Application of Questionnaire Theory to Pattern Recognition

Abstract

Analogies between Pattern Recognition and Statistical Communication Theory are pointed out. Wit:. the relative entropy between the "ideal attribute" and "heuristic attribute" sets as an index o!' performance, optimum selection of possible heuristic operators results in a recognition decision in probabilitty, with the smallest possible average number of tests during a "sequential analysis". On the other hand, a computer generation of q_:;.asi -optimal operators may be done based on n ' ..:r~e rical attributes using the same entropy index of performance. Experimental res:;.! t s are presented : Optimal operator choice is demonstrated for recognition of handwritten capital letter words and for vocoder speech recognition. Cccputer building of quasi-optimal operators is stcwn fer digitized images. They are limited to first level linear pumerical operators.