Research On Arithmetic Of Car License Plate Recognition Based On Neural Network

In the modern Intelligent Transportation System (ITS), the License Plate Recognition (LPR) is one of the most critical technologies. It has a very broad application prospect in electronic policeman, electronic toll collection system, traffic flow control, parking lot management, stolen vehicle inquiry, etc. In recent years, the research on the license plate recognition system has attracted more and more attention by all over the world, then it is a hot issue to improve the speed and accuracy of the license plate recognition algorithm.The paper mainly bases on the image pretreatment, the license plate location, tilt correction, character segmentation, feature extraction and character recognition process of the car license plate character recognition, and has carried on simulation in the environment of MATLAB.Firstly, in the part of image pretreatment, the paper has fully used digital image processing technology and math morphology technology, and has used a binarization method which is based on the self-adaption original threshold and the Ostu method.Secondly, in the location process, a method based on scan line and plate features will be used; since tilt plate image will have an adverse effect on the following character segmentation and recognition, the paper has used one tilt correcting method based on the line features of character for license plate image tilt correction.Thirdly, in the part of character segmentation, the paper has used a method based on the gray change property and vertical projection threshold to eliminate the borders of plate image; then the paper has used a method based on vertical projection, and combining with the priori knowledge of plate.Finally, in the part of character recognition which BP neural network method is used, a feature extraction method based on the pixel's number feature of character projection and the thick gridding feature will be used, then designs BP neural network using character projection's pixel feature and the thick gridding feature to recognize.