支持向量机在字符分类识别中的应用

Application of support vector machines in classification and recognition of characters

为了对数字字符和字母字符进行有效识别,提出了一种利用二值字符图像投影的特征参数构造字符特征矢量的方法,对这些特征矢量进行归一化处理并作为支持向量机的训练集.采用支持向量机和多层感知器网络对字符的特征矢量进行训练,分别构造出26个字母分类器、10个数字分类器以及36个字母-数字综合分类器.通过对字符的分类识别测试,字符识别的准确率平均为96.5%,识别速度平均为20.5ms/字符,结果表明了支持向量机在字符识别应用中的有效性.

To recognize numeral and letter characters efficiently, a novel method based on characteristic parameters of the projection of binary images was proposed to construct the eigenveetors. The eigenveetors were normalized and selected as training set of support vector machines. Through training the eigen vectors of characters using support vector machines and multilayer perception networks, 26 letter classifiers, 10 number classifiers and 36 letter-number integrated classifiers were constructed respectively. Testing results showed that the average veracity and velocity of characters recognition reached 96.5% and 20.5 ms/ character respectively, and that SVM is a promising method for characters recognition.