A Recursive Algorithm for the Complete Representation of Salient Contour Paths in a Graph

The problem of computing salient contour groupings in an image is formulated as a graph search problem where the vertices of the graph are given by high-contrast smooth contour fragments and the arcs of the graph represent hypothesized contour singularities (corners and junctions) or hypothesized \invisible" contour fragments. Whereas previous approaches to this problem assume an additive cost function and compute a unique path between two vertices, the proposed algorithm is not subject to these restrictions and is proven to represent, with high probability, all the salient image contours accurately and eeciently. This algorithm is viewed by the authors as an instance of a more general class of algorithms for hierarchical estimation which integrate knowledge and information at several levels by means of continuous and frequent interleaving of bottom-up hypothesis generation with top-down model-based veriication.