The Research Of License Plate Recognition Algorithm Based On Character Multi-feature Extraction

License plate recognition is an important part of intelligent transportation systems, based on ease and reduce road accidents and improve traffic utilization for the purpose.License plate is determined uniquely in public car identity credentials. The correct identification of the car license, can achieve at traffic junctions, highways, authorities guard against and from real-time vehicle registration, traffic statistics, and can help automate billing; on the vehicle, stolen vehicles, vehicle crime identification and interception, which is building an integral part of intelligent transportation systems.This paper researches the license plate recognition technology on the license plate character segmentation, character multi-feature extraction and character recognition character. This paper work has been completed as follows:License plate character segmentation. First, image preprocessing for the license plate,The discrimination types of it with histogram,its tilt correction by hough transform, removing the border with projection, it is license plate coarse segmentation and fine segmentation with intrinsic characteristics and vertical projection combination.License plate character feature extraction. In depth analysis of the characteristics of alphabetic and numeric characters, straight lines of letters and numbers、the character strokes%rings and ring area of four features are proposed, given these feature extraction methods,these features can accurately distinguish all numbers and letters on the license plate.Due to the complex structure of Chinese characters and multiple strokes, The13point characters features of combining the block size with the scanning line to extract Chinese features,and from10to13point feature improved, improve the recognition rate of the similar Chinese characters.License plate character recognition. The BP neural network is used for license plate character recognition in this paper.Inadequate for the BP neural network algorithm, use additional momentum and adaptive learning rate to improve its. Established four sub-networks are acyclic network, a ring network, two ring networks, and network characters. Simulation experiment achieves good results, the license plate character recognition rate reaches98.9%.