The recovery of straight picture edges from digitizations of scenes containing polyhedra ("line finding") is central to the functioning of scene analysis programs. While recognizing that recovery properly involves a computational mobilization of a great deal of knowledge-supported context, there remain some basic issues of representation which govern the way in which the primary data—grey levels—are addressed. The paper describes a parametric representation of straight picture edges and its procedural deployment in the recovery of edges from digitizations of scenes whose contents are essentially polyhedra with strong visible shadows.

This paper describes a new method for identifying a character, which overcomes certain fundamental problems arising with existing techniques. The method involves the evaluation of an autocorrelation function of the character to be recognized. This function contains maxima corresponding to straight lines in the character, and is essentially unchanged by changes in the position and orientation of the character. Variations in the size and aspect ratio of the character also leave the function sensibly constant.