TELKOM
NIKA
, Vol.11, No
.11, Novemb
er 201
3, pp. 6714
~6
721
e-ISSN: 2087
-278X
6714
Re
cei
v
ed Ap
ril 8, 2013; Re
vised J
une 2
5
, 2013; Acce
pted Jul
y
27,
2013
Visualization of License Plate Recognition System
Zhiy
i
Ruan
1
, Ying
Zou
2
, Dongming Ho
ng
3
, Lurong Wu*
4
1,3,
4
Colle
ge of Comp
uter and
Information, F
u
jian A
g
ricu
lture
and F
o
restr
y
Univers
i
t
y
, F
u
z
hou, Ch
in
a
2
Colle
ge of Ma
terial En
gin
eer
i
ng, F
u
jia
n Agri
culture a
nd F
o
r
e
str
y
Un
iversit
y
, F
u
zho
u
, Chi
n
a
*Corres
p
o
ndi
n
g
author, e-ma
i
l
:
w
u
lur
ong
1@
sina.com
A
b
st
r
a
ct
T
he imag
e of the licens
e p
l
ate is loc
a
ted
and
seg
m
ent
ed by so
me d
i
gital-
i
m
a
ge pr
ocessi
n
g
techno
lo
gies
such as gr
ay
-scale pr
oces
sing, gr
ay-sca
le stretchi
ng
and filt
erin
g, edg
e det
ectio
n
,
mor
p
h
o
lo
gica
l process
i
ng, Ho
ugh
transfor
m
a
t
ion etc.
Acco
r
d
in
g to
the
cha
r
acteristics
of the
lice
n
se
pl
ate
,
bin
a
ry
matrix o
f
character i
m
a
ge sets th
e fu
zz
y
matr
ix, and
base
d
o
n
the p
r
inci
p
l
e
of prox
imity co
mp
utin
g
space of clos
e
ness to get the fu
zz
y
pattern
recogniti
on of
characters. On the bas
is of the data of imag
e
pixels, the sa
mp
les w
h
ich
are ran
d
o
m
ly
selecte
d
und
e
r
noisy cond
iti
on an
d w
h
ich
are treated by
mor
p
h
o
lo
gica
l
process
i
ng
are
rando
mly sel
e
cted, and th
en
the sa
mpl
e
s ar
e used to test the si
mu
latio
n
a
n
d
ide
n
tificatio
n
of Back Prop
agati
on (BP)
Neur
al N
e
tw
orks. W
i
th the math
e
m
atic
al
softw
are-Matla
b
progr
a
m
min
g
, the l
i
ce
nse
plat
es are
reco
gn
i
z
e
d
.
T
he
ai
m
is
to dev
el
op th
e
Visua
l
i
z
a
t
i
o
n
o
f
user i
n
terface
in
Lice
nse Plat
e Reco
gniti
on Sy
stem.
Ke
y
w
ords
:
lic
ense p
l
ate rec
ogn
ition, i
m
a
g
e
processi
ng, fuzz
y
p
a
ttern rec
ogn
ition, BP ne
ural n
e
tw
ork, GUI
Copy
right
©
2013 Un
ive
r
sita
s Ah
mad
Dah
l
an
. All rig
h
t
s r
ese
rved
.
1. Introduc
tion
Lice
nse plat
e re
cog
n
ition
system is a
sp
e
c
ial com
puter visi
on
system
whi
c
h take
s
licen
se pl
ate
for a spe
c
ific targ
et, is the u
s
age
of comp
uter vi
sion
and p
a
ttern Recogni
tion
techn
o
logy in
the field of Intelligent tra
n
s
po
rtati
on, Li
cen
s
e Plate
Re
cog
n
ition
System is on
e of
the most imp
o
rtant pa
rt of Intelli
gent Tra
n
sp
ort Syste
m
, ITS, as a re
se
arch h
o
tspot in the fiel
d of
mode
rn tra
n
sportation, it p
l
ays an i
m
po
rtant rol
e
in t
he man
age
m
ent of turb
an
tran
spo
r
tatio
n
,
high
way, and
parki
ng lot etc.
Automatic li
cense plate
reco
gnition te
chn
o
logy [1,
2] ca
n be
divided int
o
: radi
o
freque
ncy id
entification t
e
ch
nolo
g
y, bar code
recognition te
ch
nology an
d vehicle li
cen
s
e
recognitio
n
te
chn
o
logy. Th
e form
er two
techni
que
s
are a
c
curate
an
d reliable,
but
it still n
eed
s t
o
install
relate
d
devices an
d
to e
s
tabli
s
h
ba
ckgr
o
und
manag
eme
n
t in the
vehi
cl
e licen
se
pla
t
e
recognitio
n
system; Li
cen
s
e plate
re
co
g
n
ition
syst
em
is
ba
sed
on
video te
chn
o
l
ogy ha
rd
wa
re
so
that you can i
dentify and monitor the veh
i
cle directly.
At the beginn
ing of the 1980s, forei
gn rese
arche
r
s had wide
con
c
ern ab
out the
license
plate re
cog
n
i
t
ion technol
o
g
y. In the 1
990
s,
along
with the developme
n
t of comp
uter visi
on
techn
o
logy a
nd imp
r
ovem
ent of the
co
mputer
pe
rfo
r
man
c
e, li
cen
s
e pl
ate reco
gnition
syste
m
has
bee
n sy
stemati
c
ally rese
arche
d
. Ho
wever,
th
e syste
m
ca
n not re
co
gn
ize the
Chi
n
ese
cha
r
a
c
ters i
n
the
Chi
nese
lice
n
se pl
ate
.
