International Journal of Engineering and Technical Research (IJETR)

ISSN: 2321-0869 (O) 2454-4698 (P), Volume-4, Issue-3, March 2016

Application Morphological for Vehicle License Plate

Recognition
Rong-Choi Lee, Huan-Sheng Wang, King-Chu Hung

transponder may be stolen copy and owners must be

AbstractThis paper is the use of image processing to self-absorbed, resulting in the identification easily extended.
develop an automated system to detect the license plate. The In recent years, PC hardware upgrades to improve speed
proposed license plate recognition system of this paper, recognition technology, recognition rate and speed. Vehicle
including license plate image binary, the license plate location,
Identification System are: automatic highway toll systems,
and segment the license plate characters and character
recognition four sections.
vehicle factory or building of access control, vehicle parking
In the first part of our license plate location in image field management, search for vehicle theft and arrested the
processing technology, the input gray scale image of the vehicle, illegal vehicles, said the social contribution which can
then Sobel edge detection method to find the plate edge, with the effectively save a lot of manpower and financial resources to
final morphological location to find the license plate. License waste.
plate image binary part, we use dynamic average gray value
method to find threshold, the license plate image grayscale B. Research Motivation
convert binary image. The character cutting part, we use eight Development of automated license plate recognition
neighborhood connected components labeling method to
system is intended to replace manpower license plate
segment the characters of the license plate.
In this study, at last we apply Sim Net neural network to recognition, judging from the images captured whether the
make recognition of the license plates. Finally, license plate license plate, if there locate the license plate area, then rip out
recognition success rate 95.69%. the license plate of the characters were identified. When the
recognition system is completed and hope to combine
Index Termsdynamic average gray value morphological
hardware, such as a camera or monitor instrument can capture
Sim Net neural network
images and apply research results in practice, automatic
license plate recognition after capturing the image, number
I. INTRODUCTION
interpretation, compared with human read the license plate
A. Background number, then enter the license plate number to check whether
the target is also looking for more and more efficient.
In today's Taiwan, quite frequent use of the vehicle from In recent years, the rapid growth of Taiwan's vehicles,
the traffic problems, such as derivatives: speeding, running also causing a lot of traffic problems, such as traffic
red lights, accident and motor vehicle theft cases. Each enforcement and the vehicle lost, police units to invest a lot of
vehicle has a unique license plate number, when there is a manpower looking for a vehicle, the effect is very limited. To
problem that can use to quickly identify the target vehicle's traffic enforcement, for example, by automatically speeding
license plate. Once a duplicate license plate must be corrected, and running a red light camera equipment can be offending
in order to prevent others from using false number plate vehicle camera images through high-speed snapshot down.
violations, so the original owner of the license plate punished. But before sent photos to the owners, to identify each
The current vehicle identification system can be divided violation of the vehicle's license plate number and
into two parts; it is to install a transponder in the vehicle when Also in the stolen vehicle inspection, police patrol units
the vehicle identification means is read by infrared RF or can be set up in important sections through the monitor and
microwave transmission number. The other is by taking the police look for other ways, but because of the passing cars on
actual image and then identification the license plate number the road quite a bit and the speed is quite fast, the human
from the image. cannot be in a short time within the license plate number and
When the vehicle speed exceeds the speed limit or install find out whether it is the correct interpretation of stolen
response may lead to recognition errors. To solve this vehicles. Currently inquiry stolen the owner is quite a waste of
problem the installation of monitoring systems to capture manpower. The use of license plate recognition system to
images of the vehicle license plate number to identify who identify a large number of illegal photos, find the photo
these vehicles cause poor recognition efficiency. Furthermore, number and address of the owner, you can effectively save
manpower vehicles, police officers need to visually check the
license plate number and then enter, it is an inefficient work.
Rong-Choi Lee, Department Computer and Communication, Institute C. Research of Related Literature
of Engineering Science and Technology, College of Engineering, National
Kaohsiung First University of Science and Technology 811, Taiwan, R.O.C., In the literature [1-6, 18, 19] is the location for checking
+886-7-6011000 ext 2040/86-973196786. your license plate. Where in the literature [3] need to find the
Huan-Sheng Wang, Department Computer and Communication,
National Kaohsiung First University of Science and Technology 811,
row and column thresholds, but it easily the influence of light
Taiwan, R.O.C. or at night and other factors, it can not find the perfect
King-Chu Hun, Department Computer and Communication, National
threshold. In [4] because the image is not in the same position
Kaohsiung First University of Science and Technology 811, Taiwan, R.O.C., by photography and by the impact of on roads steep slopes.

