An Improved Algorithm for Multiple Closed Contour Detection

We present an improved algorithm for the multiple closed contour detection in the multi-object scene, especially for the contours partly occluded by some others. Our method formulates the boundary fragments detected by edge detector from nature image as a graph represented by a reversal symmetry matrix, which incorporates the Gestalt principles of proximity and good continuity as the transition probability between vertices. The algorithm extracts the multiple closed contours by detecting salient cycles in the graph in a global way. It succeeds in extracting the multiple salient closed contours via several iterations even though the occlusion exists. The experiment results show that our algorithm can match or exceed the former ones when extracting isolate smooth closed contour, and do better when facing boundary occlusion cases.