2. Gra
y
transform function
bottle flow
bottle
wall(2)
bottle
wall(1)
rejection
confirm
reject
Appar
a
tus
Residual
liquid
inspection
Initial
inspection
Bottle
M
outh and
botto
m
inspection
Evaluation Warning : The document was created with Spire.PDF for Python.
TELKOM
NIKA
ISSN:
2302-4
046
Re
sea
r
ch on
Bottom
Detection in Intelligent Em
pty Bottle Inspectio
n
System
(Sile Ma)
5573
Acco
rdi
ng to
experim
ents,
whe
n
the val
ue of (
x
1
, y
1
) i
s
(
30
, 1
5
) an
d the value
of (
x
2
, y
2
)
is (
120, 200
), the results
o
f
image
s g
r
a
y
stretching
are
mu
ch b
e
tter. We
u
s
e
Can
n
y ope
rat
o
r
[10] to dete
c
t the edg
e of
the bottle bot
tom. It not
only effectively detect
s
ed
ge
s of obj
ect
s
b
u
t
also filters ou
t partial edge
point
s that a
r
e not impo
rtant, which is
helpful to the
follow-u
p
ch
ain
cod
e
traci
ng
and imag
e lo
cali
zation.
3. Location
of Bo
ttle
Bottom
In the course
of inspecting
empty bottles, shak
e or tilt of bottles will
cause the difference
of the im
age
acqui
red
ea
ch tim
e
. So i
t
is
ne
ce
ssa
ry to lo
cate t
he e
m
pty bot
tles p
r
e
c
i
s
ely to
make
sure th
e imag
es wit
h
in the
re
gion
s of i
n
tere
st e
v
ery time. In
real
appli
c
ati
on, the
sp
eed
of
the empty bo
ttle inspe
c
tio
n
can
be up
to a maximu
m of 20 bottles pe
r secon
d
. However,
the
locatio
n
p
r
ocessing
algo
rithm o
c
cupie
s
most ti
me
o
f
the wh
ole i
m
age
pro
c
e
s
sing. So
a ra
pid
and efficie
n
t location al
g
o
rithm is e
s
sential fo
r improvin
g the
overall pe
rforma
nce of the
intelligent em
pty bottle inspectio
n
syste
m
.
The pri
n
cipl
e of image po
si
tioning take
s
the bes
t feat
ure poi
nts of the empty bottles as
the locatin
g
points first. Then search f
o
r the loca
tin
g
points a
nd
get their co
ordinate
s
in ea
ch
image. Th
e o
ffset of the image
can b
e
o
b
tained by
ca
lculatin
g the
different valu
e of the locating
points of two
continu
o
u
s
image
s. At last, the sa
me
offset can b
e
made to ma
ke sure that t
h
e
regio
n
s of int
e
re
st move preci
s
ely to the detectin
g
are
a
s.
Becau
s
e
the
imag
e of
b
o
ttle bottom
is a
ci
rc
le, bottle
bottom loc
a
ting
is
same as
insp
ectin
g
for ci
rcle
an
d fin
d
s th
e
cente
r
and
radi
us p
r
eci
s
ely a
c
cording to
the
ge
ometri
c featu
r
e
of circle. At p
r
esent, the m
a
in metho
d
s
of circle
dete
c
ting in
clu
de
detectin
g
ci
rcles
with Hou
g
h
transfo
rmatio
n, fitting circles with e
d
g
e
detectin
g
and lea
s
t sq
uare m
e
thod
[11], template
matchin
g
for
active ci
rcle
s
[12, 13] an
d
so o
n
. Be
cau
s
e the
r
e i
s
int
e
rferen
ce of
antiskid vein
s in
the bottom im
age, it is difficult to find th
e edge
poi
nt
s accurately. We a
dopt ch
ain-cod
e
tra
c
king
combi
ned
wit
h
the im
prov
ed
Rand
omi
z
ed
Houg
h tra
n
sform to
de
tect the
circl
e
in thi
s
p
a
p
e
r.
This meth
od
not only re
so
lves the shortcoming
of
the slo
w
tran
sf
ormatio
n
sp
e
ed of traditio
nal
Hou
gh tran
sf
orm, but also can o
b
tain th
e circle cente
r
accu
rately.
