Knowledge representation and control in computer vision systems

The authors analyze the roles of knowledge and control in working computer vision systems, describe model-based vision approaches whereby models serve to expedite scene interpretation by providing expectations for what is likely to be seen, and examine context-free approaches wherein image features are matched against a priori specified-object descriptions. They compare knowledge representation schemes of formal logic, semantic nets, production systems, and frames with respect to procedural and descriptive capability. They discuss control strategies, highlighting issues of parallel vs. sequential control, local vs. global control, distributed vs. centralized control, and top-down vs. bottom-up approaches. The authors develop these concepts within the framework of well-known systems such as Acronym, Hearsay, and VISIONS, providing a review of the major issues in computer vision.<>