Salient Closed Contour Detection Based on Multiscale Analysis and Minimum-Angle

This paper presents a novel computational approach, named MAMA, to extract perceptually salient boundaries from a set of noisy boundary fragments detected in real images. The approach first acquire a contour map using multiscale analysis. Then constructs an relation matrix between arbitrary the endpoints of each edge. At last the most salient boundary is detected by searching the smallest angle that lines got from two points and the point's grad. The proposed approch guarantees the boundary is a global structure. We collect a varietry of images for testing the proposed approach with enouraging results.