Internati
o
nal
Journal of Ele
c
trical
and Computer
Engineering
(IJE
CE)
V
o
l.
4, N
o
. 2
,
A
p
r
il
201
4, p
p
.
27
8
~
28
4
I
S
SN
: 208
8-8
7
0
8
2
78
Jo
urn
a
l
h
o
me
pa
ge
: h
ttp
://iaesjo
u
r
na
l.com/
o
n
lin
e/ind
e
x.ph
p
/
IJECE
Implementation of Dynam
i
c Ti
me Warping Meth
od for the
Vehicl
e Number Licens
e
Recognit
ion
M
a
d
e
Su
da
rma
,
S
r
i A
r
iy
an
i
Department o
f
Electr
ical Engin
e
ering,
Faculty
of Engin
eering
Uday
an
a
University
, Bali, Indonesia
Article Info
A
B
STRAC
T
Article histo
r
y:
Received Nov 19, 2013
Rev
i
sed
Jan
4, 2
014
Accepte
d
Ja
n 28, 2014
In the era of inform
ation techno
log
y
veh
i
cl
e num
bers
identific
at
ion needs
to
be done b
y
s
y
stem automatically
.
Ther
efor
e, th
e accuracy
of
the d
a
ta
is well
documented
an
d work porses identif
ic
ation
can be done qu
ickly
.
Motor
vehicle license r
ecognition is a
recogni
tion s
y
stem b
y
comparin
g character
featur
e in
li
cens
e
pla
t
e wi
th r
e
fe
rence
fea
t
ure
which ex
ists in d
a
t
a
base.
This
s
y
s
t
em
us
es
ch
a
i
n cod
e
m
e
thod
and t
e
m
p
lat
e
m
a
tch
i
ng to
ex
tra
c
t
char
act
er
featur
e in
li
cens
e
pl
ate’s
im
age.
F
eatur
e ex
tra
c
ti
on with
chain
c
ode m
e
tho
d
will result in a
n
arra
y
o
f
dire
c
tion codes whic
h stored in d
y
n
a
m
i
c arr
a
y,
which stored
in
d
y
n
a
m
i
c arr
a
y. In th
is appl
ic
ation
test f
eatu
r
e will
be
m
a
tched with fe
ature stored in d
a
tab
a
se using d
y
namic time warping method
(DTW
) to obtain
a dis
t
an
ce va
lue
between
tes
t
fe
ature
and ref
e
re
nce fe
atur
e
,
the s
m
all
e
r th
e
dis
t
an
ce ob
tain
ed s
hows
that
both th
e fe
atur
e
s
are m
o
re
sim
ilar. Th
e res
u
lt of this s
y
s
t
e
m
is the recogn
ition of e
ach
c
h
arac
ter in
lic
ens
e
plat
e’s
i
m
a
ge. In this
s
t
ud
y, s
a
m
p
les
of licens
e
pl
at
e’s
im
ages
are
tes
t
ed w
ith
the n
u
m
b
er of res
e
arc
h
obje
c
ts
. F
r
om
the s
t
ud
y f
e
a
t
ure
extr
act
ion
is obtained wit
h
tem
p
late m
a
t
c
hing
m
e
thod provides bett
er success rat
e
compared to
feature ex
traction
with chain cod
e
method, wher
e the success
rate of feature extraction with templa
te matching
method is at 78
% whereas
featur
e
extra
c
t
i
o
n
with
cha
i
n cod
e
m
e
thod
is
a
t
6
8
%.
Keyword:
Dy
nam
i
c tim
e
war
p
in
g
Im
age m
a
tching
Im
age segm
entation
Tem
p
late matc
h
i
ng
Vehicle license
Copyright ©
201
4 Institut
e
o
f
Ad
vanced
Engin
eer
ing and S
c
i
e
nce.
All rights re
se
rve
d
.
Co
rresp
ond
i
ng
Autho
r
:
M
a
de S
u
darm
a
Jl
. Tu
ka
d
Ye
h
Ay
a N
o
.
4
6
,
D
e
npa
sar
8
0
2
2
5
,
B
a
l
i
- I
n
do
nes
i
a
Telp
./Fax
.
: +62
361
795
680
0 / +62
361
257
055
Em
a
il: su
d
a
rma@ee.u
nud
.ac.id
1.
IN
TR
OD
UCTION
Dig
ital i
m
ag
e p
r
o
cessi
n
g
curren
tly no
t on
ly rang
es
b
e
tween
d
i
g
ital i
m
ag
e ed
itin
g
b
y
usin
g
ex
isting
filter effects, bu
t also
co
v
e
ring
p
a
ttern
recogn
itio
n
tech
n
i
que au
to
m
a
t
i
call
y
su
ch
as th
e pattern
recogn
itio
n
o
f
face,
fingerpri
n
t, ha
nd
writing and c
h
aracte
r
pattern
of
pri
n
ting
result [1]
.
Ge
nerally pat
t
ern recognition is a
sci
e
nce t
o
cl
assi
fy
or d
r
awi
n
g som
e
t
h
i
ng b
a
sed o
n
q
u
an
ti
tative
m
easurement of feat
u
r
e or m
a
i
n
pr
op
ert
y
of
an
ob
j
ect. Th
e
p
a
ttern itself is an en
tity wh
ich
is
d
e
fi
n
e
d
and
can
b
e
d
e
fined
and
n
a
m
e
d
.
Th
e p
a
ttern
can
b
e
a
collection of
measurem
ent results or
ob
ser
v
at
i
on a
n
d can
be st
at
ed i
n
v
ect
or
not
at
i
o
n
or m
a
t
r
i
x
[5]
.
