Computer Vision and Natural Constraints

Computer vision, the automatic construction of scene descriptions from image input data, has just entered its second decade. Approaches have varied widely, especially in the amounts of symbolic, domain-dependent knowledge and inference that are incorporated into the vision process. Much current research addresses the extraction of physical properties of the scene (depth, surface orientation, reflectance) from images by using only a few general assumptions about the scene domain. Extraction of physical parameters is part of a hierarchy of operations needed to transform image input data to symbolic descriptions. Two other processes that serve as examples are stereo fusion and the partitioning of image phenomena into related groups. Computer vision research is influencing theories of animal perception as well as the design of computing architectures for artificial intelligence.