An integrated pattern recognition system and its application

We have designed an integrated pattern recognition system. Instead of designing a classifier for pattern recognition, a finite number of classifiers are simultaneously applied, and a multilayer artificial neural network with feedback is employed to process all the outputs of the individuals in order to obtain a more accurate classification rate. Because of the introduction of the feedback loop, the pattern recognition system becomes a nonlinear dynamic system rather than a nonlinear mapping. We obtain a sufficient condition on the absolute stability for the integrated network and derive a corresponding learning algorithm to ensure its stability. The system has been applied to totally unconstrained handwritten numeral recognition, and its performance is excellent!