128 www.erpublication.org
 Application Morphological for Vehicle License Plate Recognition

Therefore, they will not easily match to effectively license. [5] At present, the original of Taiwan vehicle image, as show
is need to create a reference model, and license plate shall be in Fig. 2.1(a), vehicle image of gray level, as show in Fig.
identified standards, and in a different location photography, 2.1(b). And the use of binary process in license plate
so easily lead to license plate recognition errors. In the recognition method, the rear license plate characters and
literature [1, 16] is to use the license plate character background. And set a threshold value, the image of the gray
recognition, confusion vulnerable to similar characters, such level value greater than the threshold value is 1, but less than
as 1 and I, 0 and O, 2 and Z, 3 and B, S and 5, and so on. the threshold value is 0, as show in Fig. 2.1(c). But the
Therefore, they are not easy make the license plate characters brightness of the license plate is susceptible to light, so the
match, resulting in license plate recognition errors. In [13-17] optimal threshold value will not be a fixed value, but is often
is for the immediate license plate recognition system. used as the mean gray-scale value to the threshold, but the
disadvantage is that if the image brightness distribution is
D. Objection
uneven, the averages calculated may differ from the actual
Now people use vehicles in addition to cars, there are light threshold value. In addition, the binary method can achieve
motorcycles, motorcycle, taxi, or the like. In addition to the best Otsu binarization of evaluation in many literature are
different colors, there are different numbers or English good, its main purpose is to find a threshold value, to make a
characters. Since the car's license plate format in general, weighted sum of the number of clusters is the smallest
from left to right is composed of two letters and four numbers. variation.
License plate format is again one of the previous two letters
B. License Plate Property
into numbers. Another license plate format is four numbers
plus the combination of two letters. And taxi, bus, truck only License plate recognition systems are doing before, due to the
five words. countries of the license plate format is not the same, and the
Background noise, such as adaptive or because of the object of the research paper is the license plate of the vehicle
footage in the background there is a lot of complexity, you can used in Taiwan, not just black and white license plate number
locate the license plate can be from any corner of the screen, but different characters.
and in the noise above and below the license plate number, (1) Color: Own car license plates in black and white, white
make captures taken out of character deformation and red taxis, tour buses 5 red and white characters, van bus
identification fails, you must first identify the results will green and white big business, own a large truck bus own white
affect the noise after cleared to make identification process. and green.
(2) The license plate number of words: Taiwan's license plate
II. STUDY METHODS numbers in English can be a combination of five or six
A license plate recognition system processes start when characters, and are located in the middle of the plate. After the
only the vehicle through the camera to take pictures of the plate was positioned objects, objects can be judged by the
photo camera input the license plate recognition system, the number of words is not the license plate. Not the same
video input of the first proper pre-treatment, and then locate arrangement of different types of vehicles in English and
the license plate. If there locate the license plate area, then cut numbers.
out the license plate character recognition carried out, (3) Change edge: The levels will be very dense because the
according to identification results to determine whether the license plate through the five or six words, and some of the
license plate format. text and background brightness changes will be relatively
Which also has a license plate character recognition large, big changes will focus on some of the pixel license
without cutting [2], but must go through the exact positioning plates.
of the plates, and the plates after correction angle and size,
location out of the plate height, just the height of the
characters, and with a fixed size sampling frame windows on
the license plate sampling, sample images to be formalized
and out of character templates same size, and then all the
characters and templates (10 digits, 26 letters of the alphabet)
carried identification, identification of the sample after frame Fig. 2.1: (a) Original vehicle Fig. 2.1: (b) Image gray level.
moves one pixel to the right. But the license plate characters image.
only six characters, but the identification number will be more
than six times, and it is more than 36 characters on each time
you make a template identification, there are 36 more than the
value of the finished, select the highest value (0 to between 1),
the closer a value representative of the closer alignment of the
template, and judges whether there is more than the threshold
value, if there are more than the threshold value is expressed Fig. 2.1: (c) Image binarization.
identify characters, if the sample does not exceed the C. Image Edge Detection
threshold value indicates that no characters box .
[3] using Pulse Coupled Neural Network to do dynamic
A. Image Binarization vehicle location, and [5] is the use of neural networks to the
In order to effectively reduce the complexity and amount of edge of a particular color combination of the license plate to
information represented by the graphic image, the image such detection, can not reduce the portion of the edge of plate.
usually is binarized.

