基于SVM的车牌字符识别算法研究

SVM可在训练样本很少的情况下获得很好的分类推广能力。首先分析了用多类SVM算法对车牌中的字符进行识别时存在不可区分的区域问题和采用模糊SVM算法解决该问题的办法，然后讨论了字符特征的提取方法，并根据我国车牌字符的特点分别设计了汉字、字母、数字、字母／数字4个基于模糊多类SVM的字符分类器。最后在MATLAB环境下，采用径向基核函数对算法进行学习训练。实验测试结果表明，该方法可以很好的提高字符识别的速率和效率。

License Plate Character Recognition Algorithm Research Based on SVM

Good classification and generalization abilities can be obtained through using SVM algorithm when there are very few training samples. Firstly, the indistinguishable problem which is caused by multi-class SVM to recognize license plate characters is analyzed, and it can be resolved through fuzzy SVM; then the extracting character features method is dis- cussed, and four characters classifier is design respectively based on SVM which are Chinese character, alphabet, digit, al- phabet/digit classifiers; finally, the algorithm is trained through radial basis kernel function under MATLAB environment. The experiments show that this method can well improve the rate and efficiency of characters recognition.