An Abductive Approach to Hand-written Character Recognition for Multiple Domains

Perceptual problems often call upon the application of a variety of knowledge sources. An automated problem solver can utilize much of this knowledge in the form of top-down guidance to improve accuracy and reduce processing time. Ironically, most automated hand-written character recognition systems, which are highly accurate, seldom use any top-down guidance. This paper describes an approach to automated hand-written character recognition that applies domain-specific knowledge such that a partial solution can be refined through top-down guidance. Specifically, the recognition task is one in which features derived from the input data are explained through higher-level hypotheses using abduction. Top-down guidance is used to improve accuracy. This approach has been applied to several domains. Individual character recognition is 70% without top-down guidance but improves to 99% with guidance.