In t
h
i
s
research
will b
e
d
i
scu
ssed
reg
a
rd
ing
m
o
to
r v
e
h
i
cle
licen
se p
l
ate recogn
itio
n
o
n
d
i
g
ital i
m
ag
e, wh
ere th
e
syste
m
is ex
p
ected
to
b
e
ab
l
e
to
reco
gn
ize letter a
nd
num
b
er character containe
d in
m
o
tor vehicle
license
plate’s im
age. Recognizing
m
o
tor ve
hicl
e
license plate i
s
indispe
n
sa
ble in t
h
e sec
u
rity syste
m
of
parki
n
g
area, track
i
n
g
a
m
o
to
r v
e
h
i
cl
e an
d
id
en
tifyin
g
a m
o
to
r v
e
hicle [3
]. Each
v
e
h
i
cle h
a
s an
id
en
tity in
th
e
form
o
f
m
o
to
r v
e
h
i
cle licen
se
p
l
ate wh
ich
is legally issu
ed
by the state.
License
plate is
also called vehicle
registration
pla
t
e, or in United States is known as a li
cense
plate [4]. The shape c
o
nstitutes a piece
of m
e
t
a
l or
pl
ast
i
c
m
ount
ed
on
m
o
t
o
r
v
e
hi
cl
e as a
f
o
rm
al
i
d
ent
i
f
i
cat
i
on
put
i
n
t
h
e fr
o
n
t
or
rea
r
of
a
vehi
cl
e
.
T
h
e
u
n
i
q
u
e
n
e
ss o
f
th
is
licen
se p
l
ate
th
at
m
a
k
i
n
g
th
is p
l
ate wid
e
ly u
s
ed
as an
id
en
tity in
v
a
rio
u
s
system
s su
ch
as
p
a
rk
ing
system
,
b
u
ild
ing
secu
rity syste
m
,
to
ll syste
m
a
nd so o
n
,
but
o
f
t
e
n t
h
ere i
s
a mistake in recognition
Evaluation Warning : The document was created with Spire.PDF for Python.
I
S
SN
:
2
088
-87
08
IJEC
E V
o
l
.
4, No
. 2, A
p
ri
l
20
14
:
27
8 – 2
8
4
27
9
sin
ce th
e curren
t
ex
isting
syste
m
s stil
l
m
o
stly u
s
ing
m
a
n
u
al syste
m
n
a
mely th
e reco
rd
i
n
g of m
o
to
r
veh
i
cle
l
i
cense pl
at
e c
o
n
d
u
ct
ed
by
t
h
e of
fi
cer i
n
or
d
e
r t
o
i
d
ent
i
f
y
t
h
e
vehi
cl
e
[2]
.
Thi
s
m
e
t
hod
h
a
s a wea
k
ness t
h
at
p
u
t
i
n
h
u
m
a
n. H
u
m
a
n has
a n
a
t
u
re
o
f
qui
c
k
l
y
b
o
re
d a
n
d t
i
r
e
d
s
o
t
h
at
e
a
sy
t
o
m
a
ke
m
i
st
akes, m
o
re
ove
r
t
y
pi
ng
process
also
re
qui
res a l
o
nge
r
tim
e
.
License
plate has se
rial num
b
er
t
h
at
i
s
the ar
ra
ngem
e
nt
of l
e
t
t
e
rs a
n
d n
u
m
b
ers
de
vot
e
d
t
o
t
h
at
vehi
cl
e.
Thi
s
num
ber i
n
In
d
one
si
a i
s
cal
l
e
d
pol
i
ce
n
u
m
b
er, a
n
d ca
n
be
i
n
t
e
g
r
at
ed
wi
t
h
ot
he
r i
n
f
o
rm
at
i
o
n
rega
rdi
ng that vehicle, s
u
ch
as color, brand, m
odel,
y
ear of m
a
nufact
u
r
e, ve
hi
cle identification num
ber or
VI
N a
nd
o
f
c
o
u
r
se t
h
e nam
e
an
d ad
d
r
ess
of t
h
e
ow
ne
r
[9]
.
Al
l
t
h
ese
dat
a
al
so l
i
s
t
e
d i
n
m
o
t
o
r
v
e
hi
cl
e
reg
i
stration
letter
wh
ich
is ev
id
en
ce letter th
at
police
num
be
r is s
p
ecified for that
vehicle.
2.
RESE
AR
CH
METHO
D
A
N
D
DI
SC
US
S
I
ON
The objectives
expected
from
th
is research
p
r
ep
aratio
n
t
o
find
ou
t th
e stag
es in
d
e
si
gn
ing
a
m
o
to
r
v
e
h
i
cle license p
l
ate
recogn
itio
n system
in
a d
i
g
ital i
m
ag
e, to fi
n
d
ou
t
th
e
p
e
rform
a
n
ce of m
o
to
r veh
i
cle
licen
se p
l
ate reco
gn
itio
n syste
m
in
a d
i
g
ital i
m
ag
e. Fu
rth
e
rem
o
re, th
e research
prep
aratio
n
is u
s
i
n
g
sev
e
ral
assu
m
p
tio
n
s
with
an
obj
ective th
at th
e d
i
scu
ssion
can
b
e
m
o
re d
i
rected an
d
t
o
sim
p
li
fy an
d
lim
it
in
g
th
e
problem
s
. The resea
r
ch is c
o
nducted
by
using m
o
to
r
vehicle license
plate’s im
age which e
x
peri
encing
pre
p
rocessi
ng
so t
h
at a
p
propriate m
o
tor ve
hicle license
plate area’s
imag
e is ob
tain
ed
t
o
b
e
seg
m
e
n
ted
.
