License Plate Location In Complex Background Based On Texture And Color

Vehicle license plate recognition(VLPR),which is an important part of modern intelligent transportation system(ITS), plays an important role in the modern traffic management, traffic monitoring, vehicle inspection and so on. So, it has a broad prospect of application. License plate location is the first step and the key technology of VLPR system. And the accuracy and stability of it have a profound impact on the whole VLPR system. After many years of research of scholars’, license plate localization technology has become the hot topic. And based on the characteristics of the vehicle license plate, there are lots of methods. But in the complex background, there are so many Objective factors, such as use time of license plate, the weather and illumination conditions,variety of environment and camera hardware, which all will have an unpredictable influence on Image quality and plate regional characteristics. Considering the complexity of image background and the uncertainty of plate position and image quality, it is necessary to do further research on it..By summarizing the development and the latest research achievements in the field of license plate location both at home and abroad, the current license plate localization technologies are divided into two kinds of methods,including locate method based on license plate structure characteristics and locate method based on the license plate color characteristics. Through comparison, this paper has analysised the advantages and disadvantages of these two methods. And then, taken these two kinds of method into account, we propose a system which combines texture and color to extract license plates in complex natural images in this paper. The proposed method is novel, as it exploits modified k-means clustering algorithm in RGB color space, which allow the number of color and cluster centers to be determined automatically, instead of definition color range. The advantages of this method are that first rule out color interference area by exploiting texture, then use color clustering to rule out the texture interference area, and overcome the instability and weak flexibility of quantified definition color.The method of this paper consists of four parts, including image preprocessing, coarse location based on plate texture, fine location based on color clustering and license plate screening. First of all, we change the color image into gray level,and remove nosie,enhance the gray mage. Then, through twice jump scanning, we firstly extract the horizontal position of license plate; secondly,we extract the vertical position by analysising characteristics of vertical edge and texture dense area from the vertical position. Come to this step, we get the coarse position of license plate,which can be called candidate areas. Then, these color candidate areas will be mapped to RGB color space. We regard the intensive degree of RGB three-channel in space as color similarity measure to improve K-means clustering algorithm.This improved algorithm is applied to get accurate license plate area. Finally, we eliminate the interferenced regions and select license plates based on the character feature of SWT and syntactic efficiently.All of our algorithms have been proved by combine vc++6.0and Matlab in the software platform. Experiments have been conducted on a large number of car images,including1130images got by traffic video and104images got by mobilephone camera, taken from various scenes and different conditions. The result shows that the rates of the two kinds of license images achieve99.646%and98.077%, the average times of locate one image is0.179s and0.921s. Experiment demonstrate that this method can keep high accuracy,strong stability and robustness,and can detect license plates of arbitrary orientations and different illumination condition.