License Plate Recognition Algorithm Based On Sparse Representation

License plate recognition technology is one of the key techniques in the intelligent traffic management, which has become a hot point in the field of research. Although the license plate recognition technology has been rapidly improved in recent years, the development of this technology is still facing many difficulties. So It has significance that how to achieve the effective license plate recognition under the complex conditions.This paper takes the license plate as the research object, and a comprehensive study in three key technologies such as license plate location, license plate binary segmentation and license plate recognition has been done. At last, based on the sparse representation theory, the stable and accurate full automatic license plate recognition system has been built up. The main work of this paper is as follows:Firstly, the license plate location method is proposed based on the multi features combination. The coarse location is done using HSV color feature of license plate, then, the Canny feature is used for scanning of the license plate and the license plate is located by using the rich edge features. Experimental results show that our method can locate the license plate accurate in the similar color between the car and the plate background.Secondly, the improved PCNN model and parameters setting method was proposed. The concept of image gray level histogram is defined, the parameters of the pre-setting threshold are determined and the modulation coefficient of the improved PCNN model is determined according to the determined pre-setting threshold. The improved PCNN model is applied to the license plate image binary segmentation. The experiments results show that this method can automatically complete the license plate image segmentation clearly and consistently, enhance expression ability and discrimination of character feature extraction and promote the overall recognition performance of the algorithm.Thirdly, a license plate recognition algorithm based on the structured dictionaries learning of sparse representation is proposed in which plate character segmentation is not needed. On the basis of the prior knowledge of the license plate character, the structure sub- dictionaries are built with HOG features extracting by using KSVD algorithm. A sparse indicator function for each sub-dictionaries is used for the character recognition. Finally, The recognition result of the whole license plate is obtained in series with the characters identified by each dictionary. Experimental results on the public license plate test data set show that, the proposed algorithm not only can improve the recognition rate on the premise of normal real-time, but also has strong robustness to the adverse factors such as light, license plate deformation and so on. And because there is no need for single character segmentation, our method can improve the efficiency of the license plate recognition algorithm while simplifying the steps of the license plate recognition, which makes the whole license plate recognition more efficient.