Design And Application Of License Plate Recognition Based On Image Processing

License plate recognition system is a very important part of intelligent traffic system, and plays all important role in electronic policeman, electronic toll collection system, parking lot management, and registration of vehicles in certain regions. This paper focuses on stationary scenes like freeways and urban roads, and proposes some practical solutions to tilt correction, binarization of image and character recognition. And the main contributions of this paper are as follows:First, some preprocessor is done, which mainly includes the image grayscale, the image enhancement, and the image binary-coding. Considering the License plate image noise is mainly impulsive noise, a de-noising method of self-adaption median filtering is proposed. This paper uses LoG operator to get the binary image of the License plate.Secondly, the License plates in our country are classified theoretically, Mainly introduces the License plate localization based on texture feature segmentation algorithm, doing edge detection with the License plates. And a method based on math morphology and color characteristics has been realized. Toward oblique images, an oblique angle is calculated by using Hough, and then the License plate image is rotated to correct direction.According to the segmented License plate area, and on the basis of the existing character segmentation method, a new way to complete the character segmentation is presented here by two steps:the first step is to remove the disturbances of the borders and rivets, then to precisely locate the upper and lower boundary of the characters. The second step is to combine the outline characters with the proportion information of the License plate and to locate each character about the boundary, and then to realize the License plate character segmentation.Finally, templates matching algorithm is used in character recognition. The character being recognized is contrasted with the template that is structured in advance. The ultimate recognition results are determined by the similarity with the template.The experimental results show that:the precision is more than 95% by using the method in this paper.