CONTEXT KNOWLEDGE AND SEARCH CONTROL ISSUES IN OBJECT-ORIENTED PROLOG-BASED IMAGE UNDERSTANDING

At each level of an image understanding hierarchy, knowledge and image operators are represented to a different degree of ion. Operators may be selected according to current knowledge, and can further be tailored to the specific task by deriving appropriate values of their parameters. We describe a logic programming image understanding environment which provides predicate and parameter selection at higher levels based on context knowledge, and at lower levels via unification and backtracking, Solutions are given to context-based selection of parameters of image understanding, to object-oriented declarative Prolog-based image understanding, and to the control structure for object classification. Issues arising from constraint propagation, attention mechanisms, and control of the search are discussed.