PC networked inference for handwritten letter recognition

The authors describe an experience-based system which in principle needs not differentiate figures and letters at least at the early stage of recognition. To differentiate a letter from figures, some context may be useful, which provides knowledge that words are presented in that situation. And if no such information is available, a human applies some other sort of inference in judging whether the specific object is a letter or not. In this paper, a human-like way of letter recognition is discussed as part of the whole recognition system with high rate of correct recognition the authors are developing