Segm
ent
a
t
i
on
pr
ocess i
s
c
o
n
duct
e
d
by
way
of
t
r
acki
n
g ea
ch
pi
xel
o
n
t
h
e im
age t
o
fi
n
d
out
t
h
e
wi
dt
h a
nd
hei
g
ht
o
f
t
h
e c
h
aract
er
so t
h
at
pr
od
uci
n
g
go
o
d
o
u
t
p
ut
t
o
b
e
fu
rt
he
r cha
r
act
er rec
o
g
n
i
t
i
on
pr
ocess i
s
per
f
o
rm
ed
[6
].
In
th
is mo
tor
v
e
h
i
cle li
cen
se
p
l
ate reco
gn
itio
n, t
h
e
research
m
a
terial co
ndu
cted
is in
clud
ing
t
h
ree
(3)
things suc
h
as the stages in perfor
m
i
ng cha
r
acter segm
entation, feat
ure e
x
t
r
act
i
on p
r
oces
s i
s
usi
ng c
h
ai
n co
de
m
e
thod a
nd te
m
p
late
m
a
tching t
o
obtain c
h
aracteristic
di
ffe
rentiator from
each segm
ented c
h
aracte
r
,
and
characte
r
recognition by m
a
t
c
hing each
ext
r
action re
sult of c
h
aracter fe
at
ure from
previous stage by
using
refe
rence
data
b
a
se.
In
t
h
e
p
r
o
g
ram alg
o
rith
m
sect
io
n
,
will b
e
d
i
scu
ssed
reg
a
rd
i
n
g th
e
p
r
o
cesses o
c
cu
rring
in th
e system
,
whe
r
e the
processes ha
ving
im
portant
relat
i
on
with each other. As
for th
e process
used in this
res
earch,
nam
e
ly charact
er se
gm
entatio
n
process aim
s
to
perform
separat
i
o
n
bet
w
ee
n
ob
ject
s i
n
o
r
der t
o
get
t
h
e
d
e
si
red
object,
feature extraction proc
ess
from
each
characte
r
that
being se
gm
ente
d
by
using c
h
ain code
m
e
th
od a
nd
t
e
m
p
l
a
t
e
m
a
t
c
h
i
ng
w
h
i
c
h
f
u
rt
her
use
d
f
o
r c
h
aract
er m
a
t
c
hi
ng
st
age
,
a
n
d
c
h
aract
er
m
a
t
c
hi
ng
p
r
oces
s.
Im
ag
e Prepro
cessin
g
is raw
d
a
ta inp
u
t
t
r
ansform
a
tio
n
to
assist co
m
p
u
t
atio
n
a
l ab
ility an
d feat
u
r
e
seeker a
nd to reduce noise. In preproces
sing, im
ag
e (signal) that bei
ng ca
ptured
by a censor wi
ll be
n
o
rm
alized
so
th
at th
e i
m
ag
e will b
e
m
o
re prep
ared
t
o
b
e
pro
cessed
in
the stag
e o
f
featu
r
e sep
a
ration
[8]. Th
e
q
u
a
lity of featu
r
e th
at
b
e
ing
p
r
o
d
u
c
ed
i
n
th
e
f
eature sep
a
ratio
n pro
cess is so
m
u
ch
d
e
p
e
nden
t
on
pre
p
rocessi
ng result.
To
obtain a
gra
y
scale im
age the
form
ula use
d
is:
,
∝
(1
)
with
I (x,
y)
is
th
e grayscale lev
e
l in
a coo
r
din
a
te ob
tain
ed
b
y
settin
g
co
lor co
m
p
o
s
ition
o
f
R (red
)
,
G
(g
reen
)
,
and
B (
b
lue)
p
r
esent
e
d
by
pa
ram
e
t
e
r val
u
es
o
f
α
,
β
,
an
d
γ
. Gen
e
rally th
e v
a
lue
o
f
α
,
β
,
and
γ
i
s
0.
3
3
.
Ot
he
r
val
u
e al
s
o
ca
n
be
gi
ve
n
fo
r t
h
e t
h
ree
pa
ram
e
t
e
rs p
r
ovi
ded
t
h
at
t
h
e t
o
t
a
l
o
f
ove
ral
l
val
u
es i
s
1.
Th
resho
l
d
i
n
g
pro
cess will resu
lt in
a b
i
n
a
ry
i
m
ag
e, th
e i
m
a
g
e h
a
v
i
ng
two
v
a
lu
es
o
f
g
r
ayscale lev
e
l, b
l
ack
and
whi
t
e
.
Ge
neral
l
y
t
h
e t
h
res
h
ol
d
pr
ocess
o
f
gr
ay
scal
e im
age to produce
binary im
age is as follows:
,
1
,
0
,
(2
)
with
g (
x
, y)
a
s
binary im
age of grayscale image
f (x, y
)
, and
T
t
o
st
at
e t
h
res
hol
d val
u
e.
T
val
u
e has a
very
i
m
p
o
r
tan
t
ro
le
in
thresho
l
d pro
cess. Resu
lt
qu
ality o
f
b
i
n
a
ry i
m
ag
e is so
m
u
ch
d
e
p
e
nd
en
t on
T
value used.
There
are
t
w
o
t
y
pes
of
t
h
re
shol
di
n
g
,
gl
ob
al
t
h
res
h
ol
di
n
g
a
n
d l
o
cal
l
y
ada
p
t
i
v
e t
h
re
shol
di
n
g
.
