Object recognition system with perceiving subjective contours

Abstract

Describes a 3-D object recognition system that uses as well as luminance edges (physical contours). In this paper, the authors suppose two kinds of subjective contours based on psychological and physiological findings: (1) Subjective contours that are automatically generated as by-products in the lower level processes of the visual system. Alignment and proximity of physical contours and line ends promote generation of such contours. The authors modeled the edge detection process involved in this kind of subjective contours by a BCS neural network (Grossberg and Carpenter, 1985) in their system. (2) Subjective contours that represent a hypothesis of segmentation with volumes. They are generated in the higher level process. The volumes have primitive shapes and they are components of object models. The authors used geons (Biederman, 1987) for describing object models. The higher level process groups both physical and subjective contours, that are detected in the lower level process, into the most probable geons, by found features such as line junctions and curvatures of lines. The authors describe the idea of a recognition system that combines lower and higher level processes. The authors applied it to partially shaded images and arrangement of lines.