3.1. Chain Code Tracing
After the ed
g
e
dete
c
tion, t
he bottom
im
age
c
ontai
ns rich e
dge i
n
formatio
n, whi
c
h
has
some
unim
p
ortant poi
nts
or poi
nts n
o
t related to
th
e location. T
hese un
ne
ce
ssary ed
ge
s
will
affect the location accu
ra
cy of the bottle bottom.
Th
erefo
r
e, after detecting th
e edge,
we
must
use
Chain
co
de tracin
g to
filter out
that
clutter
ed
ge
according
to the
cal
c
ul
ation
o
f
the pe
rimet
e
r
of chain
cod
e
.
The perimet
er ca
n be giv
en as,
0
2
e
p
er
i
m
e
t
er
n
n
(2)
Whe
r
e
n
e
is the n
u
mbe
r
of
even n
u
mb
e
r
s i
n
the
chai
n code,
n
o
is t
he n
u
mbe
r
of
odd
num
bers in
the ch
ain
cod
e
. The eve
n
numbe
rs in t
he chain
c
o
d
e
re
pre
s
e
n
t h
o
rizontal di
re
ction a
nd ve
rtical
dire
ction, and
the odd num
bers re
pre
s
e
n
t the other di
rectio
ns.
After each
ed
ge pe
rimeter i
s
cal
c
ul
ated b
y
m
eans of chain-co
de tra
c
ing, we ca
n
remove
the unn
ecessary ed
ge
s a
c
cording
to set perim
eter
th
resh
old. After
Can
n
y dete
c
tion, the lo
cati
on
edge of bottle
bottom is discontin
uou
s sometime
s. In
this pap
er, the threshold of
perimete
r
is
set
to 100 by rep
eated test
s.
3.2. Impro
v
e
d
Randomi
z
ed Houg
h Tr
ansform Alg
o
rithm for Ci
rcle Inspec
tion
Hou
gh tra
n
sf
orm h
a
s hi
gh
pre
c
isi
on a
n
d
stro
ng a
n
ti-interferen
c
e
cha
r
a
c
teri
stics, whi
c
h
can
be a
pplie
d to dete
c
t arbitrary
curve
with an
al
ytic form. Howeve
r, its obviou
s
disa
dvantag
e
is
the com
puta
t
ional compl
e
xity and the slo
w
spe
ed. The
r
efore, the Ra
nd
omize
d
Hou
g
h
Tran
sfo
r
m al
gorithm i
s
usually use
d
in t
he req
u
iri
ng rapid in
spe
c
tio
n
situation.
Ran
domi
z
ed
Hou
gh T
r
a
n
sf
orm
algo
rith
m is to
sele
ct
the small
e
st point set ran
domly
in
image spa
c
e
,
and then map it into a point in t
he param
eter
space. Becau
s
e this al
gori
t
hm
belon
gs to
m
any-to-one
m
appin
g
, it avoids th
e la
rg
e co
mputatio
nal complexit
y
of tradition
al
Hou
gh tran
sf
orm. Ho
wever, becau
se the small
e
st
p
o
int set of circle is
comp
o
s
ed of thre
e non-
collinear points of the
edg
e, the smallest poi
nt set
may be
not
on
the same
real
ci
rcl
e
when
Evaluation Warning : The document was created with Spire.PDF for Python.
ISSN: 23
02-4
046
TELKOM
NIKA
Vol. 11, No
. 10, Octobe
r 2013 : 557
1 –
5578
5574
there
are
mo
re tha
n
on
e
circle
ch
ara
c
t
e
risti
c
s in the
image
s. Thi
s
not
only int
r
odu
ce
s i
n
valid
units, but al
so be
come
s i
n
terferi
ng fa
ctors if t
he p
a
r
amete
r
list
contain
s
the
circle
pa
ramet
e
rs
cal
c
ulate
d
by this kin
d
of smallest poi
nt set. We
can solve it by testing t
he small
e
st point set. For
the sam
e
bat
ch of em
pty bottles, the v
a
lue
s
of bottom ra
diu
s
wil
l
cha
nge in
a
certai
n ra
ng
e.
