2015 Fifth

2015International
Fifth International
Conference
Conference
on Instrumentation
on Instrumentation
and Measurement,
& Measurement,
Computer,
Computer,
Communication
Communication
and Control

Research on the License Plate Recognition Based on Image Processing

2
Energy Science and engineering, Harbin Institute of
Chaofeng Lan 1a,2 2b 1 1
, Fengchen Li , Yingjian Jin , Xuemei Sui , Technology, 150001, China
Shouqiang Kang1 e-mail: [email protected];[email protected]
1
School of Electrical and Electronic Engineering, Harbin
University of Science and Technology Liping Zhang3
Harbin 150080, China 3
SonoScape, Co., Ltd, Shenzhen, ChinaCountry
e-mail: [email protected]

Abstract—In order to enhance the effects of license plate very mature and some has been put to practical use. For
recognition in intellectual transport domain, this article carries example: Israel's Hi-Teeh company's See Car can be
out positioning, partitioning and characteristics extracting on reached in 0.7s 93% recognition rate; Optasia Singapore's
license plate images that have been collected, so as to realize Singapore VLPRS mainly for license plates, primarily
the function of license plate character recognizing. In order to
boost the contrast on a license plate and to reduce noises,
suitable to Singaporean license plates and achieving an
methods of gray degree reinforcing and filtering treating are astonishing 99.7% recognition rate within 0.4s-2s; Auty
adopted to carry out gray degree transformation; in order to Australian companies have set up a traffic monitoring
accurately position the location of license plate characters, system called Safe - Cam system that can work around the
adopt Ostu algorithm to work out global threshold value, clock and seizing still pictures of vehicles running at
extract image edges, and make a second positioning on the 160km/h; processing 50 frames per second to an accuracy
license plate with projection method, thus a binary license up to 92%; The ALPHATECH in American rely on aspects
plate image can be acquired; partition characters according to such as AVI product fault tolerance performance is stronger
peak-group characteristics of vertical projecting figures, and and able to gain a foothold in the field[3-5].
adopt improved template-matching algorithm to realize
accurate and efficient license plate recognizing. The research
The research of automatic license plate recognition in
findings of this article have important theoretical values and China started late, and it occurred in the late eighty's. 1988
practical application significance. (Abstract) Dai Ying used common image processing methods to put
forward the idea that the classification of Chinese characters
Keywords-edge detection; Canny operator; morphology; recognition is conducted on the basis of Chinese character
projection segmentation; template matching; character feature extracting. In China Science and Technology
recognition University, Xinhan Huang and his team investigated the
license plate recognition system which based on the neural
I. INTRODUCTION (HEADING 1) network and template matching; In Shanghai Jiao Tong
With the continuous improvement of the level of the university, Feihu Qi and his team investigated the license
national economy, the urbanization process has been pushed plate recognition system which based on color segmentation,
forward, and the interaction between the cities is becoming in the case of unattended, which can automatically detect
more and more frequent. In order to meet people's travel the vehicle image information is particularly important [6-7].
needs, the number of private cars in our country is License plate recognition system can be arranged at each
increasing, and the private cars have become the first choice intersection, which can provide a powerful help for the
for us to travel and the most frequent traffic patterns we maintenance of traffic order. The correct identification of
used. At the same time, the development of the city is too the license plate number can provide real and effective
fast, and the traffic facilities construction is not reasonable information for the traffic management department to
enough, for example, traffic accidents, illegal driving, rush manage the traffic order. Therefore, the license plate
hour traffic jam and so on, which has brought great pressure recognition technology has great research value.
for the existing traffic management system. If you rely on II. DISCUSSION
current human resources to assist in the management of
traffic is not a very good solution to these problems, the use A. Research on image edge detection
of existing information technology to develop a information,
intelligence and automation of traffic management system Image edge information plays an important role in
will be a very effective and feasible way[1-2]. digital image processing and analysis. Edges typically occurs
Since the License Plate Recognition (LPR) technology between the target and the target, primitives and primitive,
launched in 1998, people have carried out extensive and in- object and background, or between regions and regions.
depth research. There has been a lot of foreign-related Edge detection methods typically use the traditional edge
article published, and even some technology has become detection operator, such as the Prewitt operator and Roberts

