We describe a new decomposition algorithm for two-dimensional, polygonal shapes. The algorithm first finds a set of overlapping ribbon-like subshapes ("ribbons") by grouping skeleton fragments into long smooth spines. The parts are then obtained by cutting the ribbons along lines of mutual intersection. With this approach we find part cuts that obey heuristic rules of early vision [17] without using boundary curvature.
Index Terms:
Shape Decomposition, Delaunay triangulation, Saliency Network, Perceptual Grouping
Citation:
Ralf Juengling, Lakshman Prasad, "Parsing Silhouettes without Boundary Curvature," iciap, pp.665-670, 14th International Conference on Image Analysis and Processing (ICIAP 2007), 2007