Abstract

One task in image understanding is to develop a mapping from some lowlevel abstraction to a higher level abstraction. One approach taken is to map from regions, lines, and/or surfaces to semantic labels. Achieving this task commonly involves developing a world model which defines semantic labels in terms of their lower level primitives. A matching scheme is then employed to calculate a match score between low level primitives and semantic labels, and some inference technique used to compare, contrast and combine match scores to create a consistent interpretation.

© 1987 Optical Society of America