978-1-4673-7723-2/15 $31.00 © 2015 IEEE 731

DOI 10.1109/IMCCC.2015.160
operators Krich operator, Sobel operator. But most of the down the detection speed and accuracy. So it is necessary to
results obtained is not ideal, could not be applied to practice, deal with the edge image.
by contrast the Canny operator with better performance. Is
B. The position research of the region of license plate
Jhon Canny of Canny edge detection operator was first
characters
proposed in a paper published in 1986, it makes up for the
deficiencies in other operators. Over the years, the Canny Morphology is very important in the image processing,
operator with many improvements. For example, Adaptive opening operation and closing operation is one of the most
Canny edge detection operator can increase sensitivity to the basic operations. By combining the opening operation and
image edges while suppressing noise. closing operation in a certain order, can connect the license
The process of the Canny edge detection algorithm as plate area and cut off the small connection between the
shown in Fig.1. images, and then filter the image to get the approximate
gauss location of the license plate. After the closing operation and
RULJLQDO smoothing partial
LPDJH B(x,y)
derivative
derivative opening operation to the edge detection images from Fig. 2
denoising
$ [\ (Bx,By) by morphological method, get the largest license plate
non character image. Take advantage of this feature, remove the
maximum
suppressi part which less than the license plate character area, we can
on get the complete license plate area. The result is shown in
double Fig.4.
threshold
detection preliminary
image edge
connecti edge points
ng edge
Fig.1. The Canny algorithm flow chart
The primary images that have been collected as shown
in Fig.2. Through Canny operator to process the
preprocessed images again, extracts the edges of each object
in the image. We can get the result of edge detection, as
shown in Fig. 3.

Fig.4. Complete license plate images

As is shown in Fig.4, after in order to the closing
operation and opening operation to the edge detection
images, the license plate character area is connected in one
area, and exclude the surrounding small holes, burr and other
interference, and then get the license plate character area
roughly. But, at the same time, it is clear that some areas are
Fig. 2. The original image not part of the character of the license plate. Implementing
row-column scanning according to the characteristics of the
license plate, get the precise scope of license plate characters.
Since the acquisition of the license plate to the background
color is blue, the character is white, so the blue pixels are
scanned and cropped. The license plate character image is
shown in Fig.5.

Fig.3. The license plate image of edge detection Fig.5. The color license plate image of positioning
As is shown in Fig.3, we can get the edge information As is shown in Fig.5, the background color of the license
that has grey level transformation of license plate characters plate is blue. Remove the license plate around a small
after the edge detection. We can see the outline of the highlight by implementing row-column scanning, to get full
license plate roughly and position the region of license plate color license plate images.
characters by virtue of this information. But we also get the
edge information of the object around the license plate, such C. The segmentation research of the license plate
as lights, edge information of the ground. So if using edge characters
method to locate license plate location, will be subject to The common character segmentation algorithm include
interference with other environmental information and slow template matching character segmentation, clustering
analysis character segmentation, projection segmentation

732
and so on. The character segmentation is the gray level D. The recognition research of the license plate characters
projection of the vertical direction of the license plate, and After pretreatment of license plate location, character
the change of the vertical direction of the gray level is segmentation, we get binary image of license plate of a
obtained. We will segment characters based on it precisely. single character, but to achieve the license plate final
In this paper, the method of projection segmentation and recognition, the characters binary image express need to be
template matching is used to make characters compare with discerned. Therefore, character recognition is the last step of
template in order. the license plate recognition. Usually, there are a lot of
To get the color license plate image 5 for processing, algorithms for character recognition, which are common in
firstly it is converted into grayscale images, and then get the the algorithm of neural network, feature statistic matching
binary image. So after processing seven license plate method and template matching method. In this paper, the
characters will be gotten, which are basically connected template matching method is used to discern the license plate
each other. In addition, license plate fixing rivets, license character. Firstly, characters of segmentation in Fig.7 using
plate frame and dot between the second character and the bilinear interpolation will be the size of template unified to
third character, these image factors will influence calculate the pixels in the template easily; secondly, by
template matching method, using similarity formula
subsequent processing of gray projection, so they need to be
calculate the similar degree of the templates corresponding to
removed. According to common sense, character of the area
seven characters, and select the similar degree maximum
is larger than that of the rivet, license plate frame and dots, template as a result discern the characters of license plate
so the area less than license plate character can be filtered, finally, as shown in Fig. 8.
so the binary image of license plate character can be
obtained, and then gray statistics of the binary image in
column will be get, the column gray distribution as shown
in Fig. 6.