At pre
s
e
n
t, there
are
so
me mo
re
ma
ture
pro
d
u
c
ts to
solve the i
s
sue of re
co
gn
izing
Chin
ese ch
ara
c
te
rs in
C
h
ina
suc
h
as
HW
eye
-
th
e
Ch
in
es
e
Acade
my of
Scien
c
e
s
In
stitute of Auto
mati
on HW
company, Hui
guan
g
plate numbe
r
auto
m
atic
identificatio
n
system
-ASIA Vision T
e
ch
nolo
g
y
L
t
d., some
relevant reco
gnition p
r
od
ucts
develop
ed
by Shen
zhe
n
Ke An
Xing Indu
strial
Com
p
any Limited
and th
e
Sino-
Chile
an
traffic
Electroni
cs
Comp
any Li
mited un
der t
he mini
stry of
Chin
a Info
rm
ation ind
u
st
ry. In
addition, so
me Universit
y
Departm
e
n
t
s’ laboratori
e
s, su
ch a
s
ar
tificial
-intelli
gen
ce State Key
Labo
rato
ry o
f
Tsing
hua
University, Shang
hai
Ji
ao
Tong
Unive
r
sity comput
er
sci
en
ce a
n
d
Enginee
ring
Dep
a
rtme
nt, the Depa
rtm
ent of aut
omation of Zheji
ang Unive
r
sity, have devoted
the sci
entific resea
r
ch stre
ngth into re
co
gnition
tech
n
o
logy are
a
. What’
s
more, Zhiyong Liu, who
is from th
e A
u
tomation In
stitute of Chin
ese A
c
ad
em
y of Science
s
, ha
s al
so p
ublished
relat
e
d
article
s
, an
d Aiming Hu, o
ne of the re
searche
r
s fro
m
Beihang
University, dev
elope
d a lice
n
se
plate re
cog
n
ition syste
m
o
n
the basi
s
of
the te
mplate
matchin
g
techniqu
e, and this sy
stem can
be appli
ed in
the toll stations.
Evaluation Warning : The document was created with Spire.PDF for Python.
TELKOM
NIKA
e-ISSN:
2087
-278X
Visuali
z
ation
of Licen
s
e Pl
ate Re
cog
n
ition Syst
em
(Zhiyi Ruan
)
6715
The stan
dard
-
sm
all ca
rs h
a
ve a wide ra
nge of us
e
r
s in China
so th
e aims of this article
are to re
se
arch the licen
se plate re
cog
n
ition sy
stem
in Chine
s
e
standard-small
car, to reali
z
e
the imag
e of
l
i
cen
s
e
plate
l
o
catio
n
cuttin
g
, and
then
to ide
n
tify the
cha
r
a
c
ter by f
u
zzy mo
del a
n
d
neural
n
e
two
r
k sele
ctive co
ncu
r
rent
o
perations. Othe
r types
of ca
rs can
al
so ch
o
o
se
re
co
gniti
on
databa
se i
n
accordan
ce
with the li
ce
n
s
e
plate f
eat
ure
s
a
nd
sp
e
c
ificatio
ns, to
com
p
lete li
cense
plate re
cog
n
ition.
2. Summar
y
of Licens
e
Plate Re
cogni
tion Sy
stem
2.1. Chinese
License Plate F
eatures a
nd Specifica
tions
In Chin
a, the
licen
se
plate
s
of
stand
ard
-
sm
all
cars
are white letters o
n
blu
e
, 44
0mm in
width
and
14
0mm in
lengt
h, Ratio
bet
ween th
e wi
dt
h
and
the h
e
ig
ht is
about
3.
14:1. The
r
e
a
r
e
seven
ch
ara
c
ters in th
e license plate. T
he first
cha
r
a
c
ter i
s
a
Chin
ese
ch
ara
c
te
r standi
ng for t
he
Abbreviatio
n
of
all provin
ces and
mu
ni
cipalitie
s;
the
seco
nd character i
s
one
of the 24 cap
i
tal
letters ex
cept
the letter "I" and "O", whi
c
h rep
r
e
s
ent
s the
cod
e
n
a
me of the i
s
suin
g auth
o
rit
y
.
Those letters, from the third letter to the s
e
vent
h, consi
s
t of 2
4
English
letters an
d n
u
mbe
r
s.
Each cha
r
a
c
ter in Licen
s
e
plate is ce
nte
r
ed in a wi
dth
of 45mm and 90mm in he
ight recta
ngul
ar
rang
e. Th
e
space b
e
twe
e
n
two
cha
r
a
c
ters i
s
1
2
mm.
There i
s
1
0
m
m
interval
sy
mbol "
●
"
bet
wee
n
the second
and th
e thi
r
d
ch
aracte
r,
so the
a
c
tual
interval
is
22mm. T
he
gene
ral
style
is
“
某
A·12
345
”.
