Improved Segmentation of Salient Closed Contours

Kanglin Xu

M.S., Computer Science, University of New Mexico, 2002


The segmentation we describe is based on a distribution of contours traced by particle moving in directions undergoing Brownian motion. Using a saliency measure based on the global property of contour closure, we describe an improved method for the segmentation problem. We first integrate an optimization algorithm into the segmentation algorithm to find one of the three important parameters: the particle's speed. We also analyze the properties of the affinity matrix, whose elements are the edge-to-edge transition probabilities, from a mathematical point of view and give formal proofs. In addition, we correct some problems in the existing algorithm. Our experiments show that the particle's speed found by the optimization algorithm, is reasonable. (705K compressed postscript).