基于K-L变换与SVM的数学符号识别

Recognition of Mathematical Symbols Based on K-L Transform and SVM

为了实现印刷体数学表达式的自动识别,对印刷体数学符号的识别方法进行了研究,提出了一种基于K-L变换和支持向量机的符号识别方法.该方法首先对数学符号图像进行预处理,然后对其归一化之后的图像使用K-L变换进行特征降维,最后使用支持向量机分类器对其进行识别.对同济大学版的《高等数学》上册书中出现频率最高的112个基元或符号的识别率可以达到96%以上,实验结果表明,该方法比近年报道的方法有明显优势.

In order to realize the automatic recognition of the printed mathematics expression, in this paper, the recognition method of printed mathematical symbols is studied, and the method based on the K-L transform and support vector machine is proposed. In this method, the treatment object is the symbol image which is after the pretreatment, the K-L transform is used to lower the dimension of the features of the image which is normalized, and the support vector machine is used to recognize the symbols finally. The recognition rate of 112 symbols appearing in the first book of ＂advanced mathematics＂ published by the publishing company of Tongji University can be more than 96 %, the experimental result indicates that there are more obvious advantages in this method than in the methods reported in recent years.