2.2. The Co
mposition of Licen
se Plate Reco
gnitio
n
Sy
stem
Lice
nse pl
ate
re
cog
n
ition
system i
s
ba
sed
o
n
vide
o i
m
age
acqui
si
tion technol
o
g
y, and
the system i
s
using
com
p
u
t
er tech
nolo
g
y
to sear
ch a
nd judg
e vehi
cle licen
se pl
ate. The thre
e
main p
a
rts of
this recogniti
on sy
stem a
r
e lice
n
se plat
e imag
e lo
ca
tion, licen
se
plate cha
r
a
c
ter
segm
entation
and lice
n
se plate ch
ara
c
t
e
r re
co
gnition
.
Figure 7 sh
o
w
s the
syste
m
make
s the
licen
se plate
location
cutting on the im
age for
use
r
to
rea
d
.
Figure 8
sh
o
w
s the
syste
m
ma
ke
s fuzzy pattern re
cog
n
ition o
n
t
he
cutting li
cense
plate cha
r
a
c
ter a
nd di
spl
a
y results. Fig
u
re
9 sh
o
w
s
t
he sy
st
e
m
id
ent
if
ic
ation re
sults ba
sed o
n
use
r
nee
ds
to save text as txt format, and pr
ov
ides three result
s for ea
ch characte
r in
accordance
with the size of
the possibility of fuzzy patte
rn recognition, for example,
“E:\LPRS\Example1.jp
g” i
s
re
cog
n
ized
to be t
he license plate b
y
the fuzzy pattern. The
first
c
h
ar
ac
te
r
is
“
豫吉
苏
”;
the seco
nd cha
r
a
c
ter
i
s
“DQC”;
the
third
chara
c
te
r is “V97”; the fo
urth
c
h
arac
ter is
“0QD”; the fifth c
h
ar
ac
te
r is
“
0
86
”
;
the
s
i
xth
c
h
aracter is “085
”; the
sevent
h
cha
r
a
c
ter i
s
“172”. Th
us, the entire li
ce
nse pl
ate is
most likely to
be “
苏
D
V
0
0
01”
.
Figure 1. Flowchart of System Identifica
t
ion
3. Image Preproces
sing
Image
enha
ncem
ent [3,
4] is on
e
of the m
o
st
ba
sic meth
ods of di
gital imag
e
prep
ro
ce
ssin
g, which al
so be
co
mes a quite valuabl
e tech
nology to b
e
use
d
. Image
enha
ncement
technol
ogy is often used
to improve
th
e gray level i
m
age qu
ality, and to incre
a
se
the ratio of sign
al and n
o
ise. Thi
s
techn
o
l
ogy ca
n make som
e
feature
s
of the image be
recogni
ze
d e
a
sily.
3.1. Gra
y
Image
Gray imag
e is the basi
c
st
ep for the pretr
eatme
nt of license plate
images. The
original
image contai
ns a large nu
mber of color informati
on
whi
c
h o
c
cupi
es the sto
r
ag
e spa
c
e. Diff
erent
illumination i
n
tensity re
sul
t
s in recognit
i
on pro
b
lem
s
, so gray pro
c
e
ssi
ng shou
ld be made f
o
r
colo
r imag
es.
In the gray p
r
ocessin
g
,
R
,
G
,
B
are give
n dif
f
erent weight
values
R
,
G
,
B
.
Lice
nse pl
ate
location
Char
acter
se
g
m
en
ta
ti
on
Char
acter
re
co
gn
i
t
i
on
Output
Lice
nse pl
ate
Input imag
e
Evaluation Warning : The document was created with Spire.PDF for Python.
e-ISSN: 2
087-278X
TELKOM
NIKA
Vol. 11, No
. 11, Novemb
er 201
3: 671
4 – 6721
6716
The gray value is
B
G
R
B
G
R
B
G
R
I
. In
gene
ral, wh
e
n
299
.
0
R
,
587
.
0
G
and
114
.
0
B
, we ca
n
get the best gray imag
e.
3.2. Gra
y
Stretch
Gray st
retch
is a kin
d
of
image en
ha
ncem
ent technolo
g
y. The
image g
r
ay
value is
mappe
d to th
e entire
ra
ng
e of gray lev
e
l by gray
stretch tra
n
sfo
r
mation in o
r
d
e
r to in
crea
se the
contrast
of th
e imag
e. Extendi
n
g
the
range
of gray level ma
ke
s the cha
r
a
c
te
r featu
r
e
s
m
o
re
obviou
s
[5-7] and it i
s
mo
re
con
d
u
c
ive
to furthe
r p
r
oce
s
sing. In
the imag
e which th
e pixe
l is
n
m
, the histogra
m
is
i
h
. Gray st
retch tran
sformati
on functi
onal form
ulati
on:
b
x
d
b
x
b
d
b
x
a
c
a
x
a
b
c
d
a
x
x
a
c
x
f
255
255
(1)
Get
a
which sa
tisfies the sm
allest po
sitive
integral of
10
0
mn
i
h
a
i
, and
b
values th
e
large
s
t po
sitive integral of
mn
i
h
b
i
10
9
0
,
and set the value
s
of
c
and
d
.
3.3. Median Filtering
The o
u
tstan
d
i
ng
merit
of M
edian
Filterin
g is th
at it ca
n not o
n
ly eli
m
inate n
o
ise
but al
so
prevent bl
urred edg
es. M
edian Filte
r
in
g is
a ki
nd
of pra
c
tical
non-li
nea
r im
age smoothi
ng
method, and
this method
has a goo
d
effect on eliminating noi
se in pro
c
e
ssi
ng licen
se pl
ate
image.
