Based On BP And CNN License Plate Recognition

License plate recognition has become the focus in the intelligent transportation, the application of license plate recognition system is becoming more and more widely.In this paper,introduced the four parts:Vehicle image preprocessing, license plate image positioning,license plate character segmentation and license plate character recognition in detail,which made of the license plate recognition system.License plate character recognition is an important indicators to measure the performance of license plate recognition system.,which similar license plate character recognition is the key factor that affect the recognition rate of license plate recognition system. This article focuses on the license plate character recognition, this paper puts forward a new methods,which the license plate character recognition is based on BP and CNN in stages and divide the license plate recognition process into two stages. First stage to classify the segmentation of license plate character image, which is divided into Chinese characters, dissimilar letters and Numbers, similar letters and Numbers in using BP neural network. According to the character specification of Chinese normal license plate,the total number of 13 similar letters and Numbers are 0,D,Q,U,8,B,3,2,5,S,Z,C and G in common,besides these 13 similar letters and Numbers, the rest of the letters and Numbers are classified as a dissimilar characters. For Chinese characters, dissimilar letters and numbers is identified by the BP neural network directly. The similar letters and Numbers into the second stage, which is based on the improvement of deep neural network (CNN) for identification.For various interference factors, the license plate images will appear fracture character, stain keep out, and so on in the real.The neural network (CNN),which is a good combination of three kinds structure thoughts:local receptive field, space sub-sampling and share weights, having the shift and scale invariance, and it is invariant to distortion in a certain egree, showed a good distortion fault-tolerant function. Otherwise compared with the general neural network,the CNN’s network topology has a high coincidence,and the CNN has a high efficiency in pattern classification and less training parameters. Therefore, this article using CNN deep neural network identification in the second phase of license plate character recognition. According to the characteristics of similar letters and Numbers, in this paper, increased the hidden layers and modified the parameters on the deep neural network (CNN). The designed license plate recognition system is based on MATLAB platform, realized the function of license plate recognition. Experimental results show that the character recognition rate to similar characters of license plate increased significantly using improved deep neural network (CNN), the whole recognition rate of license plate recognition system has improved considerably.