The Research And Implementation On Character Recognition Of License Plate Recognition

License Plate Recognition system is the core of Intelligent Traffic System. It is very important in the toll station, automatic payment highway fee and parking lot management, etc. LPR contains three sections: license plate localization character segmentation and character recognition. Character recognition is one of critical technology problems in LPR, which is focused on in the field. The paper has gone deep into character recognition and its corresponding technique, the main research is presented as follow:1. A location method which based on the region shape of plate and the gray change property is adopted to realize license plate location. An optimum algorithm is presented based on characters vertical projection distance for'slant correction. After rim cleaning, Singled character is picked out by projection method combining with the transcendent information.2. After the processing of smoothing and normalization, an optimum elastic meshing feature is extracted for English alphabets and digital characters, the directional line element feature is proposed based on the stroke feature extract method of the weighted congruency meshes.3. Improved BP neural network has been applied into character recognition. The research of how to design network structure,the training samples and parameters setting has been done. Momentum gene,a self-adaptive learning rate and the steep factor have been imposed to improve the defects of BP network. Experiment result shows that this algorithm can converge to optimum spot, improve the speed of convergence and the recognition nicety. 4. The research of SVM recognition is introduced. Construct Multi-classifier based on "one-against-all", an algorithm of SMO is adopted to train samples. Different kernel functions and parameters are tested to compare its performance. The method of SVM can get higher recognition rate without feature extracting.