-
一、外
文
资
料
License Plate
Recognition Based On Prior Knowledge
Abstract - In this paper, a new
algorithm based on improved
BP (back propagation)
neural
network
for
Chinese
vehicle
license
plate
recognition
(LPR)
is
described.
The
proposed
approach provides a solution for the
vehicle license plates (VLP)
which were degraded severely.
What it remarkably differs from the
traditional methods is the application of prior
knowledge of
license plate to the
procedure of location, segmentation and
recognition. Color collocation is used
to locate the license plate in the
image. Dimensions of each character are constant,
which is used
to segment the character
of VLPs. The Layout of the Chinese VLP is an
important feature, which
is used to
construct a classifier for recognizing. The
experimental results show that the improved
algorithm is effective under the
condition that the license plates were degraded
severely.
Index Terms
- License plate recognition, prior
knowledge, vehicle license plates, neural network.
I. INTRODUCTION
Vehicle
License-Plate
(VLP)
recognition
is
a
very
interesting
but
difficult
problem.
It
is
important
in a number of
applications
such as weight-and-speed-limit, red traffic
infringement,
road
surveys
and
park
security
[1].
VLP
recognition
system
consists
of
the
plate
location,
the
characters
segmentation, and the characters recognition.
These
tasks become more
sophisticated
when
dealing
with
plate
images
taken
in
various
inclined
angles
or
under
various
lighting,
weather condition
and cleanliness of the plate. Because this problem
is usually used in real-time
systems,
it
requires not only
accuracy but also fast processing. Most
existing VLP recognition
methods [2], [3], [4], [5] reduce the
complexity and increase the
recognition rate by using some
specific features of local VLPs and
establishing some
constrains on the position, distance
from
the
camera
to
vehicles,
and
the
inclined
angles.
In
addition,
neural
network
was
used
to
increase
the recognition rate [6], [7] but the
traditional recognition methods seldom
consider the
prior
knowledge
of
the
local
VLPs.
In
this
paper,
we
proposed
a
new
improved
learning
method of BP
algorithm based on
specific
features of Chinese VLPs. The proposed algorithm
overcomes the low speed
convergence of BP neural network
[8] and remarkable increases the
recognition
rate
especially
under
the
condition
that
the
license
plate
images
were
degrade
severely.
II. SPECIFIC
FEATURES OF CHINESE VLPS
A. Dimensions
According to the guideline for vehicle
inspection [9], all
license
plates must be rectangular
and
have
the
dimensions
and
have
all
7
characters
written
in
a
single
line.
Under
practical
environments, the
distance from the camera to
vehicles and the inclined angles are
constant, so
all
characters
of
the
license
plate
have
a
fixed
width,
and
the
distance
between
the
medium
axes
of
two
adjoining
characters
is
fixed
and
the
ratio
between
width
and
height
is
nearly
constant. Those features can be used to
locate the plate and segment the
individual character.
B.
Color collocation of the plate
There
are four kinds of color collocation for the
Chinese
vehicle license
plate .These color
collocations are
shown in
table I.
TABLE I
Category
of license plate
small horse power
plate
motor
truck plate
military
vehicle
and police wagon plate
embassy vehicle plate
Color
collocation
blue background and white
characters
yellow background and black
characters
black background and
the white characters
white background and black characters
Moreover, military vehicle and police
wagon plates
contain a red
character which belongs
to a specific
character
set. This feature
can be used to improve the recognition rate.
C. Layout of the Chinese VLPS
The criterion of the vehicle license
plate defines the
characters layout of Chinese license
plate. All standard license
plates contain Chinese
characters, numbers and letters which
are
shown
in
Fig.1.
The
first
one
is
a
Chinese
character
which
is
an
abbreviation
of
Chinese
provinces.
The
second
one
is
a
letter
ranging
from
A
to
Z
except
the
letter
I.
The
third
and
fourth ones are letters or numbers. The
fifth to seventh
ones
are numbers ranging from 0
to 9
only. However the first or the
seventh ones
may be red characters in special plates (as shown
in Fig.1).
After segmentation process the
individual character
is
extracted. Taking advantage
of the
layout and color collocation prior knowledge, the
individual character will
enter one of
the classes:
abbreviations of Chinese provinces
set, letters set, letters or numbers
set, number set,
special
characters set.
