一种用于手写体汉字识别的候选字加权多分类器集成方法

A Candidate Based Weighting Multi-classifier Combination Scheme for Handwriting Chinese Character Recognition

提出了一种基于候选字加权的多分类器集成方法，并将其应用于手写体汉字的识别研究中。利用４种不同的特征提取方法构造了４个独立的分类器；利用不同分类器各候选字加权处理得到的置信度函数来构造集成函数，从而将４个独立的分类器集成为一个多分类器系统。通过实验分析了几种分类器集成的方法，验证了具有一定互补性的多分类器集成对手写体汉字的识别率有较大的提高。实验结果表明、所提出的方法是行之有效的。

In this paper, a multi-classifier combination scheme based on weighting individual classifier's candidates is proposed. Four individual classifiers arc constructed with four different feature extraction approaches. A confidence function is defined for the classifier integration through weighting the similarity of different individual classifier candidate outputs. Different integration methods arc studied. Application of the multi-classifier to handwriting Chinese character recognition demonstrates that the recognition rate of tile integrated system can be improved by 2% or so, showing the effectiveness of the proposed method.