Study On Technology Of Vehicle License Plate Auto Recognition System Using Fuzzy Theories

With the fast development of electronic computer, artificial intelligence technology and image processing technology, the ITS (intelligence transportation system) become the new research field. The LPR (license plate recognition) technology is an important part of ITS, it's a hot field nowadays.The LPR technology is based on the hardware and software platform of digital image acquisition devices and computer information system, includes image data acquisition, preprocessing of image, extraction of license plate image, segmentation character image and recognition of the character image. This article mainly focuses on the methods of preprocessing of image, extraction of license plate image, segmentation character image and recognition of the character image.Because of the different illumination conditions, the article raises an image binarization method based on multi-thresholds Otsu algorithm, and good results turn out; the extraction of plate image process, including rough localization and precise extraction, is a difficulty of LPR, so a match algorithm based on chaotic theory brings up, tries to get a more accurate plate region; in the rectification of plates'image, there're two algorithms– one uses the combination of Radon Transform and Max Zero-row Count, the other one uses the Bilinear Transform to rectify the slant and distortion of image samples; with the good image samples for recognition, after segmentation of characters'images, a simple fuzzy pattern classifier would attain a nice recognition rate.The sources of the original image samples are from the following aspects: one is captured by DC and CCD installed at cross-roads or depots, the samples are colorful image of stationary or slow moving vehicles; the other one is downloaded from internet, some samples are captured by traffic monitors. The samples are captured at different time and angle, with various vehicle colors, size, illumination conditions, and there're also distortion or fuzziness samples. These make the research more difficult. After applying the steps given in this article to process hundreds image samples, the recognition correct rate would turn to be at a high level.