A reliability improvement of NN based OCR using rules and committee classifiers

The ability of Neural Networks (NN) to learn from training samples in order to generate desired decision regions, has been widely used in recent pattern recognition applications. In this paper, the cooperation of two feature families through a committee of NN based classifiers is investigated. The cooperation scheme is based on two stage classification using a combination of rule-based and statistical approaches. The corresponding results that show the significant improvement of the system reliability are also presented.