Therefore,
we ad
d a
con
s
traint
con
d
itio
n of th
e radi
us val
ue
ra
n
ge fo
r the
smallest
point
set.
Whe
n
select three
p
o
ints non-colli
nea
r rand
omly,
cal
c
ulate the
ra
dius of
the ci
rcle determin
ed
by these th
ree p
o
ints,
a
nd
com
pare
it with th
e v
a
lue
ran
ge. I
f
it is
within
the sco
pe, t
hen
contin
ue the f
o
llowin
g
step
s. If not, then
re-sel
ec
t a
n
o
t
her three p
o
i
n
ts, and
cal
c
ulate the
radi
us
value.The ste
p
s of the improved algo
rith
m are a
s
follo
ws,
(1) Set the
ra
nge
of radiu
s
value, an
d
structure
an
ed
ge p
o
int
set
D
. Initialize th
e pa
ram
e
ter
unit
set
P=
NUL
L
, the loop num
ber
K=0
, and
the numbe
r o
f
circle
s in
spe
c
ted
n=
0
;
(2) Sele
ct a smallest poi
nt set from
D
ra
ndomly, and cal
c
ulate the
radiu
s
of a ci
rcle d
e
termi
n
ed
by these thre
e points;
(3)
Jud
ge
whether th
e radiu
s
obtain
ed in the
ra
nge. If it is, then solve
circle p
a
ra
m
e
ter,
otherwise go
to (2);
(4) S
earch
a
circle
parame
t
er
p
c
me
etin
g the
con
d
ition of
ǁ
p
c
-p
ǁ
˂δ
at
P
, if find
then g
o
to (6),
otherwise go
to (5);
(5)
Ins
e
rt
p
into the parameter unit s
e
t
P
, and
set its co
rre
sp
ondi
n
g
accu
mulato
r value
as1, then
go to (7);
(6) Let
t
, the value of the
accumul
a
tor
corre
s
p
ondin
g
p
c
, add on
e
.
If
t
less tha
n
the thre
shol
d
T
,
then go to (7
), otherwi
se g
o
to (8);
(7)
K=
K
+
1
. If
K
˃
K
ma
x
, then
the algorith
m
end, otherwi
se go to (2);
(8)
p
c
is th
e
can
d
idate
circle p
a
ra
mete
r. Cal
c
ulate
the num
be
r
Mp
c
of the pi
xels on th
e circle
corre
s
p
ondin
g
to the para
m
eter. If
Mp
c
˃
M
mi
n
,
then go to (9), othe
rwi
s
e
p
c
is a false circl
e
para
m
eter. Remove th
is
parameter from
P
and go to (2);
(9) Structu
r
e
an ed
ge p
o
int
set
with the
edge
point
s o
n
the
corre
s
p
ondin
g
ci
rcl
e
of
p
c
, and m
a
ke
least
squ
a
res fitting to get the p
r
e
c
ise pa
ramete
rs
of a c
i
rc
le, then remove the
point s
e
t from
D
,
n=n
+
1
. Set
P=
NULL
,
K=
0
, and
co
ntinue to in
spe
c
t the rest ci
rcles to dete
r
m
i
ne wh
ethe
r
the num
ber o
f
the ci
rcl
e
s
h
a
s
been
in
sp
ected
re
ach t
he p
r
e
s
cribe
d
numb
e
r. If it
is, then
end,
otherwise, go
to (2).
The improve
d
algorith
m
can redu
ce
a lot of invalid ope
ratio
n
and in
cre
a
se the
comp
utation
spe
ed, and it can d
e
tect th
e circle
mo
re
accurately with less com
p
u
t
ation.
The
po
sitioni
ng
re
sults are sho
w
n i
n
F
i
gure
3.
(a
) i
s
ori
g
inal
imag
e, (b
) i
s
th
e
result
of
Image e
dge
detectio
n
, (c)
is the
imag
e f
iltered
by ch
a
i
n code
tra
cki
ng, (d
) i
s
th
e
dete
c
ted
re
sult
by improved random
Hou
g
h
transf
o
rm,
and the re
d cr
oss is th
e located ce
nter of
the circle.
