28-09-2012, 11:11 AM
An Algorithm for License Plate Recognition Applied to Intelligent Transportation System
1An Algorithm for License.pdf (Size: 1.91 MB / Downloads: 68)
Abstract
An algorithm for license plate recognition (LPR)
applied to the intelligent transportation system is proposed on
the basis of a novel shadow removal technique and character
recognition algorithms. This paper has two major contributions.
One contribution is a new binary method, i.e., the shadow removal
method, which is based on the improved Bernsen algorithm
combined with the Gaussian filter. Our second contribution is a
character recognition algorithm known as support vector machine
(SVM) integration. In SVM integration, character features are
extracted from the elastic mesh, and the entire address character
string is taken as the object of study, as opposed to a single
character. This paper also presents improved techniques for image
tilt correction and image gray enhancement. Our algorithm
is robust to the variance of illumination, view angle, position,
size, and color of the license plates when working in a complex
environment. The algorithm was tested with 9026 images, such
as natural-scene vehicle images using different backgrounds and
ambient illumination particularly for low-resolution images. The
license plates were properly located and segmented as 97.16% and
98.34%, respectively. The optical character recognition system
is the SVM integration with different character features, whose
performance for numerals, Kana.
INTRODUCTION
LICENSE PLATE RECOGNITION (LPR) plays an important
role in numerous applications such as unattended
parking lots [1], [3], [4], security control of restricted areas [5],
[6], and traffic safety enforcement [8], [9]. This task is quite
challenging due to the diversity of plate formats and the nonuniform
outdoor illumination conditions during image acquisition,
such as backgrounds [2], [10], illumination [7], [15], [22],
vehicle speeds [11], and distance ranges between the camera
and the vehicle [19]. Therefore, most approaches work only
under restricted conditions such as fixed illumination, limited
vehicle speed, designated routes, and stationary backgrounds.
LICENSE PLATE PREPROCESSING
License plate preprocessing is a necessary step in LPR,
which includes plate detection, correction, and segmentation.
The goal of detection is to locate regions of interest that are
similar to the license plate. Due to the angle of orientation, the
image may have a slant and distortion; thus, transformation or
correction of image is an important step before the character
segmentation.
Japanese License Plate Description
In this paper, a Japanese license plate is the object of study,
and some samples are presented in Fig. 1, followed by a brief
description of the proposed Japanese LPR process. The images
in the first row are captured by different illuminations, which
include daylight, night, shadow, and exposure conditions. The
second-row images indicate that the license plate images may
be colored or gray. The third-row images show the images captured
from various angles of orientation. The image resolution
may be low in the last row images due to the filming equipment.
Two types of the license plates should be considered, i.e.,
white characters/black background and black characters/white
background.
License Plate Location
The brightness distribution of various positions on a license
plate image may vary due to the condition of the plate and
the effect of the lighting environment. Since a binary method
with a global threshold cannot always generate satisfactory
results in such cases, the adaptive local binary method is often
used. The local binary method means that an image is divided
into m × n blocks, and then, each block is processed with
the binary method. In our research, two local binary methods
were adopted, i.e., the local Otsu and an improved Bernsen
algorithm. Otsu [36] is a traditional binary method, which we
used on each subblock. However, the performance of Otsu is
contingent on the illumination conditions, which greatly vary.
To resolve the uneven illumination obstacle, particularly for
shadow images, we proposed a novel binary method, i.e., the
improved Bernsen algorithm.
Character Segmentation
Real license plate images are prone to slant and distortion
due to different angles of orientation. Therefore, horizontal and
vertical correction and image enhancement are required prior to
character segmentation.
1) Horizontal Correction: In this paper, the size of the
image varies. For example, the size of a small image is about
40 × 80 pixels, whereas a large one is about 200 × 400 pixels.
First, all detected license plates needed to be resized to 100 ×
200 pixels.We then used the connected component technique to
detect large numerals and find the center position of each large
numeral. Next, we computed the tilt angle of every two central
points, and the average tilt angle was obtained. Finally, we
adopted a 2-D rotation method to correct the image according to
the average tilt angle. Fig. 13(a) shows the horizontal correction
method. The left column images are the detected license plates.
The middle column images are the resized images. The blue
frames are the large numerals detected by CCA, and a red line
across the centers of the numerals is the horizontal tilt angle of
the image. According to the tilt angle, the license plates can be
corrected [shown in the right column in Fig. 13(a)].
DISCUSSION AND FUTURE WORK
Compared with most previous studies that, in some way,
restricted their working conditions, the techniques presented
in this paper are much less restrictive. The proposed LPR
algorithm consists of three modules: 1) locating the license
plates; 2) segmenting the characters; and 3) identifying the
license characters. There are several commercial LPR systems
whose evaluation is beyond the review capabilities of this paper
due to the fact that their operation is strictly confidential, and
moreover, their performance rates are often overestimated for
promotional purposes.