Research On Pairwise Shape Matching Methods Based On Contour Points Information

In computer vision and pattern recognition, shape matching methods research has always been a major issue. Shape feature extraction and shape feature matching are essential steps of shape matching methods. These two steps play a vital role on precision and efficiency of the whole shape matching methods, especially shape feature extraction. Shape feature extraction’s result is shape representation. Based on shape feature extraction method, shape matching methods are divided into two categories. They are respectively region-based shape description techniques and contour-based shape description techniques. Among them, shape description techniques based on contour points are most popular.In this paper, we do the research on shape matching methods mainly via shape feature information analysis, including shape feature extraction and shape feature matching which are all part pairwise shape matching methods. In respect of shape feature extraction, in order to overcome classical shape context in limits of rotation and efficiency, a novel aspect shape context based on Bellman-ford called BFASC for short, is proposed.BFASC on framework of aspect shape context,uses Bellman-ford to calculate the shortest path among any two contour points. In respect of shape feature matching, to make the shape descriptor translation invariance, rotation invariance and scaling invariance, dynamic programming method is adopted to match shape feature in BFASC. The method that proved also is expected to improve matching accuracy and widen fitness that let it can achieve good results on the mixed shape sets of non-rigid objects and rigid objects. The results show that BFASC can shorten matching time and improve the matching efficiency in the feature extraction process, combined with dynamic programming, also can ensure the matching accuracy. Besides, the applied range of pairwise shape matching method proposed in this paper gets a better widened.