COMBINATION OF NEURAL NETWORK CLASSIFIERS

A neural network classifier combination method is introduced in this paper and applied to handwritten numeral recognition. Different kinds of feature sets are extracted from the same sample set and different classifiers are obtained from these feature sets. Performance function PF（S,T） is introduced to determine the two thresholds S and T which are used to obtain the best balance between error rate and reject rate. Experiment results demonstrate that this combination method can adjust PF’s parameters according to different application’s requirement, reduce classifying error rate, and improve recognition reliability.