I
n
g
l
oba
l
th
resh
o
l
d
i
ng
, all th
e p
i
x
e
ls on th
e i
m
ag
e are co
nv
erted
to
beco
m
e
b
l
ack
or wh
ite with
on
e th
resho
l
d
valu
e o
f
T
. Prob
ab
ly in
g
l
ob
al th
resho
l
d
i
ng
th
ere will b
e
a lo
t
o
f
info
rm
atio
n
lo
st du
e to
u
s
ing
o
n
l
y o
n
e
v
a
lu
e
o
f
T
for
ove
ral
l
pi
xel
s
.
To
ha
ndl
e t
h
i
s
p
r
o
b
l
e
m
i
t
can
use l
o
cal
l
y
ada
p
t
i
v
e t
h
re
shol
di
n
g
.
I
n
l
o
cal
t
h
resh
ol
di
ng
, a
n
im
age i
s
di
vi
d
e
d i
n
t
o
sm
al
l
bl
oc
ks an
d t
h
e
n
l
o
cal
t
h
re
sh
o
l
di
ng i
s
per
f
o
r
m
e
d on eac
h b
l
ock
wi
t
h
di
f
f
e
r
ent
T
val
u
e.
Dot
detection i
s
olated
from
an im
age in pri
n
ciple
occu
rr
e
d
strai
ght
fo
rwa
r
dly
.
We ca
n s
a
y
that a d
o
t
to
b
e
said iso
l
ated
if:
Evaluation Warning : The document was created with Spire.PDF for Python.
I
J
ECE
I
S
SN
:
208
8-8
7
0
8
Po
lice Num
b
er Licen
se Recogn
itio
n
u
s
i
n
g Dyn
a
m
i
c Ti
m
e
Wa
rp
ing
Met
ho
d (Mad
e
S
udarm
a
)
28
0
|
|
(3
)
whe
r
e
T
is
p
o
sitiv
e th
resho
l
d
an
d R is th
e v
a
lu
e of equ
a
tio
n:
∑
(4
)
Thu
s
, th
e iso
l
ated
do
t is a d
i
fferen
t
do
t (sign
i
fican
tly) wi
t
h
the dots around it. Line
detection of an i
m
age is
per
f
o
r
m
e
d by
m
a
t
c
hi
ng i
t
wi
t
h
m
a
sk and s
h
o
w
s a cert
a
i
n
part
w
h
i
c
h i
s
di
ffe
red i
n
a
st
rai
ght
l
i
n
e w
h
et
he
r
v
e
rtically, ho
ri
zo
n
t
ally, or lean
ing
4
5
0
(eith
er ri
g
h
t
or
left).
Mathem
a
tically can
be
form
ulated as follows:
|
|
|
|
(5
)
The direction of
im
age’s
e
d
ge
is varie
d
. There
is
a
straight a
n
d there is like a c
u
rve. T
h
e
r
e are
vari
ous
m
e
t
hods
o
f
e
d
ge
det
ect
i
on t
h
at
can
be
use
d
t
o
det
ect
va
ri
o
u
s t
y
pes
o
f
ed
ge. Eac
h
t
e
chni
que
ha
s i
t
s
o
w
n
adva
nt
age
.
O
n
e edge
det
ect
i
o
n t
ech
ni
q
u
e m
a
y
be wo
r
k
we
l
l
i
n
one cert
a
i
n
ap
pl
i
cat
i
on
b
u
t
o
n
t
h
e co
nt
r
a
ry
i
t
m
a
y
not
wor
k
opt
i
m
all
y
i
n
ot
he
r ap
pl
i
cat
ion
.
Ed
ge det
e
ct
i
on i
s
a pr
o
cess t
o
fi
n
d
o
u
t
ob
vi
o
u
s
di
f
f
ere
n
t
in
ten
s
ity ch
anges in
an
im
ag
e
’
s sectio
n. An
ed
g
e
d
e
tectio
n o
p
e
r
a
tor
is a co
n
tiguo
us/n
eigh
bor
hoo
d
o
p
e
ratio
n
,
n
a
m
e
l
y
an
o
p
er
atio
n
that m
o
d
i
f
i
es g
r
ay v
a
lu
e of
a
d
o
t
b
a
sed
on g
r
ay v
a
l
u
es
o
f
d
o
t
s ar
ound
it (
its
neighborhood) each ha
vi
ng i
t
s own weig
ht. The
weights’ values a
r
e de
pe
nde
d
on
operation t
h
at wi
ll be
perform
e
d, whereas the
am
ount
of ne
i
g
hborhood’s dots
i
n
volve
d
us
ually
is
2x2
,
3x
3,
3x4
,
7x7
,
an
d so on
.
Usual
l
y
t
h
e
o
p
erat
or
use
d
t
o
det
ect
t
h
e
fi
rst
e
dge
i
s
op
erat
or
based
o
n
gra
d
i
e
nt
(
f
i
r
st
de
ri
vat
i
o
n),
nam
e
ly
ro
bert
o
p
e
r
at
o
r
, so
bel
ope
rat
o
r
,
an
d p
r
ewi
t
t
operat
o
r
.
Th
e secon
d
i
s
o
p
e
rat
o
r
base
d o
n
seco
n
d
de
ri
vat
i
o
n
,
nam
e
ly Laplacian operat
or
and La
placian
G
a
ussi
an
o
p
erat
or
. C
h
ai
n co
de
i
s
oft
e
n
used
t
o
descri
be o
r
e
n
co
d
e
the cont
our of
an obje
ct. The
estab
lish
m
en
t
o
f
th
e ch
ain
cod
e
is startin
g
with
sp
ecifyi
n
g
th
e first p
i
x
e
l fro
m
an
ob
ject
.
