Abstract: Traditional character feature extraction methods have such shortcomings as complexity,large calculating and low degree of differentiation.In order to improve the rate of license plate character recognition and overcome the presented drawbacks,this paper introduces a grid-based secondary character feature extraction method.Firstly,the normalized character images of license plate are divided into four grids and the curvature feature is extracted;secondly,the normalized character images are divided into 32 grids,followed by the extraction of the duty ratio of character pixels,center of mass,divergence,three quantitative characteristics to describe each grid.This method combines the advantages of structural feature and statistical feature.It is easier to distinguish characters which are similar in structure of strokes.What's more,it has strong anti-interference ability and enough stability.Features from 1500 different types of normalized character binary images are extracted by different feature extraction methods.The results show that the recognition accuracy of letters and numbers can reach more than 97% by the proposed method,a significant improvement over that achieved by other methods.

Key words: character recognition;feature extraction;gridding;normalization