A multiclass classification method based on multiple pairwise classifiers

Abstract

In this paper, a new method of composing a multi-classclassifier using pairwise classifiers is proposed. AResemblance Model is exploited to calculate aposteriori probability for combining pairwise classifiers.We proved the validity of this model by usingapproximation of a posteriori probability formula. Usingthis theory, we can obtain the optimal decision. Anexperimental result of handwritten numeral recognition ispresented, supporting the effectiveness of our method.