Research On Several Problems In License Plate Recognition System

License Plate Recognition System (LPRS) is the most important link in ITS. It plays a very important role in the traffic supervision, traffic control, and vehicle management and so on. The capacity of LPRS directly affects the development of ITS.This paper is mainly devoted to study of License Plate Recognition System. It researches in terms of image acquisition, pretreatment, license plate location, character segmentation and character identification.In the part of image acquisition, I give a brief introduction on the features, differences and categories of license plate in our country. And I can take high-quality photographic of vehicle through setting up trigger and designing the hardware of license plate location.In the step of pretreatment of image, first we can transfer color images into gray scale images, and then binarize the images using the adaptive threshold of gray difference. Finally we can obtain successful binary images even at night. By comparing the traditional method and method of the adaptive threshold of gray difference, we can find that the method we use in the paper can create clearer and high-quality binary images. We can also use the model of cross dot to flat the binary images and remove the interference factors.In the step of license plate location, we can analyze the characteristics of the binary images and scan these images from bottom to top, measure the number of continuous leap points, and set the maximum scanning distance, continuous leap lines and length-width ratio, then combine the location of first character and the correction of slant plate we can find the location of the license plates correctly. If we combine print tab-location with slant correction of vehicle licenses, we can obtain the prime quality images.Character segmentation carries out two different methods to binarize the gray scale images, and enhance gray images to those vehicle licenses whose characters are unclear. So that the characters in these images can keep stroke linkage. Then we set the characters color to black, and the background color to white, and give them a specific treatment deleting short line segments and connecting region-dependent segmentation. At last we use BP neural networks to recognize binary characters and unit the putches. Chinese character, character and number are recognized by neural networks singly. Based on the analysis on input and output vectors and features of characters, we can create the structure of neural networks and software realizing method.