(a)
(b)
(c
)
(d)
Figure 3. The
result
s of the
bottom positi
oning
4. Defe
ct
Det
ection o
f
Bo
t
t
le Bo
ttom
In defects de
tecting proce
ss, the imag
e of
bottle bottom is divided into two
region
s,
s
h
ow
n as
F
i
gu
r
e
4
,
acc
o
r
d
in
g
to its
c
har
a
c
ter
i
s
t
ics. ‘
A
’ is th
e
regi
on
contai
ning
antiskid
vein
s,
whi
c
h have
g
r
eat influe
nce
on dete
c
tion
results.
The
r
efore, spe
c
ial
algorithm m
u
st be u
s
e
d
to
filter out the
veins a
nd th
en the Blo
b
algorith
m
ba
sed on
conn
e
c
ted d
o
mai
n
is u
s
ed
to de
tect
defect
s
. ‘B’ is the surfa
c
e region
whi
c
h c
an be dete
c
te
d by Blob algorithm directl
y
.
In this pape
r, Fourie
r tran
sform
s
is u
s
ed to
remov
e
the reg
u
lar texturing of a bottle
bottom imag
e. There i
s
a
certain
relati
onship
bet
we
en reg
u
larity texture and F
ourie
r spe
c
trum
Evaluation Warning : The document was created with Spire.PDF for Python.
TELKOM
NIKA
ISSN:
2302-4
046
Re
sea
r
ch on
Bottom
Detection in Intelligent Em
pty Bottle Inspectio
n
System
(Sile Ma)
5575
[14]. Therefo
r
e, acco
rdi
n
g
to t
he ch
ara
c
teri
stics of
antiskid vein
s, we filtered
out the re
g
u
lar
texture in the image by sel
e
cting a
n
opti
m
um radi
us i
n
Fouri
e
r spe
c
trum.
Figure 4. The
partition of a bottle bottom detectio
n
regi
on
4.1. T
w
o
Dimensional Dis
c
rete Fourier
Transform
Suppose the
size of a two
-
dime
nsi
onal
image is
N×
N
, and the gra
y
value of pixel (
x,
y
)
in the image
is
f
(
x,
y
)
. x=
-N/2,…,N/2,
y
=
-N/2,…,
N
/2
. The two
dimen
s
ional
discrete F
o
u
r
ier
trans
form is
,
22
22
1
(,
)
(
,
)
e
x
p
2
(
)
/
NN
NN
xy
F
uv
f
x
y
j
u
x
v
y
N
N
(3)
Whe
r
e
u
and
v
are freq
uen
cy variabl
es, and
u,
v=-N/2,…,
N
/2
. The two
-
dim
ensi
onal Fo
u
r
ier
transfo
rm
can
be expre
s
se
d usin
g com
p
lex numbe
r,
(,
)
(
,
)
(
,
)
Fu
v
R
u
v
j
Iu
v
(4)
Whe
r
e
22
22
(,
)
(
,
)
c
o
s
2
(
)
/
NN
NN
xy
R
uv
f
x
y
u
x
v
y
N
22
22
(,
)
(
,
)
s
i
n
2
(
)
/
NN
NN
xy
Iu
v
f
x
y
u
x
v
y
N
Therefore, im
age po
we
r sp
ectru
m
ca
n b
e
defined a
s
,
2
22
(,
)
(
,
)
(,
)
(
,
)
P
uv
F
u
v
R
u
v
I
u
v
(5)
Whe
r
e the a
m
plitude fun
c
tion,
ǀ
F(u, v)
ǀ
,
is image spe
c
trum.
If there are d
e
fects in a
n
image, the fr
e
quen
cy of gray chan
ge wi
ll be low, an
d
P(u,v)
will co
ncentrate in low fre
quen
cy area.
Otherwise
, If there is p
e
riod texture in
the image, the
freque
ncy of
gray ch
ang
e
will be hig
h
,
and
P(
u
,
v
)
will
concentrate in hi
gh f
r
equency areas.
Therefore,
we can filte
r
o
u
t pe
riodi
c te
xture a
c
cordi
ng to th
e
distribution
of
spectrum
ene
rgy.
