Research On Methods Of Automatic Recognition License Plate Characters

License Palate Recognition(LPR) system is a special computer intelligent system for car plate.It is an important research subject in the field of intelligence traffic field on the base of computer vision and Image Processing.This system can be used to vehicles management,automatic charge,avoid run a red light and the safety of vehicles,traffic flow control,traffic control and induction in airport station,harbor, sub district.It has a bright future.Generally, The software system of LPR mainly comprises modules of image preprocessing, plate location,character isolation,tilt correction and character recognition.As a result of experiment and time limit,The main designs of this paper are as follows: plate location,tilt correction, character isolation,character recognition. The thesis brings forward some effective strategies and completes some experiments basing on these five modules.①Image preprocessing: It introduces briefly some image preprocessing technologies used in this paper, including image gray processing,image enhancement, mathematical morphology, image binarization, edge detection, etc. Their experimental results are also showed.②Plate location: An algorithm based on mathematical morphology and projection is used to locate the license plate after extensive analysis and comparison on the existing methods. Image preprocessing is first adopted, which is implemented by Top-Hat transformation, mathematical morphology based on gray image. Connected components analysis is then performed to detect candidate regions. At last, the real license plate areas are obtained according to the projection of the candidate regions. Experimental results show that it can realize the fast and accurate location and segment the license plate from the complex image.③Tilt correction and characters segmentation: On the base of exact plate location using minimum variance of edge projection and Radon transformation to correct the tilted plate. After that, an algorithm based on projection dichotomy and prior knowledge is used to segment the characters. Experimental results show that the method can realize the fast and accurate segmentation. In addition, this method is not sensitive to the character adhesion, character fracturation, and the character tilt.④Character recognition: Considering the limited samples for experiment and the shortcoming of the traditional methods, an algorithm based on the combined features and SVM was proposed for license plate character recognition. First, the features of plate characters were detected by one dimensional wavelet transform and Pseudo-Zernike moments. Then by bilinear search method the features were used to train the SVM classifier that the leaning is the best than others. Finally, the plate characters were classified by the SVM classifier. A large amount of samples are used for system test. Overall recognition accuracy is 97.6%. Through contrastive research, the method presented is proved to be effective and advanced.Based on research of the theory,this paper has realized the corresponding algorithms using MATLAB programming tool.