Caelli, T. and Bischof, W. F. (1993). Learning structural descriptions of patterns and objects. Proceedings of the First Asian Computer Vision Conference, Osaka, Japan, Nov 23-25, pp. 684-687.
Learning structural descriptions of patterns (or objects) involves the generation of rules which capture the properties of pattern parts and their relationships which can best classify patterns and enable generalisation to new unseen cases. In this paper we consider a new technique termed Conditional Rule Generation (CRG) which generates generalised descriptions of patterns as lists of part and part relational feature bounds which cover all class examples. The method "precompiles" descriptions of patterns as strings of unary and binary feature states which attain theses goals.
Back to publications.