The Research Of Vehicle License Plate Recognition Algorithm Based On Confusable Characters Discriminant Neural Network

With the development of intelligent transportation and information technology in the21st century, computer aided become the scientists research direction when people deal with the traffic problems. Follow by car sales and traffic growth in our country, dealing with road accident and violation of traffic police by using traditional methods have become impractical. Intelligent transportation technology is introducing by information technology and control technology and computer technology, and using the advanced public transportation system, advanced traffic information service system, advanced vehicle control system, as well as electronic toll collection system, the emergency rescue system in intelligent transportation system.As a vehicle management sub-system of intelligent transportation system, the discriminant vehicle license plate becomes an important part of transportation. License plate recognition system mainly rely on the computer image processing technology, pattern recognition and intelligent computing technology. The license plate images are extracted from the images of video stream, and is segmented and recognized by the characters.License plate recognition technology still exist many difficulties, such as license plate scraping, denoising of license plate, character recognition and system performance requirements, and so on. In a1080P picture which is captured on standard4meters high camera, the license plate part only had about120*35pixel size. In this pictures, there are a large number of natural background and vehicles and interferences. There are also a lot of noise in license plate images, including overexposure, tangerine, license plate frame, plate stained noise and so on.This paper mainly research to the domestic and foreign research methods recently. After choosing appropriate license plate extraction, character segmentation, character recognition algorithm, and improving on the original method, the recognition average time is within300ms, digital and English character accuracy is above95%, the license plate recognition rate is about80%.Research and improvement of the paper mainly in the following aspects:(1) License plate extraction ways:After Using the Sobel operator vertical edge detection and noise elimination to remove most of the vehicle noise and environmental noise, we used the improved binary image fast rectangular to identify undetermined position of license plate, and used the license plate rectangular feature and color feature to find properly license plate location.(2) Character segmentation:After stretch in gray and binarization, the license plate images get more clearly. By using simplified projection features and registration template, and using certain fault-tolerant algorithm to divide characters correctly.(3) Character recognition:By comparing the two kinds of character image processing and feature extraction methods, we finally choose the coarse grid features character feature extraction and the projection feature extraction method. By using two simple structure of the neural networks to identify the Chinese character, English and Numbers, and through the confusable characters discriminant neural network to distinguish the confusing characters. This method can improve the recognition rate and speed of recognition characters.