基于机器视觉的分级车牌字符识别方法

为提高车牌字符识别率,提出一种考虑整体和局部特征,分别采用两级SVM分类器的识别方法,其工作模式为:第一级分类器针对所有字符,在识别结果属于形似字符的情况下,送入对应的第二级分类器作进一步识别。第一级分类器提取字符图像整体的各网格比例作为SVM的分类特征。将形似字符分为5组,分别对应的5个SVM构成第二级分类器。通过分析各组内字符笔画特征的局部相异性,提取相应网格中字符轮廓的垂直、水平投影方差、比例等特征并进行特征融合作为相应SVM分类特征。实验结果表明,该方法字符平均识别时间为23.45 ms,且在形似字符的识别率和总体识别率方面均优于模板匹配、神经网络和同类的分级识别方法,是一种有效的方法。

Method of hierarchical license plate character recognition Based on Machine Vision

To enhance the license plate character recognition rate, a method which uses a two-stage classifier of SVM (Support Vector Machine) is proposed, based on the whole and local features. The first-stage classifier aims at all characters. Send the characters to the corresponding second-stage classifier for further recognition if their identify results belong to the confused characters. The first-stage classifier extracts the whole grid rates of the character images as the classification features of SVM. The confused characters are divided into five groups, and then five corresponding SVM constitute the second-stage classifier. Through analyzing the local differences of the character stroke features in each group, extract the features like vertical projection variances, horizontal projection variances and proportions, which belong to the character outline of the grids. After that, process them with feature fusion to make up the classification features of SVM. The experimental results show that the recognition time is 23.45ms. The method has higher recognition rate of the confused character and the overall recognition rate than the template matching methods, neural network approaches and other previous hierarchical recognition methods.