First, we get an
o
p
timum radiu
s
,
r
opt
, in
Fouri
e
r
po
we
r spe
c
trum,
a
nd the
n
remo
ve the fre
que
ncy
element
s of
the cente
r
an
d
out
side
of optimum
ra
dius f
r
om
Fou
r
i
e
r
sp
ectrum.
Last, the
ima
g
e
without re
gul
ar texture can
be obtaine
d by inverse F
o
urie
r tran
sform.
Evaluation Warning : The document was created with Spire.PDF for Python.
ISSN: 23
02-4
046
TELKOM
NIKA
Vol. 11, No
. 10, Octobe
r 2013 : 557
1 –
5578
5576
4.2. Selectin
g the Op
timum Radius
Suppo
se the
size of an image is
N×
N
, a
nd define e
n
e
r
gy mean
E
(
r
),
22
2
1
()
(
,
)
uv
r
r
Er
P
u
v
N
(6)
Whe
r
e
N
r
i
s
the numb
e
r of
the frequen
cy elements in
side the ci
rcl
e
with radi
us
r
, and
P
(
u,v
) is
the power sp
ectru
m
.
E
(
r
) d
enote
s
the avera
ge
value of the energy
inten
s
ity of all frequen
cy eleme
n
ts insi
de
the ci
rcl
e
wit
h
ra
diu
s
r
, a
nd
r=0,1,2,3,
…,N/2
. Defining
E
(
r
)
as ve
rtical
axis a
n
d
r a
s
hori
z
o
n
tal
axis, we can
obtain averag
e
energy cu
rve. Calculate the
slo
pe an
gl
e of the curve
,
1
()
(
)
()
t
a
n
(
)
E
rE
r
s
r
s
(7)
Whe
r
e
s
is sp
acin
g, and
ψ
(
r
) is the
slop
e
angle of the curve
E
(
r
) wh
en the radi
us
is
r
.
The cu
rvature of curve
E
(
r
) can b
e
obtai
ned u
s
ing the
equation a
s
follow,
1
()
()
(
)
()
t
a
n
(
)
rr
r
s
kr
s
s
(8)
Whe
r
e
k
(
r
) i
s
the cu
rvature of the cu
rv
e
E
(
r
) wh
en t
he ra
diu
s
is
r
. Then the o
p
t
imum radi
us
r
op
t
can b
e
obtain
ed,
ar
g
m
a
x
(
)
opt
r
rk
r
After obtainin
g
the optimu
m
radiu
s
, Fou
r
ier tra
n
sfo
r
m
can be exp
r
e
s
sed a
s
,
22
2
ma
x
0
,
(
,
)
0
(,
)
(
,
)
,
if
u
v
r
o
r
u
v
Fu
v
Fu
v
o
t
h
e
r
w
i
s
e
(9)
4.3. Images Res
t
ore u
s
in
g In
v
e
rse Fourier Trans
f
orm
In the Fourier
s
p
ec
trum, set the frequenc
y
el
ement
s at the
ce
nte
r
an
d o
u
tsid
e
of the
optimum radi
us as ze
ro. The remai
n
in
g
freq
uen
cy element
s within
the opti
m
um
radiu
s
are
defect inform
ation. Usi
ng inverse Fou
r
i
e
r tran
sf
orm, we can obtai
n the
origin
al image which has
been filtere
d
interfere
n
ce
information
su
ch a
s
re
g
u
larity texture on the pre
m
ise of retai
n
ing
useful info
rm
ation. The eq
uation of inve
rse F
o
u
r
ier transfo
rm is,
22
22
1
(,
)
(
,
)
e
x
p
2
(
)
/
NN
NN
uv
f
xy
F
u
v
j
u
x
v
y
N
N
(10
)
After obtainin
g
the imag
e
without regul
arity
texture, we
can
get the bin
a
ry im
age by
binari
z
atio
n processin
g
, an
d then obtain
the chara
c
te
ristic d
a
ta of defect correctly using Blob
analysi
s
.
