Recognition of license plate character based on wavelet transform and generalized regression neural network

Recognition methods of Chinese characters and similar characters are the main factors which affect the rate of license plate character recognition system. In this paper, a hybrid method based on wavelet transform and generalized regression neural network (GRNN) is proposed. For Chinese characters, a block projection with wavelet packet transform is adopted to extract the feature and a clustering algorithm is presented to reduce the dimension of feature vector. Moreover, the method of the characteristic regional recognition is introduced to recognition the similar characters. Besides, the clustering method based on wavelet transform is used to extract the feature of letter and digital characters. Furthermore, GRNN with powerful non-linear mapping capability and high fault-tolerant ability is designed as character classifier in the multi-network character recognition system. Finally, the result shows that the recognition rate of whole network could reach up to 91% and the classification algorithm has better robustness.