The ba
sic p
r
i
n
cipl
e of the median filter i
s
to
sub
s
titute the averag
e value of the variou
s
points in an a
r
ea of the poi
nt for the value of
point in digital seque
nce, that is to say,
1
a
,
2
a
,
,
n
a
are in de
sce
nding o
r
de
r
n
i
i
i
a
a
a
2
1
.
And let
n
a
a
a
med
b
,
,
,
2
1
, then
b
is the average va
lue of seq
u
ence. Two
-
d
i
me
ns
io
na
l s
e
qu
e
n
c
e
ij
a
is
the gray valu
e of ea
ch im
age p
o
int. Wi
ndo
w 2-d me
dian filter
is
A
r
s
a
med
b
r
j
s
i
A
i
,
,
.
4. License Plate Extrac
tio
n
First
of all, license plate
lo
cation
nee
ds
to ma
ke e
dge
extractio
n
fo
r gray i
m
age
b
y
usin
g
Can
n
y, and then elimin
ating noi
se thro
ugh mo
rph
o
l
ogy pro
c
e
s
sing, the two
pro
c
e
s
ses
are to
make th
e im
age
s form a
conn
ecte
d a
r
ea. Acco
rd
in
g to the ch
a
r
acte
rs and
spe
c
ification
s
of
licen
se
plate
and
ph
otog
raphi
c angl
e, given th
at t
he fa
st
chan
ge of f
r
eq
ue
ncy i
s
in
ed
ge
grap
hics
of the li
cen
s
e
pl
ate, and
sel
e
cting
the l
e
ngth a
nd
wi
dth ratio
an
d
the
size of
the
con
n
e
c
ted re
gion at the sa
me time, we extract lice
n
se plate bina
ry image.
4.1. Binar
y
OSTU threshold segm
en
tation is a very
cla
s
sic
automatic th
resh
old segm
entation
algorith
m
. Its ba
sic pri
n
ci
ple i
s
that th
e hi
stogr
am i
s
divid
ed int
o
two
group
s from
a
cert
ain
threshold.
When the va
ria
n
ce
between
the two g
r
ou
ps i
s
the
max
i
mum then th
e co
rrespon
ding
threshold
con
s
ide
r
ed to be
the best thresh
old.
OST
U
the thre
sh
old se
gmenta
t
ion avoid so
me
disa
dvantag
e
s
such a
s
lon
g
co
mputatio
n time, low
effic
i
enc
y
, high c
o
s
t. It is
very effec
t
ive way
in solving
sin
g
le thre
sh
old
segm
entation
of gr
ayscal
e
image
s. OST
U
bin
a
ry is
b
a
se
d on
OST
U
threshold
seg
m
entation; it pro
c
e
s
ses th
e image wi
th
accordan
ce t
o
optimal thre
shol
d limits.
4.2. Edge De
tec
t
ion
Edge d
e
tecti
on ta
ke
s adv
antage
of ed
ge e
nhan
cive
ope
rato
r to
unde
rline
the
edg
e of
the pa
rt of
th
e ima
ge. Afte
r the
inten
s
ity of pixe
l
ed
ge
is defin
ed, th
e poi
nt
set
ca
n be
extra
c
te
d
Evaluation Warning : The document was created with Spire.PDF for Python.
TELKOM
NIKA
e-ISSN:
2087
-278X
Visuali
z
ation
of Licen
s
e Pl
ate Re
cog
n
ition Syst
em
(Zhiyi Ruan
)
6717
by the
set li
mits. The
co
mmon
dete
c
tion al
gorith
m
s are
Canny
o
perato
r
,
Rob
e
rts ope
rato
r,
the
LOG
operator, Wallis oper
ator, edge detection
me
thods based
on fractal
theory etc. Based on
Can
n
y operator, the gra
d
ie
nt and di
re
cti
on Angle of b
ound
ary are:
y
x
G
G
f
,
x
y
G
G
y
x
arctan
,
, (
x
f
G
x
,
y
f
G
y
)
(2)
Can
n
y op
erat
or i
s
to
see
k
l
o
cal
maximu
m value
of th
e ima
ge
gra
d
i
ent, so
to
sm
ooth the
image i
s
n
e
cessary.
Usi
n
g 5×5 Ga
ussi
an filter a
nd t
he Sigma
pa
rameter
co
ntrol the filter, a
n
d
usin
g two thresh
old se
gm
entation
s
en
han
ce
s
imag
e segm
entati
on. Can
n
y method can be
tter
balan
ce
the
edge
dete
c
ti
on a
nd
noi
se
su
ppressio
n
. It is the
mo
st effective
d
e
tection
met
hod
provide
d
by Matlab Imag
e Processin
g
Toolbox
function
s. It is base
d
on the optimization
algorith
m
and
it is less susceptibl
e
to no
ise interfe
r
e
n
c
e.
