A Novel System Design of License Plate Recognition

In this paper, we have proposed a novel approach of license plate recognition system by adopting comprehensive features of license plate. Firstly, we originally combine the stroke width of license plate character with specified colors of license plate to segment the license plate region. Then, we use linear fitting method and projecting method to rectify the position of slant license plate. Furthermore, we utilize connecting region searching method to segment English and digital characters, and creatively utilize relocation method to segment the Chinese character in order to reduce distortion. At last, we adopt number of holes and peripheral distance curves to recognize English or digital characters, and creatively adopt affine moment invariants and stroke statistic feature to recognize Chinese characters. In the experiment, we test our algorithm using lots of car images having various backgrounds and different illumination situations, and the accuracy of license plate segmentation reaches as high as 96.5% and the average accuracy of license plate character recognition is 93%.