B
a
s
e
d
on
t
h
e
pi
xel
o
b
j
ect
chai
n c
ode
i
s
est
a
bl
i
s
hed
by
f
o
l
l
o
wi
ng
t
h
e
di
rect
i
o
n
rul
e
o
f
c
h
ai
n
co
de.
Based
on t
h
e
chain c
o
de, t
h
e analysis to a
n
object ca
n b
e
d
one
by
cal
c
u
l
a
t
i
ng t
h
e
per
i
m
e
t
e
r, area, a
n
d
t
h
e
fo
rm
/shape fac
t
or.
Th
e
p
e
rim
e
ter states th
e leng
t
h
o
f
th
e
fram
e
p
r
od
u
c
ed
. Peri
meter is calcu
lated
with th
e fo
rm
u
l
a as fo
llows:
√
2
.
(6
)
For
c
h
ai
n c
o
de
o
f
07
7
0
7
6
4
5
5
45
3
0
1
2
3
3
4
20
1 st
at
e
d
a
b
o
v
e, t
h
e
fram
e
l
e
ngt
h i
s
:
1
0
1
1
√
2
2
5.56
The i
n
cluding
of
√
2
fact
o
r
in d
e
term
in
in
g
of
P
i
n
o
d
d
c
o
d
e
, beca
use
o
d
d
co
de
has
di
ag
onal
di
rect
i
o
n.
Th
e
calculation of area based
on
c
h
ain
c
o
de ca
n
be stated as
fol
l
ows:
C
ode
0:
Area
= Ar
ea +
Y
;
C
ode
1:
Area
= Ar
ea +
(Y
+
0.
5)
C
ode
2:
Area
= Ar
ea
;
C
ode
3:
Area
= Ar
ea
– (
Y
+
0.
5)
C
ode
4:
Area
=
Ar
ea
Y
;
C
ode
5:
Area
=
Ar
ea –
(
Y
– 0.
5)
C
ode
6:
Area
=
Ar
ea
;
C
ode
7:
Area
=
Ar
ea
+ (Y
–
0.
5)
Sha
p
e
fact
or
i
s
de
fi
ne
d as
f
o
l
l
o
ws:
(7
)
Since
S
is a ratio
b
e
tween
p
e
rim
e
ter an
d
area th
en
S do
es n
o
t
p
r
esen
t a d
i
m
e
n
s
io
n
qu
an
tity, so
th
at
S
has
in
v
a
rian
t
p
r
operties to
t
h
e scale,
ro
tatio
n, an
d tr
an
slatio
n
,
wh
ich con
s
titu
tes a v
e
ry usefu
l
feature
characte
r
istic.
Tem
p
late
matc
h
i
ng
is a p
r
o
c
ess to
find
an
ob
j
ect in
t
h
e whole objects i
n
side an im
age. Te
m
p
late is co
mpare
d
with
th
e who
l
e ob
j
ects and
if th
e tem
p
late
is fit with an un
kno
wn
ob
ject in
th
e im
a
g
e th
en
t
h
e
o
b
j
ect
i
s
marked as a te
m
p
la
te. The com
p
arison bet
w
een tem
p
late
and the whole objects in
t
h
e image can be
done
by
calculating
t
h
e diffe
re
nce of
t
h
e distance,
as follows:
,
∑∑
,
.
2
(8
)
Evaluation Warning : The document was created with Spire.PDF for Python.
I
S
SN
:
2
088
-87
08
IJEC
E V
o
l
.
4, No
. 2, A
p
ri
l
20
14
:
27
8 – 2
8
4
28
1
Wi
t
h
f
(
j
,k
)
stati
n
g
th
e im
ag
e wh
ere th
e object lo
cated
which
will b
e
com
p
ared
with
te
m
p
late
T
(
j
,k
)
whe
r
eas
D
(
m
,n
)
sta
ting the
distance
be
tween tem
p
late and object on th
e im
ag
e. In
g
e
n
e
ral th
e
size o
f
tem
p
late is far
sm
a
ller fro
m
i
m
ag
e’s size. Id
eally, te
m
p
lat
e
is to
b
e
said
m
a
tch
i
n
g
wit
h
obj
ect on
the i
m
ag
e if
D
(
m
,n)
= 0,
h
o
wev
e
r con
d
i
tio
n
lik
e t
h
is is d
i
fficu
lt to
b
e
fu
lfilled
let
alon
e if tem
p
late i
s
a gray
scale imag
e. Therefore, t
h
e
ru
le
b
e
ing
u
s
ed to
state th
at a te
m
p
late is
m
a
t
c
h
i
ng
with
o
b
ject is:
,
,
(9
)
with
L
D(
m,n
)
i
s
a t
h
res
h
ol
d
val
u
e.
Dyn
a
m
i
c Ti
me
W
a
rp
ing
(DTW
) is a m
e
t
h
od
to
calcu
l
ate th
e d
i
stan
ce b
e
tween
two
ti
m
e
series d
a
ta. The
adva
nt
age
o
f
DTW
from
other distance
m
e
thods
is its ability
to
calculate
the distance betwee
n two dat
a
vecto
r
s with
di
ffe
rent
length. The distance
of
DT
W
bet
w
ee
n t
w
o vect
ors i
s
cal
cul
a
t
e
d f
r
o
m
opt
im
al
warpi
n
g
p
a
th
o
f
bo
th
v
ectors.
Of sev
e
ral techn
i
ques u
s
ed
to
cal
cu
late
DTW
, th
e m
o
st reliab
l
e on
e is d
y
n
a
m
i
c
pr
o
g
ram
m
i
ng m
e
t
hod.