4.3. Mathem
atical Morph
o
log
y
Processing
Corro
s
io
n an
d inflation are
two of the mo
st fundam
en
tal operatio
ns in mathemati
c
al
morp
holo
g
y [8]. They can
be define
d
as:
A
a
b
B
b
A
a
a
B
A
,
,
,
b
a
c
B
b
A
a
c
B
A
,
,
(3)
The pro
c
e
s
s of Corro
s
ion befo
r
e
Inflation is calle
d o
pen ope
rati
on, which
can elimi
nate
small obje
c
ts.
This
p
r
o
c
e
ss
ca
n
al
so
sep
a
rate
th
e boun
dary of
object an
d
can
smooth
obj
ects
i
n
del
icate
pla
c
e
s
.
The pro
c
e
ss of op
eratio
n inflation bef
ore
Corro
s
i
on
is call
ed cl
osed operation t
hat can
fill
a
tiny
hollow of the
object. It not only
can
con
n
e
c
t the adjacent obje
c
ts but
can smooth
th
e
b
ound
ary of t
he o
b
ject,
which
ena
ble
the
fractu
re of co
ntour line to b
e
made up.
5. License Plate Chara
c
te
r Segmenta
ti
on
5.1. Hough T
r
ansformatio
n
After the li
cense pl
ate i
m
age
is lo
cated,
it som
e
times
ha
s different deg
ree
of
tilt
phen
omen
on.
In su
ch
ca
se, it need
s to
make ge
om
etric
co
rrectio
n
on th
e plat
es by
cla
s
sical
Hou
gh tran
sform. In th
e
Hough, th
e
curve or
strai
ght
line
with give
n shap
e in th
e ori
g
inal
ima
g
e
spa
c
e is
conv
erted into a p
o
int in the Hough sp
ac
e. Hough is to tra
n
sfor
m the de
tection proble
m
of curve o
r
st
raight line in
the Origin
al image
spa
c
e
to the peak p
o
int of the transfo
rm spa
c
e.
Hou
gh tran
sform lin
ea
r po
lar e
quatio
n i
s
exp
r
e
s
sed
as
sin
cos
y
x
,
x
y
arctan
.
,
refers to
pola
r
coordinate
s
vector
pe
rpe
ndicular to th
e line,
is the
length
of the
vector,
is
the angle b
e
twee
n the vector and the ax
is po
sitive direction.
5.2. Vertical Projection M
e
thod
In fact, the q
uality of the origin
al imag
e,
the ca
mera angl
e, the
effect of the
Hou
g
h
transform and
other
factors will
cause a
certai
n
degree of influence on th
e divided
character.
In
the image p
r
oce
s
sing, the
noise
can n
o
t be com
p
le
te
ly eliminated. So it is more effe
ctive to
detect the ch
ara
c
ter p
o
siti
on with vertical proje
c
tion
method.
After image
prep
ro
ce
ssi
ng, licen
se
plate locat
i
on and ext
r
actio
n
, and
rotation
transfo
rmatio
n, the image of licen
se plat
e is tr
eated b
y
morphol
ogy
processin
g
. Then it need
s to
record the
vertical
p
r
oje
c
tion of the
value
of the
binary i
m
ag
e
col
u
mn
hig
h
lights in tu
rn.
Acco
rdi
ng to the prio
ri kn
o
w
led
ge of the
licen
se plate
area a
nd p
r
o
j
ection m
e
tho
d
s, the lice
n
se
plate is
scan
ned b
a
sed o
n
a certain t
h
re
shol
d.
After the
previo
us p
r
o
c
e
ss,
we
che
c
k th
e
begin
n
ing
an
d the
end
po
sition
of ea
ch
ch
ara
c
te
r.
T
he ave
r
ag
e o
f
each
cha
r
a
c
ter wi
dth i
s
u
s
ed
to be the
wid
t
h of cha
r
a
c
t
e
r an
d to red
u
ce
noi
se im
pact. Fin
a
lly, the seve
n ch
ara
c
ters a
r
e
got
throng
cutting
from the start of the chara
c
ter to
the en
d of the avera
ge width of th
e cha
r
a
c
ter.
Evaluation Warning : The document was created with Spire.PDF for Python.
e-ISSN: 2
087-278X
TELKOM
NIKA
Vol. 11, No
. 11, Novemb
er 201
3: 671
4 – 6721
6718
6. License Plate Chara
c
te
r Reco
gnitio
n
6.1. Chara
c
ter Template
This a
r
ticle ta
ke
s a 24 x 4
8
JPG format
cha
r
a
c
ter im
age a
s
templ
a
te ,then re
spectively
set up Chi
n
e
s
e characte
rs ( the color
chara
c
te
r im
ag
e as sh
own in Figure 2), English letters
(as
sho
w
n in Fig
u
re 3
)
, and th
e Arabi
c num
eral (as
sho
w
n in Figure 4) characte
r dat
aba
se.
Figure 2. Chi
nese Ch
ara
c
t
e
rs
Figure 3. English Alpha
beti
c
Ch
ara
c
te
rs
Figure 4. Ara
b
ic Nume
ral
Cha
r
a
c
ters
6.2. Fuzzy
Pattern Chara
c
ter
Reco
gni
tion
Fuzzy patte
rn re
co
gnition
, it is mai
n
l
y
cla
ssified
by dire
ct m
e
thod
s a
c
cording to
maximum me
mbershi
p
prin
ciple fo
r indiv
i
dual ide
n
tification. For the
grou
p model
, it is identified
by indire
ct method
s and i
s
classified a
c
co
rding to "ch
oose the nea
rly principl
e".
