Knowledge Driven Pattern Recognition Based on Modeling Human Perception

摘要

Research has been conducted in recent years to provide the human/robotic decision-maker with the kind of co- mprehensible perceptual imaging of the decision-making world, which the human "expert" may develop through his senses and perception([1,5-10,12,13). For 3D physical decision-making worlds, objects in ordinary pictures taken from unknown viewpoints, even when parts of the object are undetectable or hidden from view, may be recog- nized/classified by "formal description schema"-models [5-10] derived by modeling the human expert's perception of characteristic elastic constraints (percepts) and of subjective assessments of conformity/resemblance measures. Segmented curves and two-dimensional images may be important perceptual structures in themselves, a well as being essential ccmponents for subsequent detection of collinearity, parallelism, connectivity, and other perceptual groupings. Most current edge, or surface, detectors detect edge-points, or uniform-greyness regions, (image locations through which an edge, or a surface, is judged to pass), and possibly link these together into lists on the basis of proximity. In this paper we propose a technique for recognition of edges, surfaces, and their perceptual groupings, which is based on the interactive modeling of human perception of characteristic elastic constraints and of their spatial relations. We present an interactive human-computer environment and methodology, to help us develop a computational understa- nding of human visual perception, of perceptual organi- zation, and of knowledge-driven recognition. We propose a carputational approach to achieve visual recognition /classification of physical objects from two-dimensional images (i.e. using only 2D data), taken from unknown viewpoints, without preliminary reconstruction of depth information or surface orientation from the visual input. A particular segmentation method is proposed for the composition and the development of oomprehensible formal descriptions of patterns.