License Plate Recognition System Is The Key Technology Research

License plate recognition (LPR) is one research hotspot in the field of pattern recognition. It is also an important component in the intelligent transportation system and has a very wide range of applications.In a typical license plate recognition system, after image/video acquisition and preprocessing, license plate is located, and then character in the license plate is segmented, finally the plate characters is identified. In this thesis, license plate location, character segmentation and recognition technologies are mainly studied.In the license plate location stage, a novel method is adopted which combines edge analysis and color statistics. The framework is divided into four parts, preprocessing, coarse position, precise position and non-plate rejection. In order to improve the speed of vehicle license plate location, one improvement is carried out, which is based on the combination of SVM classifier and the aspect ratio for false license plate elimination.In the char split stage, a novel approach for complex illumination license plate binarization method is adopted. In this method, the R, G, B color component redistributed to construct a new and more proper gray image, according to the characteristics of domestic color plates. Then, the spatial homomorphic filtering techniques ia used to enhance the robustness under different lighting conditions, and the binarization is fulfilled with the combination of global and local thresholds.To resolve problems such as incomplete or inaccurate license plate detection, a boundary movement method is proposed based on the feedback of the character number and the character width.In the char recognition stage, two kinds of char recognition method are studied. First, template matching recognition algorithm is studied, witch is widely used and has good real-time performance, while which has low recognition rate. Then, sparse representation based char recognition method is explored, which has relatively high recognition rate but poor real-time performance.Finally, a LPR system was accomplished on the VC++platform by MFC tools, which verifies the effectiveness of our methods.