A Neural Network Based Multiple Expert System Model for Conflict Resolution

The paper describes a Neural Network and Multiple Expert System Model for Conflict Resolution of Unconstrained Handwritten Numerals. The basic recognizer is the Neural Network. The Neu-ral Network classifier is a combination of Modified Self-organizing Map (MSOM) and Learning Vector Quantization (LVQ). It will solve most of the cases, but will fail in certain confusing cases. The Multiple Expert System, the second recognizer, resolves the confusions generated by the Neural Network. This Expert System increases the confidence level of each decision made by the neural network recognition system of the first stage and corrects the possible substitution , thus resulting in a most reliable system. The results obtained from this architecture are compared with comments collected from an experiment conducted with a group of human experts specialized in unconstrained handwritten character recognition. The developed system is giving the same confusing pair as that given by the group of human experts and it also resolves the confusion.