129 www.erpublication.org
 International Journal of Engineering and Technical Research (IJETR)
ISSN: 2321-0869 (O) 2454-4698 (P), Volume-4, Issue-3, March 2016
And [6] using wavelet transform to detect license plate, these
are for license plate location moves from the image edges.
The plate is generally the easiest method is to detect
directly the edge of the image to grayscale Fig. 2.1(b), which
may be used Sobel Filter matrices Mx and My, using follow (1)
and (2) formula, they can find out from the part of the image in
a horizontal or vertical edges, as show in Fig. 2.2.

1 1 1
M x = 0 0 0 (1)
1 1 1 (a) Image erosion. (b) Image dilation
Fig. 2.3: erosion and dilation operation
1 0 1
M y = 1 0 1 (2)
1 0 1
E. The Eight Neighborhood Connectivity Labeling

g1 g2 g3 Some of the binary pattern is adjacent pixels together, and

G= g 4 g 6
together is part of the same object. License plate location or
g5 (3)
characters do when you use to cutting, its purpose is to image
g 7 g8 g 9 capture out separately for each item. Using the method of
eight neighborhood connected components labeling at the
(a) Image
eight open.on the pixels, it(b)
directions is Image close.whether the
determined
9 same object.
Px = G M
n 1
n Xn (4) Fig. 2.4: Open and close operations.

Table 1: Eight neighborhood connected components labeling

9 1 4 6
Py =
n 1
G n M yn (5) 2
3
P
5
7
8

P= Px + Py (6) We used eight neighborhood items marked connectivity

issues to take into account the order of the license plate
G represents a point on the map and the pixel grayscale characters, as show in Fig 2.6, the leftmost characters start
eight neighboring pixels value, multiplied by the sum of Mx number. So we find from top to bottom, and then from left to
and My respectively after the Px and Py, using (4) and (5) right.
formula, the sum is the result of edge detection, as (5) formula Eight neighborhood connectivity labeling objects, we
P. should determine whether the object is connected around, and
whether the object has done linked marked. If there are no
marked pixels and pixel itself is connected to the same thing,
if there is already marked, you must indicate the two identical
objects become the same thing.

Eight connectivity labeling process, from the search

direction of a non-zero pixel P (i, j) do labeled, Table 1
represents the position in the following Table 2, black plaid
represent P5, it also represent P (i, j), its expressions as
Fig. 2.2: Image edge use Sobel method. formula 7.

D. Morphology
Table 2: Represents the Table 1 position.
Morphology of the use of images to analyze the shape of
the structure, in order to strengthen the structural P1 P4 P7
characteristics of the image as is license plate localization. P2 P5 P8
Morphology have (1) Erosion, as show in Fig 2.3(a). (2) P3 P6 P9
Dilation, as show in Fig 2.3(b). (3) First erosion and then
expand (opening) method, as show in Fig 2.4(a). (4) First In Table 2 Mathematical expressions, the following
expand and then erosion (closing) method, as show in Fig expression.
2.4(b).

130 www.erpublication.org
 Application Morphological for Vehicle License Plate Recognition

P1 P4 P7 Pi 1, j 1 Pi 1, j Pi 1, j 1 1 1 1
P2 P5 P8 Pi , j 1 Pi , j Pi , j 1 (7) M y 0 0 0
1
(9)
4
P3 P6 P9 Pi 1, j 1 Pi 1, j Pi 1, j 1 1 1 1

The eight neighborhood connectivity labeling of four ( P1, Then for a license plate in the text and the background of
P2, P3, P4) positions. In accordance with the order of some of the high brightness contrast, given a threshold binary,
inspection, when P (i, j) not indicated number and relative the value of each pixel on the vertical edge of the map, the
positions P1, P2, P3, P4 number are zero, then P (i, j) for a threshold value of 1 or greater, the image is white. Is less than
new object give the new object number. If Pn the position of the threshold value was 0, the image is black. The original
non-zero, then the same as P (i, j) the number and Pn. If i was image is turn into the binary pattern.
not continued to check the remaining four positions, P5 also
been compiled.
After eight neighborhood connectivity labeling
procedure, to obtain upper, lower, left, right boundary
coordinates and pixels of objects. After searching for the end
of the article you can find a number 1, and then get object ID
to get the object boundary coordinates.

