基于核聚类的手写金融汉字识别方法

根据手写体金融汉字的特点，利用核聚类方法将原始样本特征映射到高维特征进行聚类分组，对每一组使用一个支持向量机二值分类器进行分类，并用这些二值分类器组成决策树的结点，构成一个决策分类树。给出了金融汉字的分组方法和决策树的生成算法，提出利用交叠系数来控制交叠，可以克服错分积累，提高分类准确率。实验结果表明，采用该方法，手写体金融汉字识别的速度和正确率都达到了实用的要求。

Handwritten Financial Chinese Characters Recognition Approach Based on Kernel Clustering Algorithm

According to the characteristics of handwritten financial Chinese characters, original features are mapped to higher dimension by applying kernel clustering. Each group is classified by a support vector machines (SVM) classifier. These binary classifiers are seen as the nodes of decision tree, and construct a decision classifying tree. The algorithm of grouping financial Chinese characters and creating decision tree is given. The method of controlling overlap by overlap coefficient is proposed and it can overcome misclassification accumulation. Experimental results show that, with this approach, the speed rate and accuracy rate of recognition meet the requirements.