An Example of Learning in Knowledge-Directed Vision
Bruce A. Draper, Allen R. Hanson · Series in machine perception and artificial intelligence · 1992
Introduction The goal of image understanding systems is typically the identi#cation of objects in visual imagery and the establishment of the three-dimensional relationships among the objects and the viewer. It is a generally accepted premise that, in many domains, the timely and appropriate use of relevant knowledge can substantially reduce the combinatorially explosive search encountered in establishing 'instance-of' relationships between image data and its interpretation#s#. Because of the variety and scope of knowledge pertinent to vision, the acquisition of both object models and interpretation strategies remains a major outstanding problem in model-based image understanding. While many vision algorithms at the low and intermediate levels are available, successful use of knowledge in image understanding requires a careful hand-crafting of the knowledge base. Typically this requires specifying, for each object class, both the description of the generic object as well as o