III. STUDIES THE ARCHITECTURE

Fig. 3.2: Image vertical edge map. Fig. 3.3: Image vertical edge
Process license plate recognition system is divided into map of Sobel method.
three parts: (1) License plate location: From the image in the
region of the license plate may be positioned out. (2) Before you get the binary edge image map, using the
Character cut: to determine whether the area has been targeted principle of morphological, will fill the gap between the edges
for the license plate, and capture the license plate characters. of the plate, while the other noise filter to clear its action is
(3) Number identification: Identification in license plate divided into three steps: (1) License plate image in the
characters. horizontal direction of closing. (2) License plate images after
breaking the horizontal direction (Opening). (3) License plate
A. License Plate Location images of breaking through the vertical direction (Opening).
License plate recognition in image input part contains the C. Determine the License Plate of Image
license plate after treatment license plate location, license
plate may be produced objects, similar to the license plate and Finally, the use of the eight neighborhood connected
then determine the shape of the object, were cut characters. components labeling method for the original image to find the
location of the license plate.
B. Edge Detection Every object record images upper, lower, left, right and
Enter the license plate photos are color images (R, G, B). the number of pixels. Because the license plate will be taken
The color images into grayscale image in G. G (i, j) = 0.3 R as the angle and distance affect the aspect ratio of the license
(i, j) + 0.59 G (i, j) + 0.11 B (i, j) + 0.5, the value is plate. After the morphological of break and close three steps
between from 0 to 255. Then from grayscale images G edge will affect the object area ratio. In order to do the exact
detection, edge detection use Sobel filter available in Fig. 3.1. character and the least restrictive license plate cutting width
and height? With the following three criteria to select a plate
objects.

height lower upper 1 (10)

width right left 1 (11)

width
3 6 (12)
height
Fig. 3.1: (a) Original image. (b) Image gray level.
(c) Image of binarization. (d) Sobel method to edge map.
Totall Pixels
0.6 (13)
width height
Sobel filter image after beginning operation of Mx get
vertical and My edge map, using follow (8) and (9) formula, column
as show in Fig. 3.2 and Fig. 3.3 the license plate image width (14)
8
grayscale map to find vertical edge map.
The upper, lower are image of top, below, width is image
1 0 1 of width, height is image of high, the Totall Pixels word is
1
Mx 1 0 1 (8) image the number of pixels of sum.
4 When (13) value closer 1, indicating that this item will be
1 0 1 closer quadrilateral. The aspect ratio (12) falls within a range

131 www.erpublication.org
 International Journal of Engineering and Technical Research (IJETR)
ISSN: 2321-0869 (O) 2454-4698 (P), Volume-4, Issue-3, March 2016
the object will be closer to the shape of a rectangular plate. Use binary license plate do binarized as show in Fig.2.2
Objects meet these conditions, there may be as part of the (c). The main purpose is to reduce the image binary image of
license plate image. the large amounts of data volume, with a threshold to P
But some parts, such as the left-most digit of 7 or right grayscale image converted into binary. Whereby the complex
most character is L, are likely to be closed will be filtered out background images and objects to be captured separated.
before do opening action part of the character, again just to be
able to fully capture the background, not the background is 0 P( i , j ) 255 (21)
the text interrupted during binary reverse judgment, make the
background even as an object. 1, if P( i , j ) threshold
In order to accurately capture the license plate characters, Binary( i , j ) (22)
the need to expand the scope of license plate location. First set 0 , other
H and W, using follow (15) and (16) formula. How to find the threshold is a big problem, most of the
height paper is the use of Otsu binary on to find the optimal threshold
H (15) value, and in order to save the amount of calculation can also
5
be used when the average gray-scale threshold value.
From the table can be found in the threshold value as
3 height 6 h width environmental conditions change in the recorded image, so
W (16)
2 you can not use the fixed threshold value, in order to reduce
the amount of computation, the average value of the use of
Renew: gray-scale values as a threshold, you can get very good effect.
Whether the image average grayscale value of binary
Upper Upper H (17) method, as show in Fig. 3.5(b), we are used a threshold for the
image of binary, you cannot succeed in capturing the
character or the character and background separation. We
Lower Lower H (18)
employ dynamic mean value binary method make the image
binary operation, as show in Fig. 3.5(c).
Left Left W (19)

