Recursive handwriting recognition using epistemic utility theory

Abstract

In most handwriting recognition systems, the processing steps of segmentation, recognition, decision making and postprocessing are serially taken. These systems usually use the resources exhaustively in each stage of the serial engine. Every stage is tuned to maximize global performance. However, such a paradigm is prone to miss the global optimal point because the individual stages often do not have all the information available. A more interactive model is desirable. In this paper we describe such a model extending the concepts of satisficing decision mechanisms. Our experiments show proof of validity of the proposed model.