Whe
n
the id
entified obje
c
t is not a sp
ecific
p
a
ttern,
but is a fuzzy set of do
main. the
identificatio
n
probl
em b
e
comes a
pro
b
lem of
solv
ing
close d
e
g
ree
bet
wee
n
the fu
zzy set
.Obviously, set cha
r
a
c
ter t
e
mplate a
s
a
catego
ry, an
d awaitin
g
re
cog
n
ition cha
r
acte
r a
s
obj
ect,
the ch
ara
c
te
rs may b
e
u
s
ed to dete
r
m
i
ne a fu
zzy
matrix. Then,
based o
n
th
e prin
cipl
es
of
cho
o
si
ng ne
a
r
ly, the close
ness is
comp
uted and
cate
gori
z
ed.
Acco
rdi
ng to
the characte
r feature
s
, givi
ng
ea
ch
calculated p
o
int o
f
binary im
ag
e some
weig
ht (as
sh
own in Fig
u
re
5 to provisio
n fuzzy matri
x
collectio
ns,
and the form
ula is:
1
,
1
1
,
1
1
,
1
1
,
1
05
.
0
,
1
,
1
1
,
1
,
1
.
0
,
4
.
0
,
~
j
i
BW
j
i
BW
j
i
BW
j
i
BW
j
i
BW
j
i
BW
j
i
BW
j
i
BW
j
i
BW
j
i
A
(4)
Figure 5. Fuzzy Matrix Cal
c
ulatio
n Wei
g
ht Diagram
Providing that
i
A
~
refers t
o
kno
w
n
cat
egory fu
zzy
sub
s
et withi
n
the universe of
discou
rse, if
obje
c
t to be identified meet the Formula
~
~
1
~
~
,
max
,
B
A
N
B
A
N
j
t
j
k
, then we
con
s
id
er that
~
B
is
c
l
os
es
t to
k
A
~
,
~
B
is cla
s
sif
i
ed t
o
k
A
~
pattern. Combined
with fuzz
y matrix
formula by u
s
ing the hamm
i
ng app
roa
c
h
degree
s its calcul
ation formula is:
n
j
m
i
k
k
j
i
B
j
i
A
mn
B
A
N
11
~
~
~
~
,
,
1
1
,
(5)
Evaluation Warning : The document was created with Spire.PDF for Python.
TELKOM
NIKA
e-ISSN:
2087
-278X
Visuali
z
ation
of Licen
s
e Pl
ate Re
cog
n
ition Syst
em
(Zhiyi Ruan
)
6719
6.3. Neural Net
w
o
r
k Ch
ar
acte
r
Re
cog
n
ition
The appli
c
ati
on of neural netwo
rk [9] in
im
age re
cog
n
ition acco
rdi
ng to data proce
s
sing
type can be
divided into the neu
ral n
e
t
work alg
o
rith
m based on
pixel data an
d feature d
a
ta.
Neu
r
al net
wo
rk recognition
technolo
g
y base
d
on the image pixel d
a
ta take
s the high dime
nsi
o
n
of the original
image data a
s
a neu
ral net
work trai
ning
sampl
e
s.
Some algo
rithm, su
ch a
s
forward feedb
ack
ada
ptive neural network, Hopfiel
d
ne
ural net
work,
RAM
neu
ral
n
e
twork, SO
FM ne
ural
network,
cell
ular
neural n
e
twork,
are
ba
se
d
on
the
pixel to identify image
s. First,
to construct a training sampl
e
set whi
c
h
also
con
s
titutes th
e inp
u
t ve
ctor and
the
targ
et vecto
r
of
th
e trai
ning
re
q
u
ired.
Acco
rdi
ng to
the
nee
ds
of identificatio
n to, the image pixel is pro
c
e
s
sed
by bi
nary figure, random
n
o
ise interferen
ce a
nd
morp
holo
g
ica
l
pro
c
e
ssi
ng.
The imag
e pixel value is
0 or 1, as n
eural
netwo
rk input vector
to
establi
s
h
corresp
ondi
ng training
set.Secon
d, to esta
blish a th
ree
-
layer fee
d
-fo
r
wa
rd n
e
two
r
k.
The feed fo
rward an
d ba
ckwa
rd BP n
eural
network
is teste
d
on
the ba
sis
of the inp
u
t and
the
target ve
ctor formed
by training
sam
p
l
e
s.Thi
r
d,
to t
e
st the
re
cog
n
ition of cha
r
acter data
b
a
s
e
plus n
o
ise image, then to realize the re
c
ognition of ch
ara
c
ter of po
sitioning cutting.
7. Visualizati
on of Identi
fication Sy
stem
Grap
hical
User Interfa
c
e
s
(GUI
) [10
-
12] is a
wa
y of huma
n
-comp
u
ter int
e
ra
ction
operation p
r
o
v
ided by the Matlab. It not only can
facilitate the user'
s
ope
ratio
n
, but also can
gene
rate exe
c
utabl
e file wi
thout installin
g Ma
tlab ru
nn
ing on Wi
ndo
ws o
perating
system
s.
Interface of
lice
n
se pla
t
e re
cog
n
itio
n sy
stem
consi
s
ts of t
he ima
ge, p
o
sitioni
ng
segm
entation
and pattern reco
gnition th
ree pa
rts.