Usi
ng the me
thods p
r
o
p
o
s
ed above,
we
can d
e
tect
d
e
fects
co
rrect
l
y even if they exist in
and their g
r
a
y
value close
to the antiskid vein
s area
. The experi
m
ental re
sult
s are sh
own in
Figure 5. Fig
u
re 5
(a
) is t
he ori
g
inal im
age. (b
) i
s
th
e Fou
r
ier
sp
e
c
trum
of ima
ge (a
). (c) i
s
the
image of filtering re
gula
r
texture usi
n
g
optimum
ra
dius. (d
) is the image of
Inverse Fou
r
ier
transfo
rm an
d (e) i
s
the im
age after Bin
a
rization p
r
o
c
essing.
The re
sult
s of bottle bottom defect
s
reco
gnition b
a
se
d on the
above algo
rithm are
s
h
ow
n
in
F
i
gu
r
e
6
.
Evaluation Warning : The document was created with Spire.PDF for Python.
TELKOM
NIKA
ISSN:
2302-4
046
Re
sea
r
ch on
Bottom
Detection in Intelligent Em
pty Bottle Inspectio
n
System
(Sile Ma)
5577
(a)
(b)
(c
)
(d)
(e)
Figure 5. The
partition of a bottle bottom detectio
n
regi
on
(a) Original images
(b) D
e
tectio
n
result
s
Figure 6. The
defect dete
c
tion of bottle bottom
5. Conclusio
n
The
detection of bottle bottom is one of th
e key technologies i
n
the intelligent bottle
insp
ectio
n
system
ba
sed
on the
ma
chi
ne visi
on,
an
d it h
a
s a
direct effe
ct o
n
insp
ectio
n
sp
eed
and
accu
ra
cy of the
syste
m
. In this
pa
per,
we
studi
ed the
bottle
bottom lo
cat
i
on a
nd d
e
fe
ct
detectio
n
m
e
thod
s o
n
the
basi
s
of p
r
e
-
pro
c
e
ssi
ng
a
nd e
dge
in
sp
ection
of th
e
bottom ima
g
e
.
In
the location
of bottom, we use the ch
ain cod
e
tracing method to filter out th
e unwa
n
ted
edge,
and ad
opt th
e improve
d
Ran
dom
Hou
gh Tra
n
sfo
r
m
to inspe
c
t ci
rcle. T
he imp
r
oved
Ran
d
o
m
Hou
gh Tran
sform can redu
ce a lot of invalid ope
rati
on
by increa
sin
g
the con
s
trai
nt conditio
n
s
of
radiu
s
,
whi
c
h
ca
n
sig
n
ifica
n
tly improve
t
he
spe
ed
an
d p
r
e
c
isi
on
of location. In
the d
e
tectio
n
of
bottom defe
c
ts, we m
a
inly
discu
s
sed h
o
w to eli
m
ina
t
e the effect
of antiskid ve
ins o
n
dete
c
ti
on
results. Fou
r
i
e
r tran
sform techni
que is
use
d
in this pape
r to rem
o
ve the regu
lar texture b
y
cal
c
ulatin
g o
p
timum ra
diu
s
an
d the d
e
fect det
e
c
ti
on is a
c
hi
eved by Blob
analysi
s
at l
a
st.
Experiment result
s sh
ow t
hat the algo
ri
thm use
d
in this pa
pe
r ca
n improve th
e efficien
cy a
nd
accuracy of d
e
fect dete
c
tio
n
and ha
s hig
h
pra
c
tical val
ue.
Referen
ces
[1]
Sile MA,
Huiquan WANG, Zengben
HAO, et al.
Ap
plic
atio
n
researc
h
of
machi
ne v
i
sio
n
t
e
chn
i
qu
e i
n
intelligent empty bottle inspection syst
em
. T
he 8th W
o
rld C
ongr
ess
on Inte
lli
gent
Contro
l a
n
d
Automatio
n
(W
CICA). Ji’na
n
, Chin
a. 20
10: 4
462-
446
6.
[2]
Minh
ua LAN. Appl
icatio
n
of Contro
lli
ng
T
e
chno
log
y
on
the Empt
y B
o
ttle Insp
ectio
n
.
Me
ch
an
i
c
al
&
Electrical E
ngi
neer
ing T
e
c
h
n
o
lo
gy
. 201
1; 40(8): 133-
13
6.
[3]
Zhilia
ng
Wang
, Jian Ga
o, C
hua
n
x
ia
Jia
n
,
Yu
Ce
n, Xi
n Che
n
.