Right Right W (20)

With renew Upper, Lower, Left, Right to retrieve the

object, using follow (17) and (20) formula. It has expanded to (a)Original license plate. (b) Binarization license plate.
capture the character completely fall within the scope of the
text and the background is not part of the interruption.

(c) Dynamic mean value of binary license plate.

Fig. 3.5: Original and Mean value binarization and dynamic

mean value binarization license plate map.
So we are doing a dynamic binary, the main aim is to make
each pixel has its own threshold, in order to achieve
Fig. 3.4: License plate extract. separation of text and background. The main approach is to
calculate the image around each pixel of the gray scale mean
value to each pixel when the threshold value, can improve the
above problems.
D. Segment the License Plate Calculation of dynamic binarization threshold, first obtain
the height and width of the image. Where in the height is equal
After it find the license plate, segment the license plate, as
to the number of images row, column width equal to the
show in Fig. 3.4.
number of images. When we want to find a threshold, to
E. Characters Extract calculate the average value of the yellow grayscale range, set
the threshold value greater than or equal to a threshold value,
Although the second part of the license plate recognition
less than the threshold value is set to 0.
aims to capture characters, the characters in the license plate
capture objects out before capturing characters, must do the
Using the above methods to obtain binary pattern,
reverse binarized determination, recalculation capture out of
although the increase is very computationally intensive large
the number of characters, this item may be determined
increase, the resulting binary pattern of the text and
whether the license plate.
background can completely cut out.
F. License Plate Image Binarization
G. Binary License Plate Reverses Determination

132 www.erpublication.org
 Application Morphological for Vehicle License Plate Recognition

Generally speaking, we have chosen to do black and white vector has had tremendous value, and thus control the overall
characters cutting, so we must be able to determine the license output, and therefore the proper normalization is necessary, in
plate after binarization will produce black and white or black this study for each input neuron will formalize -1 to the
and white. And choose the black and white characters that we interval 1.
need to do the cutting. Each character map feature values upon acquisition, you
First Object, after binarization produce a binary image, let can get the feature vector values, when the identification
B1 is a binary image. It is additionally generated Inverse characters can use the following matrix multiplication
binary image B2 (B2=1-B1). Binary map obtained after B1 representation, using follow from (23) to (26) formula..
and B2 respectively communicated with eight neighborhood
connectivity labeling way to find the largest object, this object X x1 x 2 x n (23)
may be part of the license plate backgrounds.
Y y1 y 2 y n
If B1 has the largest objects, use B2 is what we need black
(24)
and white binary pattern, so that B3 = B2, which can be
obtained from the scope of the license plate background B1. If
the object has the largest area of B2, so B3 = B1, can be Y F(X W) (25)
obtained from the scope of the license plate background B2.
From B3 binary to identify bit in the range of the license plate e x e x
background objects, to retain articles having a certain height F( x ) ,1 F( x ) 1 (26)
and width, and capture them. Rip out the text if there are five e x e x
or six, then this thing is very likely that the license plate, the
X is a feature value vector, W is a weight matrix, Y are
identification numbers do. If the number is not captured five
output units, F(x) is an activation function, can limit the
or six, it means that this item is not a license plate. The image
output neurons between -1 and 1. Required identification
do eight neighborhood connected components labeling
weight matrix W, available learning stage to produce weight
objects, you can get information on each object. Identify all
matrix. W's in the learning process must first set the X and T.
objects in the highest height value Max H do reference, will
Suppose a learning collation characters S samples, the
be less than (0.7 * Max H) objects as noise cleared, the stay
samples were each character must be classified, an A input
may be text objects are part.
characters X, you must make A representative output neuron
I. Identification Number Plates 1, the other output neurons classified as -1, and the
classification results into T, T are target values.
Before the license plate number identification process,
(1) Set the initial value of the X, T.
you must first capture the character map out action after it do
(2) After the initial value is set, the first to produce an initial
normalization, remove the characteristic values from the
weight matrix W, using follow (27) and (28) formula.
character map. The use of feature values to identify the
learning stage or stages, so that the initial value of the former
X X / norm(X) (27)
SimNet network learning taken by each license plate
characters and produce appropriate weights, while the latter
W XT T (28
use these weights to identify the work.
)
J. Imaging Feature Values
(3) The weight matrix W learning and correction phases,
With the most basic way to get from the image distribution
using follow from (29) to (32) formula.
characteristic values, the following steps: capture the
character of the binary pattern, the image size is row x
Y X W (29)
column. The characters of image formalized into a fixed
image size. Preset feature matrix zero matrix of size 40x24.
Characteristic matrix sampling the binary data into the feature dY T Y (30)
points on the graph, and then the feature vector X is changed 0
into -1, characteristic value acquisition is complete. dW X T dW (31)
K. SimNet
W W dW (32)
In recent years, neural network class played on classified
information is a very important role. Neural network [7 -12]
are based on human thought and brain activity and construct a (4) Repeat (3) step of the process, after 10,000 cycles of
mathematical model similar to brain function. Neural network operation or when the error is less than the critical value
has a similar human brain cells to each other by a number of dY end of the learning phase.
simple and continuous composed of neurons, and parallel (5) The right weight matrix obtained by dividing the Norm
processing operations between them, and its architecture has (X) obtained number of new matrix, as follow (33) formula.
its regularity. And bypass SimNet network, which for a
two-tier neural network, namely input layer and the output W W / norm(X) (33)
layer; input layer for receiving various characteristics of the
input and the output layer is used to represent the input by
those who are classify the case. In input layer and output layer Updated finished W matrix, this matrix W can be used in the
to complete the link. In order to avoid a value in the input identification license plate number.