It provides im
age
and di
splay
whi
c
h the u
s
ers
sele
ct to reco
gnize, image
licen
se plate
positio
ning segmentatio
n, and fuzzy pat
tern re
co
gniti
on
and ne
ural in
terface
netwo
rk
re
spe
c
tively. The recog
n
ition re
sult
s can b
e
outp
u
t as sho
w
n
in
Figure 6, the initial interface.
Figure 6. Initial Interface of
Recognitio
n
Syst
em
Figure 7. The
Positioning
Cutting Interf
ace of
Lice
nse Plate Recognitio
n
System
Figure 8. Re
cognition Syst
em Interface
Fi
gure 9. The
Result Saving Interface of
Re
cog
n
ition System
Evaluation Warning : The document was created with Spire.PDF for Python.
e-ISSN: 2
087-278X
TELKOM
NIKA
Vol. 11, No
. 11, Novemb
er 201
3: 671
4 – 6721
6720
Figure 7 sh
o
w
s the
syste
m
make
s the
licen
se plate
location
cutting on the im
age for
use
r
to
rea
d
.
Figure 8
sh
o
w
s the
syste
m
ma
ke
s fuzzy pattern re
cog
n
ition o
n
t
he
cutting li
cense
plate cha
r
a
c
ter a
nd di
spl
a
y results. Fig
u
re
9 sh
o
w
s
t
he sy
st
e
m
id
ent
if
ic
ation re
sults ba
sed o
n
use
r
nee
ds
to save text as txt format, and pr
ov
ides three result
s for ea
ch characte
r in
accordance
with the size of
the possibility of fuzzy patte
rn recognition, for example,
“E:\LPRS\Example1.jp
g” i
s
re
cog
n
ized
to be t
he license plate b
y
the fuzzy pattern. The
first
c
h
ar
ac
te
r
is
“
豫吉
苏
”;
the seco
nd cha
r
a
c
ter
i
s
“DQC”;
the
third
chara
c
te
r is “V97”; the fo
urth
c
h
arac
ter is
“0QD”; the fifth c
h
ar
ac
te
r is
“
0
86
”
;
the
s
i
xth
c
h
aracter is “085
”; the
sevent
h
cha
r
a
c
ter i
s
“172”. Th
us, the entire li
ce
nse pl
ate is
most likely to
be “
苏
D
V
0
0
01”
.
8. Conclusio
n
In this pap
er, it just pays attention to t
he licen
se
plate of the stand
ard
-
sma
ll cars.
Thro
ugh
digi
tal image
te
chn
o
logy
su
ch
as
the
i
m
age
gray
level, the e
dge
dete
c
tio
n
,
morp
holo
g
ica
l
pro
c
e
ssi
ng
and
Hou
gh transfo
rm, to
realize the im
age of li
cen
s
e plate lo
cati
on
and
cha
r
a
c
te
r segme
n
tation. Using th
e data of
ch
ara
c
ter te
mpl
a
te and
sel
e
ctively opera
t
e
con
c
u
r
rently betwe
en fuzzy model
an
d neu
ral net
work to reco
gnize Chi
n
e
s
e, English, a
nd
English
and
numbe
rs
of the licen
se
plate re
sp
e
c
tively. To make li
ce
nse plate lo
ca
tion,
segm
entation
and
re
co
gni
tion expe
rim
ents
by 16
i
m
age
s
colle
cted, but b
e
cause of th
e
big
differen
c
e i
n
backg
rou
nd i
m
age, the
r
e
are
one
cann
ot po
sitioning
, 2 pie
c
e
s
po
sitionin
g
mi
stake,
and the re
st of the image is able to su
cce
ssfully
loca
te and cut. We just get the cha
r
a
c
ters after
the fuzzy patt
e
rn recognitio
n
and the cha
r
acte
rs are
wi
thin the three
results p
r
ovid
ed.
For
other types of moto
rs, th
e reco
gnition
syste
m
ca
n al
so
reali
z
e
lice
n
se
plate
positio
ning cutting, and specifi
c
ation
s
.
In acco
rdan
ce with the licen
se plate
chara
c
te
risti
c
s and
cho
o
si
ng the
corre
s
po
ndi
ng ch
ara
c
te
r databa
se, the lice
n
se pl
ate can al
so
completely
be
recogni
ze
d. For some othe
r types of ca
r su
ch as th
e licen
se pl
ate of the coa
c
he
s “
某学
A·123
4
”
,
A large cont
ainer trucks licen
se plate
“
某挂
A·123
4
”
licen
se plat
e licen
se pla
t
e of consul
ate
“
某
领
O·123
4
”
,
we ju
st need
to identify the spe
c
ific characters in diffe
ren
c
e
s
.
As the
syste
m
discu
s
sed
in this p
ape
r
can
not only
realize the im
age
s of licen
se pl
ates
positioning, segmentation,
but
also
can facilitate the recogniti
on of
Chi
n
ese c
haracter. What’s
more
after a
m
ende
d, the
system
can
also
a
c
hiev
e the
cha
r
a
c
ter po
sitioni
n
g
, cutting
an
d
recognitio
n
cutting and
re
cog
n
ition of li
cen
s
e
pl
ate i
n
Ja
pan
ese
and
Ru
ssi
an,
by aiming
at
the
licen
se pl
ate spe
c
ification
s
spe
c
ific imp
r
ovem
ent syst
em in Jap
an, Ru
ssi
a and ot
her count
rie
s
.
