Research On Character Segmentation In License Plate Recognition System

The study of Intelligent Traffic System(ITS) is increasingly becoming a focus with the development of science technology and the construction of urban traffic. License Plate Recognition(LPR)technology is one of the most important parts in the Intelligent Traffic System(ITS). License plate recognition technology has important practical value in many aspects of the traffic management, such as the charging system in highway and parking lots, the management of intelligent zone about entering and leaving, the querying and tracking the target vehicle and so on.Typical license plate recognition technologies mainly include three key parts: the license plate location, character segmentation and character recognition. This paper focuses on the license plate segmentation techniques’ study. Some analysis and new methods in the process of plate segmentation are displayed.In the process of license plate character segmentation mainly have four parts which is license plate image preprocessing, image binaryzation,skew correction, character segmentation. In the image pre-processing, the thesis adopts a dual histogram equalization method that based on the exposure threshold, which can effectively deal with the darker picture and enhance the license plate image contrast. In the process of image binaryzation, a binaryzation method which is based on the Otsu method and minimum error classification method is adopted. The key of this thesis’ binaryzation method is setting up a selection mechanism which takes the larger one as the threshold binarization threshold. This strategy can enhance the adaptability of the binarization method and get a good result. In the process of tilt correction, the thesis adopts the method with domain analysis and Radon transform. It can solve the problems that the characters adhesions together and the license plate frame obscure. In the process of character segmentation, the thesis adopts the method with vertical projection and Connectivity domain analysis. This method first locates the second and the third characters through the prior knowledge of license plate and vertical projection, then segments the part that is all the character of the left from the second character with vertical projection and the other part that is all the character of the right from the third character with connectivity domain analysis. The results show that this method can deal with the unconnected of Chinese characters.