T
h
e di
st
ance of
DT
W
can
b
e
calcu
lated
with fo
rm
u
l
a as fo
llo
ws:
,
,
(1
0)
,
,
1
,
1,
1
,
1
(1
1)
0,
0
0
,
0,
∞
∞
,
0,0
0
,
∞,
0
∞
(1
2)
1
,2,3
…
;
1
,2,
3
…
The
value
in t
h
e colum
n
of
(i, j
)
i
s
seen as
ad
di
t
i
on
val
u
e
o
f
war
p
i
n
g
pat
h
f
r
om
t
h
e col
u
m
n
of
(1
,
1
)
u
n
til
(i,
j
)
.
Co
lu
m
n
with
t
h
e
v
a
lu
e
o
f
γ
(i
,
j
)
(
1
< i <
m,
1 <
j <
n)
is c
a
lled
su
mm
ed
d
i
stan
ce m
a
tri
x
. Th
e
fo
llowi
n
g
is th
e
exam
ple of s
u
mmed distance
m
a
trix.
3.
R
E
SU
LTS AN
D ANA
LY
SIS
3.
1.
S
y
ste
m
T
e
sti
n
g
Th
e testin
g
o
f
su
ccess rate fro
m
m
o
to
r v
e
h
i
cle licen
se p
l
ate recog
n
ition
syste
m
is co
n
ducted
b
y
way
o
f
co
m
p
aring
t
h
e
resu
lt ob
tain
ed
b
y
th
e syste
m
(ob
j
ecti
v
e) with
t
h
e resu
lt o
f
our reason
ing
itself (su
b
j
e
ctiv
e).
Reco
gn
itio
n syste
m
wh
ich
i
s
b
e
i
n
g m
a
d
e
is tested
wit
h
5
0
im
ag
es an
d 38
1 ch
aracter ob
j
ects.
Wh
ere th
e
outc
o
m
e
of re
cognition system
is producing t
w
o
ou
tput
s nam
e
ly recognition outc
om
e with chai
n code
m
e
t
hod a
nd t
e
m
p
l
a
t
e
m
a
t
c
hing m
e
t
hod
. Sy
st
em
t
e
sti
ng i
n
cl
udes t
h
ree t
h
i
ngs
, nam
e
l
y
segm
ent
a
t
i
on, c
h
ai
n
co
d
e
feat
u
r
e reco
gn
itio
n, and
te
m
p
late
m
a
tch
i
n
g
feature reco
gn
itio
n.
3.
2.
Ob
ject
Se
par
a
ti
on
Thi
s
res
u
l
t
t
e
st
prese
n
t
e
d i
n
Tabl
e 1 a
n
d Fi
gu
re
1, i
s
co
n
duct
e
d t
o
fi
n
d
out
sy
st
em
success rat
e
i
n
anal
y
z
i
ng
ob
je
ct
i
n
t
e
gri
t
y
i
n
t
h
e im
age by
di
vi
di
ng eac
h o
b
j
ect in
th
e i
m
ag
e in
to
its own spa
ce.
Object
sep
a
ration
process will sim
p
l
i
fy recogn
itio
n p
r
o
cess, sin
ce th
is process i
s
on
e
o
f
d
e
termin
an
t factors in
th
e
success of this
recogn
ition syste
m
[10].
Tabl
e
1. T
h
e
v
a
ri
at
i
on
of
t
h
re
shol
di
n
g
val
u
e
s
an
d
wi
dt
h re
f
e
rences
.
No
T
h
r
e
sholding
W
i
dth Refer
e
nce
Seg
m
entation (%)
1
95
5
57
2
95
10
80
3
95
15
79
4
127
5
66
5
127
10
93
6
127
15
94
7
159
5
66
8
159
10
85
9
159
15
86
Evaluation Warning : The document was created with Spire.PDF for Python.
I
J
ECE
I
S
SN
:
208
8-8
7
0
8
Po
lice Num
b
er Licen
se Recogn
itio
n
u
s
i
n
g Dyn
a
m
i
c Ti
m
e
Wa
rp
ing
Met
ho
d (Mad
e
S
udarm
a
)
28
2
Th
e test is co
nd
u
c
ted
with
sev
e
ral test im
ag
es and
w
i
t
h
i
m
age t
h
res
h
ol
di
n
g
param
e
t
e
r val
u
es
an
d
di
ffe
rent
wid
t
h
referen
c
es. The seg
m
en
tatio
n
resu
lt o
f
t
h
e syst
em
will b
e
co
m
p
ared
with
o
b
j
e
ctiv
e reaso
n
i
n
g
. Th
e
success rat
e
of
t
h
e sy
st
em
i
n
segm
ent
a
t
i
on st
age wi
t
h
t
h
e
v
a
ri
at
i
on o
f
t
h
re
shol
di
n
g
val
u
e
s
and
di
ffe
re
nt
wi
dt
h
referen
ces is sh
own
in th
e
fo
llo
wing
graph
i
c fro
m
th
e d
a
ta
o
f
test resu
lt.
Fi
gu
re
1.
The
gra
p
hi
cs o
f
se
g
m
ent
a
t
i
on t
e
st
3.