OLED Defect
Insp
ection
S
y
st
e
m
Devel
opm
ent
throug
h In
dep
end
ent C
o
mp
one
nt An
al
ysi
s
.
T
E
LKOMNIKA Indo
nes
ia
n Jo
urna
l
o
f
Electrical E
ngi
neer
ing,
20
12,
10(8): 230
9-2
319.
Evaluation Warning : The document was created with Spire.PDF for Python.
ISSN: 23
02-4
046
TELKOM
NIKA
Vol. 11, No
. 10, Octobe
r 2013 : 557
1 –
5578
5578
[4]
MI Chao, LIU
Hai
w
e
i
, Z
H
AO Ning, SHE
N
Y
ang. A
Shi
p
C
a
rgo H
o
l
d
Insp
ection A
ppro
a
c
h
Usin
g Las
er
Vision S
y
stem
s.
T
E
LKOMNIKA Indon
esia
n Journ
a
l of Elec
trical Eng
i
ne
eri
n
g
. 201
3, 11(1)
: 330-33
7.
[5]
Li
w
e
i Z
H
ANG, Han
g
Z
H
ANG. A Method
of On-Li
ne
Glass B
o
ttle Detecti
on
W
i
th Comput
erized Vis
i
o
n
.
Co
mp
uter Engi
neer
ing & Sci
e
nce
. 200
9; 31(
10): 36-3
8
.
[6]
Jianx
i
n YIN, Hengnian QI, Ha
ilin FENG, et
al. A method f
o
r
w
o
od surface def
ect
detect
i
on based on
mixed te
xture f
eatures.
Jour
n
a
l of Z
heji
a
n
g
A & F
University
. 2011; 28(
6): 937-
942.
[7]
Cha
o
Gao, Xi
nke Li, Yon
g
c
a
i Guo, et al. Mach
i
ne visi
on
detecting s
y
st
em for bridg
e
cabl
e defect.
Scienc
epa
per Onlin
e
. 201
1; 6(10): 775-
78
0.
[8]
Meih
ong S
H
I, W
engu
an
g
W
A
NG. F
abric Defect Detec
t
ion Al
gorithm
Based
on B
l
ob Al
gorithm
.
Moder
n Electro
n
ics T
e
chn
i
qu
e
.
2010; 24: 29-
32.
[9]
Min
X
U
, Wanyou T
A
NG, Qianli MA,
et a
l
.
Rese
arch
of P
r
intin
g
D
e
fect
On-lin
e D
e
tecti
on B
a
sed
o
n
Blob Al
gorithm.
Packagi
ng En
gin
eeri
n
g
. 20
1
1
; 32(9): 20-2
3
.
[10]
Can
n
y
JOH
N
. A Computati
ona
l Appro
a
c
h
to Edge D
e
tection.
Pattern Ana
l
ysis
and Mac
h
in
e
Intelli
genc
e
. 19
86; 8(6): 67
9-6
97.
[11]
Yunh
on
g W
A
NG, Yong Z
HU,
T
i
eni
u T
A
N.
Biometrics Per
s
ona
l Identif
ic
ation Bas
ed o
n
Iris Pattern
.
Acta Automatic
a
Sinic
a
. 200
2; 28(1): 1-10.
[12]
Dau
g
man
JOH
N
. Hig
h c
onfid
ence
visu
al r
e
c
ogn
ition
of per
sons b
y
a
test of
statistical
in
dep
en
denc
e.
Pattern Analys
i
s
and Mach
in
e Intelli
genc
e
. 19
93; 15(1
1
): 114
8-11
61.
[1
3
]
D
a
ug
ma
n
JOHN
.
Hig
h co
nfid
ence
reco
gniti
on
of pers
ons
by ir
is p
a
tterns
. IEEE 35th Internation
a
l
Carn
aha
n Co
nferenc
e. Lon
do
n,
Engla
nd. 20
01: 254-
26
3.
[14]
Bin HE, T
i
an
yu
MA.
Visual C+
+
Digital I
m
a
g
e
Processi
ng
. B
e
iji
ng, Posts &
T
e
lecom Press
.
2002: 9
2
-
93.
Evaluation Warning : The document was created with Spire.PDF for Python.