Fig.8. The result of character recognition

As is shown in Fig.8, the recognized license plate
character is consistent with the acquired character, which
indicates that the method is effective and practical.

Fig. 6. The license plate characters grey-scale image III. DISCUSSION

As is shown in Fig.6, the gray level projection can get This paper investigates four processes, among which are
the change of the vertical direction of the gray level. Each processing, partitioning, character segmentation and
gray level jump point is the left and right edge of characters. character recognition on license plate images that have been
The part which the gray value is greater than zero is the collected, so as to ultimately realize the function of license
character area, and the part that equal to zero is the distance plate character recognizing. Pretreatment of license plate
between characters. image removes the color plate and enhances gray-scale
Segment character by using projection partitioning image contrast; extract the edge information of image, and
algorithm and remove all black rows and columns by do open and close operation of the edge in order. The license
scanning, and then cut out the character. The result after plate character area is connected into one area. And filter the
scanning and clipping is shown in Fig.7. surrounding part of the license plate. Obtain the general area
of license by row and column scanning; using projection
segmentation method, with gray projection in column
direction of the plate, get each character on the left and right
sides of the edge coordinates, and segment characters in
order; using template matching method, unify the size of the
characters segmented and plate of Chinese characters,
Fig.7 Characters of segmentation numbers and letters needed firstly. Characters discerned
As is shown in Fig.7, the license plate characters have compare with plate. The greatest similarity with plate is
been accurate positioned and segmented from image. After a result. This paper is programmed on the MATLAB platform
series of process such as grayscale, binary, morphological to realize the automatic recognition of the license plate.
opening and closing operation, increase the resolution of
characters. ACKNOWLEDGMENT

733
 This work was supported by Natural Science Youth
Foundation of Heilongjiang Province of China
(No.QC2015082).

REFERENCES
[1] YANAMURA 㸪 GOTO M 㸪 NISIIIYAMA D 㸬 Extraction and
tracking of the license plate using hough transform and voted block
matching[J] 㸬 IEKli IV2003 Intelligent Vehicles
Symposium㸬2003㸸243-246㸬
[2] S. Rastegar㸪 R. Ghaderi㸪G. Ardeshipr㸪N. Asadi㸬 An intelligent
control system using an efficient License Plate Location and
Recognition Approac[J]㸬International journal of Image Processing
Volume 㸬2009㸸12-16㸬
[3] S.H. Mohades Kasaei㸪S.M. Mohades Kasaei㸪S.A. Monadjemi㸬 A
Novel Morphological Method for Detection and Recognition of
Vehicle License Plates[J]㸬American Journal of Applied Sciences
6 㸬2009㸸66-70㸬
[4] S. Ozbay㸪E. Ercelebi㸬Automatic Vehicle Identification by Plate
Recognition[J] 㸬 World Academy of Science 㸬 Engineering and
Technology㸬2005㸸150-158㸬
[5] O. Khalifa, S. Khan, R. Islam. Malaysian Vehicle License Plate
Recognition 㸬 The International Arab Journal of Information
Technology[M]㸬Suleiman Vol. 4㸬2007㸸359-364㸬
[6] Rafael C. Gonzalez, Richard E. Woods. Digital Image Processing[M].
Second Edition.Beijing: Electronic Industry Press㸪2003: 41-496.
[7] Mi-Ae Ko, Young-Mo Kim, License Plate Surveillance System Using
Weighted Template Matching[C]. IEEE Proceedings of the 32nd
Applied Imagery Pattern Recognition Workshop, 2003, 10: 269 ୍
274.

734