辽
BA9083
Chinese
character
Letter
Letter
or
number
(a)Typical layout
Number
辽
BB092
警
Chinese
character
Letter
Letter
or
number
Special red
character
(b) Special character
Fig.1 The layout of the Chinese license
plate
III. THE PROPOSED
ALGORITHM
This
algorithm
consists
of
four
modules:
VLP
location,
character
segmentation,
character
classification
and
character
recognition.
The
main
steps
of
the
flowchart
of
LPR
system
are
shown in Fig. 2.
Firstly
the
license
plate
is
located
in
an
input
image
and
characters
are
segmented.
Then
every
individual
character
image
enters
the
classifier
to
decide
which
class
it
belongs
to,
and
finally
the BP network decides which character
the
character image represents.
Image acquisition
Plate
location
Characters
segmentation
classifier
Chinese
character
Letter
Letter
or
number
Number
Special character
Characters recognition
Fig.2
The
flowchart
of
LPR
system
A. Preprocessing the license plate
1) VLP Location
This process
sufficiently utilizes the color feature such as
color collocation, color
centers
and distribution in the plate
region, which are described
in section II. These color features
can
be
used
to
eliminate
the
disturbance
of
the
fake
plate
’
s
regions.
The
flowchart
of
the
plate
location is shown in
Fig. 3.
Characters edge detection
Binary image segmenting
Candidate image detection
Vehicle
plate
extraction
Fig.3 The
flowchart of the plate location algorithm
The
regions
which
structure
and
texture
similar
to
the
vehicle
plate
are
extracted.
The
process is described as
followed:
(1)
(2)
Here, the Gaussian variance
?
is set to be
less than W/3
(W is the
character stroke width),
so
P
1
gets its maximum
value M at the center of the stroke.
After convolution,
binarization is
performed
according
to
a
threshold
which
equals
T
*
M
(T<0.5).
Median
filter
is
used
to
preserve the
edge gradient and eliminate isolated
noise of the binary image. An
N
*
N
rectangle
median filter is set, and N represents
the odd integer mostly close to W.
Morphology closing operation can be
used to extract the
candidate region. The confidence
degree of candidate region
for being a license plate is verified
according to the aspect
ratio and
areas.
Here,
the
aspect
ratio
is
set
between
1.5
and
4
for
the
reason
of
inclination.
The
prior
knowledge of color
collocation is used to
locate plate region exactly. The locating
process of
the
license plate is shown in Fig. 4.
Fig. 4 The whole process of locating
license plate
2) Character
segmentation
This
part
presents
an
algorithm
for
character
segmentation
based
on
prior
knowledge,
using character
width, fixed
number of characters, the ratio of height to width
of a character,
and so on. The flowchart of the
character
segmentation is
shown in Fig. 5.
License plate image
preprocessing
Obtain binary image
Vertical projection
Eliminate space mark
Fig. 5 The flowchart of the character
segmentation
Firstly,
preprocess
the
license
the
plate
image,
such
as
uneven
illumination
correction,
contrast
enhancement,
incline
correction
and
edge
enhancement
operations;
secondly,
eliminating
space
mark
which
appears
between
the
second
character
and
the
third
character;
thirdly,
merging
the
segmented
fragments
of
the
characters.
In
China,
all
standard
license
plates contain only 7 characters (see
Fig. 1). If the
number of
segmented characters is larger than
seven, the
merging process must be performed.
Table II shows the
merging
process. Finally,
extracting the
individual character
’
image
based on the number and the width of the
character.
Fig. 6 shows the
segmentation results. (a) The incline and broken
plate image, (b) the incline and
distort plate image, (c)the serious
fade plate image, (d) the smut license plate
image.
TABLEII
Get
Nf
If
NF>
MaxF
For each
character segments
Calculate the medium
point
M
i
For each two consecutive medium points
Calculate the distance
D
K
Calculate the
minimum distance
Merge the
character segment k and the character
segment k +1
NF =
NF
-
1
End of algorithm
where Nf is the number of
character segments, MaxF is the
number of the license plate, and i is
the index of each
character segment.
The
medium point of each segmented character is
determined
by:
（
3
）
where
S
i
1
is the initial coordinates for the
character segment,
and
S
i
2
is the final coordinate
for
the character segment.