133 www.erpublication.org
 International Journal of Engineering and Technical Research (IJETR)
ISSN: 2321-0869 (O) 2454-4698 (P), Volume-4, Issue-3, March 2016
L. Experiment process according to size ratio. During the experiment, the character
image height adjustment for the 40 pixels (image size is
During the experiment, the license plate recognition
24x40), weight value ratio will be adjusted to 30 pixels
system is divided into two stages, the first stage and the
(image size is 18x30) obtained false positive rate can be
second stage of the plate positioning the cutting process are
reduced at the end, similar characters (such as 8 and B, 0, and
the same characters, the goal is make the image cut out the
D, etc.) are able to correctly identify.
license plate characters, not the same part of the third stage,
after the completion of license plate location and character
IV. VEHICLE LICENSE PLATE RECOGNITION OF
cutting, the first characters of the binary pattern directly to the
RESULTS
archive, and then do the training sample plates. The other is to
make the character binary pattern character recognition. Experiment recognition results, all license plate images
232, correction license plates recognition have 222 successful,
M. Input Image recognition rate about is 95.69%. Most of characters have
Experiments with a single license plate license plate faded or scratched, resulting in incomplete capture.
photos is fixed proportion of 500666 pixels (row = 500, Identification fails reflective image on a license plate
column = 666). Experiments photograph license plate characters, resulting in incomplete capture. Most characters
location is not necessarily located in the center of the image, it have faded or scratched, resulting in incomplete capture is
will not affect the outcome of the license plate location when noise into the non-character. A license plate location is
positioning, but the camera angles and distance will affect the too large, failed to capture. Number of characters captured
results of positioning. less than four or more than six, was determined not to license
plates. There are problems in removing the screw part by
N. License Plate Location other non-character objects. Impact, resulting in some of the
The license plate of location, as show in Fig.4.1 white part. character image is cut off. Range of license plate location is
too small; the license plate image is incomplete. This
recognition rate are higher than [9, 10] articles in reference, as
show in Table 4.

Table 3: Experiment data.

