A neural network framework for low-level representation and processing in computer vision

A goal of computer vision is the construction of scene descriptions based on information extracted from one or more 20 images. A reconstruction strategy based an a three-level representational framework is proposed. The first representational level, the primal sketch, makes explicit physical characteristics of the scene through detection of illuminance changes and their geometrical distribution and organization. Physical characteristics appear at several spatial scales and a multiresolution analysis helps in eliminating spurious edges. The second representational level, the raw 2.50 sketch, makes explicit the orientation and rough depth at edge location of the visible surfaces. A multiresolution neural network stereo algorithm is designed to compute the disparity at each edge location and at all the resolution levels. Matching is facilitated by a hierarchical focusing mechanism. The third representation level, the full 2.50 sketch, makes explicit the orientation and depth estimate at all the visible surface coordinates. Depth information between the edges is computed with a local shape-from-shading algorithm. A constraint satisfaction network fuses stereo and shading data