The
distance between two consecutive medium points is
calculated
by:
（
4
）
Fig.6
The
segmentation
results
B. Using specific prior
knowledge for recognition
The
layout
of
the
Chinese
VLP
is
an
important
feature
(as
described
in
the
section
II),
which can be used to
construct
a
classifier for recognizing. The recognizing
procedure
adopted
conjugate gradient
descent fast learning method,
which is an improved
learning method of BP
neural
network[10].
Conjugate
gradient
descent,
which
employs
a
series
of
line
searches
in
weight or parameter
space. One picks the first
descent direction and moves along that
direction
until
the
minimum
in
error
is
reached.
The
second
descent
direction
is
then
computed:
this
direction the
“
conjugate
direction
”
is the
one along which the
gradient does not change its
direction
will
not
“
spoil
”
the
contribution
from
the
previous
descent
iterations.
This
algorithm adopted topology 625-35-N as
shown in Fig. 7. The size
of input value is 625 (25*25 )
and
initial weights are with random values,
desired output values
have
the same feature with
the input values.
Fig. 7
The
network
topology
As Fig.
7 shows, there is a three-layer network which
contains working signal
feed forward
operation and reverse
propagation of error
processes. The target parameter is t and
the length
of
network output vectors is n. Sigmoid is the
nonlinear transfer
function, weights are initialized
with
random values, and changed
in a direction that will reduce the
errors.
The algorithm was
trained with 1000 images of different
background and illumination most
of
which
were
degrade
severely.
After
preprocessing
process,
the
individual
characters
are
stored. All characters used for
training and
testing have
the same size (25*25 ).The integrated
process for
license plate recognition consists of
the following steps:
1) Feature
extracting
The feature vectors from
separated character images have
direct effects on the recognition
rate. Many methods can be
used to extract feature of the image
samples, e.g. statistics of
data
at
vertical
direction,
edge
and
shape,
framework
and
all
pixels
values.
Based
on
extensive
experiments, all
pixels
values method is
used to construct feature vectors. Each
character was
reshaped into a column of 625
rows
’
feature
vector. These feature
vectors are divided into two
categories
which can be used for
training process and testing process.
2) Training model
The
layout
of
the
Chinese
VLP
is
an
important
feature,
which
can
be
used
to
construct
a
classifier for training, so five
categories are divided. The
training process of numbers is
shown
in Fig. 8.
Fig. 8 The architecture of a neural
network for character recognition
As
Fig. 8 shows, firstly the classifier decides the
class of
the input feature
vector, and then
the
feature
vector
enters
the
neural
network
correspondingly.
After
the
training
process
the
optimum parameters of the net are
stored for recognition. The
training and testing
process is
summarized in Fig. 9.
Input character vector
for
recognition
Neural
network
Target output
error
(a)
Training
process
Input character vector
for recognition
Neural
network
output
(b)Testing process
Fig.9 The
recognition process
3) Recognizing
model
After training process there are
five nets which were
completely trained and the optimum
parameters were stored.
The untrained feature vectors are used
to test the net, the
performance
of
the
recognition
system
is
shown
in
Table
III.
The
license
plate
recognition
system
is
characterized by the
recognition rate which is defined by
equation (5).
Recognition rate =(number of correctly
read characters)
/
(number of
found characters)
(5)
TABLE III
Class
Number
Letter
Chinese
character
Number and letter
Special character
IV. COMPARISON OF THE
RECOGNITION RATE WITH OTHER
METHODS
In
order
to
evaluate
the
proposed
algorithm,
two
groups
of
experiments
were
conducted.
One group is to compare the
proposed method with the BP based
recognition method [11]. The
result is
shown in table IV. The other group is to
compare the proposed method
with the method
based on SVM
[12].The result is shown in
table V. The same training and
test data set are used.
The
comparison
results
show
that
the
proposed
method
performs
better
than
the
BP
neural
network
and SVM counterpart.
TABLE IV
Method
Our method
BP
Chinese character
96%
94
．
5%
TABLE V
Method
Our method
SVM
Chinese character
96%
93
．
7%
V. CONCLUSION
Number
99.5%
99.5%
Letter
97.4%
95.7%
Number
99.5%
97.6%
Letter
97.4%
89.8%
Recognition
99.5%
97.4%
96%
97.3%
98.2%