0-7803-9514-X/06/$20.00 ©2006 IEEE ICIEA 2006
A Novel Approach for License Plate Character Segmentation
Feng Yang
a
, Zheng Ma
a
, Mei Xie
b
a
School of Communication and Information Engineering,
b
School of Electronic Engineering,
University of Electronic Science and Technology of China (UESTC).
No.4, Section 2, North Jianshe Road, Chengdu, P.R.China, Post Code: 610054
Email: yangfeng_34@163.com, zma@uestc.edu.cn , mxie@uestc.edu.cn
Abstract
Character segmentation is an important step in license
plate recognition (LPR) system. In this paper, a novel
character segmentation method of license plate is
presented combining Laplacian Transformation, region
growing and prior knowledge of license plate. In the
proposed methodology, image preprocessing is
performed to the license plate at first, and the character
region in license plate is enhanced in the following.
Then the edges of the characters are detected by using
Laplacian Transformation and the candidate regions of
characters are located by using region growing
algorithm. And the character segmentation regions are
determined by using prior knowledge of license plate.
Finally the characters are segmented from original
license plate and binarization is performed to the
characters, which can make it more efficient for
character recognition in OCR system. The proposed
method in character segmentation is fast and accurate,
and is tolerant to license plate with deformations,
rotations, plate frame, rivet, the space mark, and so on.
And promising results have been obtained in
experiments on Chinese license plates.
1 Introduction
Automatic recognition of car license plates plays an
important role in traffic surveillance systems. Such
systems, which are applied in parking areas, highways,
bridges and tunnels, can help a human operator and
improve the overall quality of a service. Any situation
requiring the automatic control of the presence and
identification of a motor vehicle provided with a license
number may represent a potential application. The LPR
(License Plate Recognition) algorithm consists of three
steps: license plate locating, character segmentation and
character recognition. This paper presents a new
algorithm for character segmentation.
There are many factors that cause the character
segmentation task difficult, such as image noise, plate
frame, rivet, space mark, plate’s rotation and
illumination variance. Our algorithm uses Laplacian
Transformation, region growing algorithm and the prior
knowledge of license plate and overcomes the
difficulties mentioned above.
The proposed methodology is composed of six steps:
image preprocessing, enhancement of character regions,
edge detection of the characters, location of the
candidate regions of characters, determination of
character segmen-tation regions, characters
segmentation and binarization.
Till now, some works have been done on character
segmentation [1-8]. Compared with the method of
image binarization [1][2], this algorithm uses the
information of intensity and solves the problems of
abruption and conglutination of characters that are the
drawbacks of image binarization. And because of using
Laplacian Transformation, region growing and prior
knowledge, the segmentation is more accurate and
robust than the simple projection method [3][4].
Experimental results reveal that the proposed
methodology is very effective for the segmentation of
license plate’s characters.
The rest of the paper is organized as follows: section
2 discusses about the prior knowledge of Chinese
license plate’s region, section 3 introduces the algorithm
in detail, section 4 shows the experimental results and
finally section 5 concludes the paper with references.
2 Prior Knowledge of Chinese License
Plate
A representative Chinese license plate is shown in Fig.1.
In China, different vehicle has different type of license
plate(LR), such as bus and truck has LP with the black
and yellow, car and light van has LP with blue and
white, police and military vehicle has LP with black, red
and white, and diplomat vehicle has LP with black and
white. As we known, LP has many features compared
with other regions in the same image [9]. They are: 1)
LP is composed of two or three colors; 2) the outline
size of LP is 440*140 (mm), and ratio of width/height is
about 3; 3) base on LP’s statistic, the area of characters
probably occupy 20% of entire LP area; 4) the interval
of adjacent symbol usually is 12mm, and the distance of
the second to the third is 34mm. The size of character is
45*90(mm). In order to extract characters from LP
quickly and accurately, these features must be used
effectively.