Research And Application Of License Plate Recognition Algorithm Based On Improved LM-BP Neural Network

License plate recognition system is an important part of intelligent transportation system, and it is also a hot spot of domestic and foreign scholars’ research. By preprocessing of license plate location and extraction, the license plate image processing, license plate feature extraction, license plate character recognition technology to identify vehicles grades, color and information. It also combines many techniques such as image processing and pattern recognition, which can be widely used in the field of highway automatic toll collection, road monitoring, community and vehicle management system. In the practical application, license plate recognition still exist many bottlenecks, such as rainy day, fog and other complex environment under low recognition rate and low speed; even the license plate recognition is not ideal in situation of exposure degree is too low or too high, etc. In order to solve these problems, this paper makes a deep study of license plate character segmentation and character recognition, puts forward some improved algorithm, and uses the VC++6.0 platform to develop the license plate recognition system in this thesis and the related experiments are carried out.In this paper, a variety of license plate recognition algorithm based on the analysis, and based on this proposed the corresponding improved algorithm, while the license plate image preprocessing, tilt image correction and other content. In order to eliminate the pollution, improve the binarization accuracy, this paper presents binarization algorithm based on the virus evolutionary genetic, which is a kind of improved genetic algorithm, in order to maintain the diversity of population and the virus mechanism combined with the genetic algorithm. The adaptive threshold Canny edge detection algorithm for edge detection, and match the Radon transform method, in the detection to the edge of the information to accurately draw the license plate tilt angle, the correction in order get the correct image. On this basis, in order to reduce the blurring of the license plate image segmentation error rate, this paper improve the vertical projection of the segmentation method. In order to improve the complex environment (rainy, foggy and dim, etc.) of the character recognition rate, this paper presents a kind of improved LM-BP neural network recognition method of license plate characters, the method according to the characteristics of China’s current domestic license plate system, combined with LM algorithm to improve the traditional BP neural network, and increase the parameters modified Levenberg Marquardt algorithm, which can make the traditional BP neural network slow convergence speed and easy to fall into local minima of the two disadvantages are solvedFinally, the paper carries on the related test to the vehicle license plate recognition system. Test results show that the system can effectively achieve the license plate location, character segmentation and recognition, to achieve the desired recognition results and convergence rate, the overall recognition rate of the system reached 95.66%.