Salient Contour Matching for Object Detection
Wei Bi, Yongping Zhang, Weiguo Huang, Guanqi Gao · 2016
In order to solve detection difficulties caused by the variable contours of the object and complex scene, the salient contour based method is proposed in this work. We directly extract image contours by globalized probability of boundary, and utilize the improved Otsu for adaptive threshold processing to attain the salient contours. Then, the refinement of fan shape model is used to match the salient contour for detecting object, its mirror flip and rotated object by the three similar matching strategies. The approach of the salient contour extraction can effectively remove the detailed edges and noise. Meanwhile, the problems caused by unstable object contour are resolved using the refined Fan Shape Model. The experimental result of ETHZ shape classes validated the effectiveness of the proposed method.