Experiment Successfully /all Recognition rate
sheets
Result 222/232 95.69

Fig. 4.1: White is license plate of location. Table 4: Recognize rate compare.
.
Paper No. Recognize rate (%) sort number
O. License Plate Edge and Extraction
In Fig 4.2 is the license plate use Sobel method edge and This article 95.69 1
at upper is extracted the license plate.
[11] 94.2 3
[12] 95.33 2

V. CONCLUSIONS
In this paper, in order to identify the license plate
grayscale value change text and background for many of shots
out of the picture (including license plates of different colors),
to observe the change of its edge part of the text, and then use
Fig. 4.2: License plate Sobel method edge and extraction. the morphological processing, the license plate area capture
it. No fixed way to search windows license plates, because the
photographs in the vehicle, it is impossible to remain at the
same distance, the angle can also cause changes in the license
P. Character Recognition
plate.
In the license plate character recognition identification
Identification license plate recognition success rate
process is the most important part of this research on the
imaging features video input values will be affected. In the
character recognition using SimNet network as identification
previous part of the license plate location out of the license
purposes. In this study, using a mix of English and
plate, license plate images are not just in the background and
Mathematics neural network, the packet into 10 digits (0 to 9),
text, Histogram not simply divided into two clusters, and
and 25 kinds of letters of the alphabet (26 letters of the
usually gray-scale values of other non-text, and then through
alphabet, the numbers 0 deduct the same letters O).
the usual binary image will become noise, filtered through
Must first character image normalization and
some non-text processing, such as: internal purge of
standardization of action is mainly to adjust prior learning
non-background noise, noise connected with screws, and find
SimNet network characters or character recognition height,
the true height of the text.
and then adjust the width of the characters in the image

134 www.erpublication.org
 Application Morphological for Vehicle License Plate Recognition

