Finding salient points of shape contour for object recognition

Object recognition via shape matching has been a fundamental topic in robot vision. It is intuitive that the performance of recognition will be improved if the salient feature points of shapes are found while the redundant points are filtered out. In this paper, we propose an adaptive contour evolution (ACE) algorithm to capture the salient feature points of shape contour. The redundant contour points are removed from the original shape, which makes the shape contour compact and representative. The degree of evolution in our method can be controlled adaptively for various applications. In this work, the shape context (SC) descriptor is adopted to represent shapes. The proposed method can be used to process shape contours for any existing shape descriptors. A framework of shape recognition based on ACE and SC is proposed, where the dynamic programming (DP) algorithm is employed for shape matching. The using of the evolved shape contour not only reduce the computing cost of shape matching, but also increase the robustness to local noise. The conducted experiments validate the capability of the proposed method. The comparable results on benchmark datasets indicate that the shape recognition accuracy is essentially improved.