Algebraic inference of pattern similarity
Julian R. Ullmann · The Computer Journal · 1967
It is common in automatic pattern recognition to recognize an unknown pattern as the same as the known pattern to which it is most similar. This paper explores the idea that if appropriate pairs of parts are chosen, then the similarity of a pair of patterns is the sum of the similarities of these pairs of parts. In a simplified recognition problem, it is found that appropriate pairs of parts are chosen automatically when a purely algebraic technique is used for inferring the similarities of unknown to known patterns. This is a critical step in a research programme aimed at automatically finding suitable systems of features for recognizing hand-printed and cursively written characters.