Some studies uneven brightness distribution license plate [15] C. D. P. Crovato and A. Schuck*, The Use of Wavelet Packet
Transform and Artificial Neural Networks in Analysis and
images, the use of Otsu binary method, may be because most
Classification of Dysphonic Voices, IEEE Transactions on
of the license plate the shadows caused by the characters are Biomedical Engineering, Vol. 54, No. 10, October 2007, pp.
connected, or uneven illumination can not capture the full 1898-1990, (2007) IEEE.
character. In order to complete separation of text and [16] Wu Wei, Yuzhi Li, Mingjun Wang, Zhongxiang Huang. Research on
background, use dynamic binary method to calculate the Number-Plate Recognition Based on Neural Networks, Dept. of
Road and Traffic Engineering,Changsha Communications University.
threshold value of each pixel, so when faced with the license (2001).
plate on the shaded image or uneven illumination, but also can [17] H. Sheng, C. Li, Q. Wen, and Z. Xiong, Real-Time Anti-Interference
rip out the full text. Finally, the character image, using Location of Vehicle License Plates Using High-Definition Video,
SimNet perform character recognition test result. Finally, IEEE Intelligent Transportation Systems Magazine, (2009).
[18] L. Hai, and G. Shrestha, Real Time License Plate Recognition at the
license plate recognition success rate about 95.69%. Time of Red Light Violation, College of Automation, Harbin
Engineering University, Harbin, China, International Conference on
REFERENCES Computer Application and System Modeling (ICCASM 2010), Vol 2,
pp. V2-243-V2-247, (2010).
[1] A. Conci., J. E. R. d. Carvalho and T. W. Rauber, A Complete System
[19] A. W. G. C. D. Wijetunge and D. A. A. C. Ratnaweera, Real-Time
for Vehicle Plate Localization, Segmentation and Recognition in Real
Recognition of License Plates of Moving Vehicles in Sri Lanka, 2011
Life Scene, IEEE Latin America Transactions, Vol. 7, No.5,
6th International Conference on Industrial and Information Systems,
September (2009).
ICIIS 2011, pp. 82-87, Aug. 16-19, (2011), Sri Lanka.
[2] Mario I. Chacon M., Alejandro Zimmerman S. License Plate Location
[20] L. Hai, and G. Shrestha, Real Time License Plate Recognition at the
Based on a Dynamic PCNN Scheme, DSP & Vision Lab, Chihuahua
Time of Red Light Violation, College of Automation, Harbin
Institute of Technology.(1995).
Engineering University, Harbin, China, International Conference on
[3] P. Kulkarni, A. Khatri, P. Banga and K. Shah, A Feature Based
Computer Application and System Modeling (ICCASM 2010), Vol. 2,
Approach for Localization of Indian Number Plates, Dept. of
pp. V2 243247, (2010).
Electronics and elecommunication, Smt. Kashibai Navale College of
[21] C. A. Rahman, W. Badawy and A. Radmanesh A Real Time
Engineering, University of Pune, India, IEEE, pp. 157162, (2009).
Vehicles License Plate Recognition System, Proceedings of the IEEE
[4] Y. Song, A Level Set Based Method of License Plate Localization,
Conference on Advanced Video and Signal Based Surveillance
Key Laboratory of Intelligent Information Processing, Institute of
(AVSS03), (2003) IEEE.
Computing Technology, Chinese Academy of Sciences, IEEE IRI
2010, Las Vegas, Nevada, USA, pp. 403-404, August 4-6, IEEE,
(2010).
[5] X. Wei, Seeds Algorithm in Vehicle License Plate Location System, BIOGRAPHIES
Computer and Communication Engineering School of Weifang
University, Weifang, China, IEEE, Second IITA International Rong-Choi Lee was born in Taiwan, Republic of China, the birth date of
Conference on Geoscience and Remote Sensing, vol. 1, pp. 441443, September 20, 1961, I am a student at National Kaohsiung First University
(2010). of Science and Technology, Computer and Communication Engineering, the
[6] F. Faradji, A. H. Rezaie, M. Ziaratban, A Morphological-Based research field of image processing, message encoding. I hope to do some
License Plate Location, ICIP2007, pp. I-57 I-60, (2007) IEEE. research on the human life has improved.
[7] P. M P, M. S. A.l Majid, S. Yaacob, M. H. F. Rahiman and R. P.
Krishnan, Statistical Time Energy Based Damage Detection in Steel Huan-Sheng Wang is a professor of the Republic of China Kaohsiung
Plates Using Artificial neural Networks, School of Mechatronics First University of Science and Technology, the main research is image
Engineering, University Malaysia Perlis, Perlis, Malaysia, processing and signal processing, and vehicle license plate to identify.
International Colloquium on Signal Processing & Its Applications
(CSPA) , IEEE, pp. 33 - 36, May (2009). King-Chu Hung is a professor of the Republic of China Kaohsiung First
[8] Ching-Tang Hsieh, Yu-Shan Juan, Kuo-Ming Hung. Multiple University of Science and Technology, the main research is image
License Plate Detection for Complex Background, Department of processing and signal processing, and breast cancer to identify and ECG
Electrical Engineering, Tamkang University, Department of symptoms to identify.
information Management, Kainan University, Taipei, Taiwan.(2005).
[9] M.-K. Wu, J.-S. Wei, H.-C. Shih, and C. C. Ho, License plate
detection based on 2-level 2D Haar wavelet transform and edge density
verification, Department of Electrical Engineering, National Yunlin
University of Science and Technology, Douliou, Yunlin 64002,
Taiwan, ROC, IEEE International Symposium on Industrial
Electronics (ISlE 2009), Seoul Olympic Parktel, Seoul, Korea, pp.
1699-1704, July 5-8, (2009).
[10] N. Chen and L. Xing, Research of License Plate Recognition based on
Improved BP Artificial neural Network, College of computer science
Xian Polytechnic University Xian, China, International Conference
on Computer Application and System Modeling, Vol. 11, pp. V11-482
-V11-485, (2010).
[11] Rong-Tsai Lee and King-Chu Hung, Real-time vehicle license plate
recognition based on 1-D discrete periodic wavelet transform, 2012
International Symposium on Computer, Consumer and Control, IS3C
2012, pp. 914-917, (2012).
[12] R. T. Lee, K. C. Hung, and H. S. Wang, Real Time Vehicle License
Plate Recognition Based on 2D Haar Discrete Wavelet Transform,
International Journal of Scientific & Engineering Research, Volume 3,
Issue 4, April-2012, ISSN 2229-5518, IJSER (2012).
[13] B. Shan, Vehicle License Plate Recognition Based on Text-line
Construction and Multilevel RBF Neural Network, College of
Automation and Electronic Engineering Qingdao University of
Science and Technology Qingdao, 266042, China, Journal of
Computers, vol. 6, no. 2, pp. 246-253, FEBRUARY (2011).
[14] S. N. H. S. Abdullah, M. Khalid and R. Yusof, License Plate
Recognition using Multi-cluster and Multilayer Neural Networks, pp.
1818-1823, (2006) IEEE.

135 www.erpublication.org