Ackn
o
w
l
e
dg
ement
The resea
r
ch
wo
rk
wa
s j
o
intly sup
p
o
r
ted by
Fujia
n province Innovation P
r
actice of
Und
e
rg
ra
duat
e in Chin
a (G
rant No. 11
1Z
C12
60).
Referen
ces
[1]
Shan
Di
ng. T
he R
e
se
arch
of Lic
ense
Pl
ate
C
haracter
Reco
gn
ition
Based
on
Ne
ural
Net
w
o
r
k
Ensemb
le. Ma
ster'
s
degree t
hesis of Sha
n
d
ong N
o
rmal U
n
iversit
y
. 2
0
1
1
: 1-5 (in C
h
in
ese
)
.
[2]
Xi
ao
jin
g Z
h
ang
. Rese
arch
of
Vide
o B
a
sed
V
ehicl
e Detecti
o
n
a
n
d
Lic
ens
e Plate Rec
ogn
iti
on
S
y
stem
.
Master'
s
degre
e
thesis of T
a
i
y
uan U
n
ivers
i
t
y
of
T
e
chnol
og
y.
2011: 1-6 (i
n
Chin
ese).
[3]
Ibaa Jama
l, M Usman Akra
m, Anam T
a
ri
q.
Retina
l Image Prepr
ocess
i
ng: Backgr
o
u
nd an
d No
i
s
e
Segmentation.
T
E
LKOMNIKA Indon
esi
a
Jour
nal of Electric
al
Engin
eeri
ng.
2
012; 10(
3): 537
-543.
[4]
Xi
ng
hua H
e
, Yuan
yu
a
n
Z
h
o
u
, Ji
y
a
ng W
a
ng, etc.
Image Processi
ng
abo
ut MAT
L
AB7.X. Bei
jin
g:
Peop
le'
s
Posts and T
e
lecom
m
unic
a
tions Pr
ess. 2006 (i
n C
h
in
ese).
[5]
Yongc
ha
o Ch
e
n
.
T
he Rese
ar
ch of Lice
nse
Plat
e Rec
o
g
n
it
ion Bas
ed o
n
Digita
l
Image
Processi
ng.
Master'
s
degre
e
thesis of W
uhan U
n
ivers
i
t
y
of
T
e
chnol
og
y.
2006: 9-1
0
(in
Chin
ese).
[6]
Jingj
ia
o L
i
, Li
h
ong Z
h
ao, Ai
xi
a W
ang. P
a
ttern Rec
o
g
n
itio
n
.
Beiji
ng: Pu
bli
s
hin
g
Ho
use
o
f
Electronic
s
Industr
y
.
2
0
1
0
(in Chi
nes
e).
[7]
Jinmin
g S
u
, Y
ong
li W
a
ng. G
r
aph
ic Imag
e
abo
ut MA
T
L
AB. Beiji
ng: P
u
blish
i
n
g
H
ouse
of Electro
n
ics
Industr
y
.
2
0
0
5
(in Chi
nes
e).
[8]
Pei
y
o
u
Ha
n, Gui
y
un D
o
n
g
. Image T
e
chnol
og
y. Xi’a
n: No
rth
w
e
s
tern Po
l
y
tec
hnic
a
l U
n
i
v
ersit
y
Pr
ess.
200
9 (in Ch
in
e
s
e).
[9]
Patricia Me
lin,
Victor Herrer
a
, Dan
n
ie
la R
o
mero
, etc. Genetic Optim
i
z
a
tion
of Neur
a
l
Net
w
orks fo
r
Person R
e
cog
n
itio
n base
d
o
n
the Iris.
T
E
LKOMNIKA Indon
esia Jo
urn
a
l
of Electrical
Engin
eeri
n
g
.
201
2; 10(2): 30
9-31
8.
Evaluation Warning : The document was created with Spire.PDF for Python.
TELKOM
NIKA
e-ISSN:
2087
-278X
Visuali
z
ation
of Licen
s
e Pl
ate Re
cog
n
ition Syst
em
(Zhiyi Ruan
)
6721
[10]
Cha
o
Ch
en. A
pplic
atio
n E
x
a
m
ples S
u
ccinc
tl
y
a
b
o
u
t MAT
L
AB: Image P
r
ocessi
ng a
n
d
GUI Desig
n
.
Beiji
ng: Pu
blis
hin
g
Hous
e of Electron
ics Ind
u
str
y
. 20
11 (in
Chin
ese).
[11]
Z
heng
jun
L
i
u.
Scie
ntific C
o
mputin
g a
n
d
Simulati
on
a
b
out MAT
L
AB. Beiji
ng: P
u
b
lis
hin
g
H
ous
e of
Electron
ics Ind
u
str
y
. 20
04 (in
Chin
ese).
[12]
Hua
n
ji
n Li
u, Hui W
ang, Pe
ng
Li
, etc. N
T
i
ps about MAT
L
AB. Beijin
g:
Beij
i
ng Un
iversit
y
o
f
Aerona
utics
and Astron
auti
cs Press. 2011
(in Chi
nese).
Evaluation Warning : The document was created with Spire.PDF for Python.