Research Of Character Recognition Of License Plate

License Plate Recognition (LPR) system is one of important part of the intelligent transportation system. It has been applied to the areas of illegal vehicle monitoring, automatic charging in high-speed road junctions, traffic control and guidance,vehicle management and security alar.According to the current problems of low recognition rate and high complex computation,this paper uses the methods of combined features and chain code to recognise characters. The main contents are as follows:(1) License plate pretreatment and character segmentation.This paper realizes license plate location by using the color information of license plate. The processing including image enhancement, skew correction, binarization, etc.After removing the frame of license plate, this paper uses the methods of based on plate region prior knowledge and statistical pixel to achieve a single character segmentation.(2) Feature extraction of license plate character.Normalization processing of the characters before feature extraction.This paper puts forword combined features as the features of license plate character,that is the combination of grid feature,projection feature and structural feature. By experiment it determined the best combination of the three weight factor.In this combined state, t recognition rate of the system is the highest.(3) Character recognition based on combined features. This article has designed 3 layers BP neural network used for license plate character. Combined features under the best combination of weight coefficient are the inputs of the neural network,using PSO algorithm to optimize network structure and improve network convergence speed. Experimental results show that recognition rate is high and convergence speed is fast by the combination of combined features and PSO-BP network for character recognition.(4) Character recognition using boundary chain code. This paper extracts the chain code feature of characters by the method of boundary chain code based on morphological thinning. According to the differences of chain code for the license plate character, it designs a multi-level classifier. Experimental results show that the chain code features can effectively reflect the morphological features of the character.Feature extraction method is simple and memory capacity is small.(5) This paper holds that the method of combined feature recognition has a strong anti-interference and fault tolerance,but the chain code features have a great influence by the character size and the degree of tilt.Generally, the first method is better than the second.