3. Ch
ai
n C
o
de
Fe
ature
Re
cog
n
i
t
i
o
n
Test resu
lt
o
f
t
h
is reco
gn
ition syste
m
will b
e
co
m
p
ared
with
th
e resu
lt of o
b
j
ectiv
e
reaso
n
i
n
g
. Test
resu
lt
o
f
ch
ai
n
co
d
e
feat
u
r
e
reco
gn
itio
n system
with
DTW
can
b
e
seen
in th
e tab
l
e
b
e
low. Th
e
fo
llo
wi
n
g
is the
gra
p
hics that
present t
h
e s
u
cc
ess rate
of the
syste
m
in
recognizing c
h
a
r
ac
ter with chai
n
code
feat
ure
,
whe
r
e
m
a
t
c
hi
ng p
r
oc
ess i
s
usi
n
g
DTW
. Fro
m
th
e d
a
ta in
th
e
graph
i
c of test resu
lt shows i
n
Figu
re
2
,
can
b
e
co
n
c
l
u
d
e
d
t
h
at th
e reco
gn
itio
n with
ch
ai
n
cod
e
feat
u
r
e,
where m
a
tch
i
n
g
pro
cess is u
s
i
n
g
DTW
ha
ving
s
u
ccess
rat
e
o
f
67
% a
n
d
reco
g
n
i
t
i
on e
r
r
o
r
3
3
%.
T
h
e
err
o
r
in c
h
a
r
ac
ter rec
o
g
n
itio
n
is d
u
e t
o
se
ver
a
l refe
rence
fe
atures
h
a
v
i
n
g
sim
i
larities with
test
featu
r
es, so
t
h
at sev
e
ra
l
errors
o
ccurred
i
n
ch
aracter
reco
gn
itio
n pro
c
ess. Th
e
failures also occurred
in
t
h
e stage
of
c
h
ain
code
feat
ure e
x
traction, syste
m
failu
re in
featu
r
e ex
traction
stage
is at 1% a
n
d s
u
ccess rate at
99%.
Fi
gu
re
2.
The
gra
p
hi
cs o
f
rec
o
g
n
i
z
i
n
g c
h
ai
n
co
de
feat
ur
e w
i
t
h
DT
W
.
3.
4.
Tem
p
la
te Ma
tchin
g
Fe
a
t
ure Rec
o
g
n
ition
Test resu
lt
o
f
t
h
is reco
gn
ition syste
m
will b
e
co
m
p
ared
with
th
e resu
lt of o
b
j
ectiv
e
reaso
n
i
n
g
. Test
resu
lt of te
m
p
late
m
a
tch
i
n
g
featu
r
e
recognitio
n
syste
m
with
DTW
ca
n be see
n
i
n
t
h
e t
a
bl
e bel
o
w. T
h
e
fo
llowing
is the g
r
ap
h
i
cs
o
f
test resu
lt d
a
ta th
at p
r
esen
t
the success
rate of the
syst
e
m
in r
ecognizing cha
r
acter
with
tem
p
late match
i
n
g
feature,
wh
ere m
a
tch
i
ng
pr
ocess
i
s
usi
n
g
dy
nam
i
c t
i
m
e
warpi
n
g
.
Evaluation Warning : The document was created with Spire.PDF for Python.
I
S
SN
:
2
088
-87
08
IJEC
E V
o
l
.
4, No
. 2, A
p
ri
l
20
14
:
27
8 – 2
8
4
28
3
Fi
gu
re
3.
The
gra
p
hi
cs o
f
sy
s
t
em
t
e
st
m
a
t
c
hi
ng
feat
ure
wi
t
h
DTW
.
Fro
m
th
e Figure 3, co
n
c
l
u
d
e
d
th
at the recog
n
ition
with
tem
p
la
te
m
a
tch
i
n
g
feature
o
f
78
% an
d and
2
2
% recog
n
itio
n error. Th
e erro
r in
ch
aracter reco
gn
itio
n is du
e t
o
sev
e
ral
referen
ce feat
u
r
es hav
i
ng
si
m
ilarit
i
es with
test feat
u
r
es,
so
t
h
at sev
e
ra
l
erro
rs o
c
curred in
ch
aracter reco
gn
itio
n pro
c
ess.
3.
5. Resul
t
An
al
ysi
s
The
result of t
h
is threshol
ding
pr
ocess will
be c
r
ucial i
n
success rate
of t
h
e system
in segm
entation
p
r
o
cess, feat
ure ex
tractio
n
an
d
reco
gn
ition
.
For ch
ar
acter reco
gn
ition
with ch
ai
n
co
d
e
feature,
wh
ere
m
a
t
c
hi
ng p
r
oc
ess i
s
usi
ng
DTW
has a s
u
ccess rate of 67% and rec
o
gnition error of 33%. Cha
r
acter
recognition
with chai
n code fe
ature is
also e
x
perie
n
cing the
failure in
feat
ure extraction,
whe
r
e system
f
a
ilure
for feature
ext
r
action process
is
1%.
Th
e failure in
ch
aracter recog
n
ition
with
ch
ain
cod
e
f
eat
u
r
e is du
e to
2 facto
r
s, first is th
e failu
re i
n
feat
ure e
x
t
r
act
i
on a
nd se
co
n
d
i
s
an e
r
r
o
r i
n
pe
rf
orm
i
ng
reco
g
n
i
t
i
on.
A
n
er
ro
r i
n
c
h
a
r
act
er rec
o
gni
t
i
on i
s
occurre
d
because feature in test im
age
has some similarities with m
o
re than
one feat
ure i
n
re
fere
nce dat
a
base
.
Wh
ereas th
e
failu
re in
featu
r
e ex
traction
is
d
u
e
t
o
th
e im
p
e
rfect of
o
b
j
e
ct’s edg
e
. For ch
aracter
reco
gn
itio
n
with
tem
p
late
match
i
n
g
feature,
whe
r
e m
a
t
c
hi
n
g
pr
ocess i
s
usi
n
g
DTW
,
has a s
u
ccess
rate of 78% a
n
d e
r
ror
i
n
reco
g
n
i
t
i
on
of
22%
. A
n
er
r
o
r i
n
cha
r
acter recognition
is occurre
d beca
use
feat
ure i
n
test im
age has som
e
si
m
ilarit
i
es with
m
o
re th
an
one f
eat
u
r
e i
n
re
f
e
rence
dat
a
bas
e
[
7
]
.
