License Plate Detection And Recognition Algorithm Research

License plate recognition(LPR) is an important research topic of digital image processing and pattern recognition emploied in intelligent transportation system (ITS), LPR system is significant to save the lives and the properties of the chauffeurs and the passengers and to improve the stability and the authority of the law execution in transportation.This paper accomplishes the research and design of LPR system's algorithm. Firstly, the paper introduces the research status of the LPR technique and analyzes the basic theorems and primary technique of image processing, and on the base of them, it dose the main research in the algorithms of license plates detection, tilt correction, binarization, character dividing, character recognition in the LPR modules.The paper presents a new method of license plate detection based on fuzzy set by employing color characteristics and texture characteristics of the license plate in the module of license plates detection. The proposed method applies color-pair zone detection, color-pair edge detection, chrominance maximum statistic and intensive small edge enhancement to input images, and the results are aggregated by fuzzy set operators so that the license plates could be extracted accurately. The method was employed to images taken under various conditions and the results show that the average accuracy of license plate extraction is 98.0%. Especially, the method is also robust to images with poor illumination or complex backgrounds. Also the paper presents an improved method of license plates tilt correction, which combines edge detection and rotation projection to calculate the tilt angles and corrects the titling images of the license plate steadily by bilinear interpolation rotation. In the binarization module of the license plates, the paper presents an improved method of Otsu binarization threshold value by using the concept of the P method for reference and the thought of bringing the histogram morphology. At last, the paper adopts the two stage multi-mode recognition scheme to the divided characters of license plate based on the global features, which use brim distance detection and the image match method based on mismatch weight penalizing match model recognize the characters of license plate efficiently. The results show that the accuracy rate of the algorithm's character recognition can reach to 95.5%.