On the geometric interpretation of image contours

摘要

In this paper we suggest a computational model for the 3D interpretation of a 2D view based on contour classification and contour interpretation. We concentrate on those contours arising from discontinuities in surface orientation. We combine a generic surface description well suited for visual tasks with a model of the image formation process in order to derive image contour configurations that are likely to be interpreted in terms of surface contours. Next we describe a computer algorithm which attempts to interpret image contours on the following grounds. First, an image analysis process produces a description in terms of contours and relationships between them. Second, among these contours, we select those which form a desired configuration. Third, the selected contours are combined with constraints available with the image formation process in order to be interpreted in terms of discontinuities in surface orientation. As a consequence, there is a dramatic reduction in the number of possible orientations of the associated scene surfaces.