4.
CO
N
C
LUS
I
ON
Th
e recogn
ition
system
p
r
o
c
esses p
e
rformed
b
y
b
i
n
a
ry
p
r
o
cess, seg
m
en
tatio
n
,
feat
ure ex
traction
wi
t
h
c
h
ai
n
co
d
e
m
e
t
h
o
d
a
n
d
t
e
m
p
l
a
t
e
m
a
t
c
hi
n
g
m
e
t
hod
,
and
feat
ure
m
a
t
c
hi
ng
pr
oces
s usi
n
g
dy
nam
i
c t
i
m
e
warping m
e
thod. Acc
u
racy level of chai
n code feature
rec
o
gnition is at 67% , and accura
cy level of tem
p
late
m
a
t
c
hi
ng feat
u
r
e rec
o
g
n
i
t
i
on
i
s
at
78%. T
h
e
resul
t
resea
r
c
h
m
eans succe
ssful
l
y
co
n
duc
t
i
ng t
o
m
o
t
o
r
vehi
cl
e
licen
se p
l
ate
reco
gn
itio
n.
REFERE
NC
ES
[1]
Ar
disasm
ita,M.S.,
Pen
g
o
l
aha
n Citra
Dig
i
ta
l Da
n Ana
l
i
s
is Ku
ann
ta
tif Da
lam Karakterisa
s
i Citra
Mi
krosk
opi
k
.
,
J
o
u
r
nal
M
i
kr
os
ko
pi
dan
M
i
kr
oanal
i
s
i
s
,
V
o
l
3
N
o
.
1
,
2
0
0
0
,
IS
SN
1
4
1
0
-
5
59
4
[2]
Fau
z
iah
;
Iw
an
W
a
h
yudd
in
, M.
2
009
.
Meto
d
e
Jaring
an
Sa
ra
f Ti
ru
an Pen
j
eja
k
a
n
Ba
lik un
t
u
k
Peng
ena
l
an
Hu
ru
f Cet
a
k pad
a
Citra D
i
g
ita
l
. Artificial, ICT Research
Cen
t
er
UNAS. Vo
l
u
m
e
3
;
1
Jauari
2009. Ja
karta :
Univers
itas Nasional
[3]
Heer
de
n, R
.
P.
20
02
.
Hi
d
d
e
n
Mark
ov Mo
del
s
f
o
r Ro
bust
R
ecog
n
i
t
i
on
of
Vehi
cl
e Li
cenc
e
Pl
at
es
. South
Africa : Un
iv
ersity o
f
Pretoria
[4]
Lilian
a
; Satia
Bu
dh
i, G;
Hen
d
ra.
2
010
.
S
e
g
m
en
tasi Pla
t
No
mo
r Kenda
raan
d
e
ng
an Men
ggu
na
kan
Met
ode
R
u
n
-
L
e
ngt
h
Sm
eari
n
g
Al
gori
t
hm
(
R
LSA)
. Juru
san
Tekn
ik
In
form
at
ik
a, Fak
u
l
tas Tekno
log
i
Indu
stri, Un
iv
ersitas
Kristen
Petra
[5]
Petzo
l
d
,
C.
201
0.
Prog
ra
mm
in
g Windo
ws Ph
on
e
7
.
Micros
oft Press: United
States of
America
[6]
Pur
n
om
o, H.
M
;
M
unt
ans
a
,
A.
2
0
1
0
.
Kon
s
ep
Pen
g
o
l
aha
n Citra
Dig
ita
l dan
Ekstraksi Fitu
r
.
G
r
aha
Ilm
u
: Yogyakarta
[7]
Su
to
y
o
, T; Mulyan
to
, E; Suhar
t
on
o, V
;
Dwi N
u
rh
ayati, O; W
i
j
a
n
a
r
t
o. 20
09
.
Teori
Pe
n
gol
ah
a
n
C
i
t
r
a
Dig
ita
l
. A
ndi
Of
fset
:
Y
o
gy
akart
a
.
Evaluation Warning : The document was created with Spire.PDF for Python.
I
J
ECE
I
S
SN
:
208
8-8
7
0
8
Po
lice Num
b
er Licen
se Recogn
itio
n
u
s
i
n
g Dyn
a
m
i
c Ti
m
e
Wa
rp
ing
Met
ho
d (Mad
e
S
udarm
a
)
28
4
[8]
Wahy
on
o,
S.E
.
20
09
.
Id
en
tifika
s
i No
mo
r Po
lisi Mo
b
il Men
ggu
na
ka
n
Meto
d
e
Jaring
an
Sa
ra
f Bu
a
t
an
Lear
ni
n
g
Vect
or
Qu
a
n
t
i
z
at
i
o
n
.
Juru
san
Tekn
ik
In
fo
rm
atik
a,
Un
iv
ersitas Gun
a
d
a
rm
a
[9]
h
ttp
://id
.wi
k
ip
ed
ia.org
/wik
i/No
m
o
r
_p
o
lisi
[10]
h
ttp
://p
engo
lahan
c
itra.co
m
/
in
d
e
x.php
?
o
p
tion
=
co
m
_
con
t
ent&task
=v
iew&id
=42
&
Itemid
=2
6
Evaluation Warning : The document was created with Spire.PDF for Python.