License Plate Character Segmentation And Character Recognition Algorithms And Realization

Along with the development of the economic and the progress of the society, cars have become a means of transport, and more and more cars also cause road traffic congestion. How to solve the problem of road congestion has become a problem to be solved cannot wait. Intelligent Transportation System came into being in this context, and motor vehicle license plate recognition system plays an important role in the ITS. It can be applied not only in the parking and toll station, but also applied at the crossroads. It not only can bring good economic benefits, but also can bring good social benefits.This paper uses digital image processing,pattern recognition and computer vision technology, and proposes the license plate recognition system at the crossroads. Main tasks are character segmentation and character recognition.a. Character segmentation. After locating license plate, we can subsequently deal with character segmentation. Because the located license plate may tilt, so before character segmentation, we need to correct the tilted plate. We can detect angle by Hough transform, and correct the tilted image by bilinear interpolation. Before character segmentation, we need to locate license plate further, and remove the border of license plate. We take the method of the sum of lines to remove the upper and lower border, and use vertical projection to remove left and right border, then we can get a license plate without of noise and interference. We take gray projection method based on separator, with the features of characters of license plate. We firstly segment five characters after the separator, then segment two characters before the separator. This approach has good robustness, and can be achieved very good results to the border removed not completely.b. Character recognition. Characters of different sizes make it difficult to identify, so character normalization is done before character recognition. Because boundaries between the licenses plate characters and the background is blurted. It will be difficult to get useful information from binary image. We will get gray-scale image used as support vector machine training samples. We make the normalized gray-scale images of characters end to end by line. We change the matrix of characters image into row vector as eigenvector. We take one-to-one multi-classification support vector machine, and combine the features of characters of license plate. We divide into three classifications, the character classifier, letter classifier, number and letter classifier. This can not only improve the character recognition rate, but also save recognition time.