A new multi-classifier combination scheme and its application in handwriting Chinese character recognition

Abstract

In this paper, a multi-classifier combination scheme based on weighting individual classifier’s candidates is proposed. Four individual classifiers are 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 are studied. Application of the multi-classifier to handwriting Chinese character recognition demonstrates that the recognition rate of the integrated system can be improved by 2% or so, showing the effectiveness of the proposed method.