Research On Some Key Issues In Shape Matching

Shape matching based on shape feature has become an important research direction of computer vision and pattern recognition, shape matching had been widely applied in a lot of fields. Such as image retrieval, character recognition,object recognition, medical image analysis, face recognition, robot navigation, s ensor network and other areas.Shape matching as a key problem in computer v ision and pa-ttern recognition,although had been researched for decades by man y scholars,th-ere are still many unresolved problems in shape matching,such as articulation and partial occlusion. In this paper we mainly aim on the descripti on of shape contour feature and the description of skeleton feature.The main works and achievements are as follows:1. Considering the contour of shape, this paper proposes a shape matching method which based on the inner distance shape context.First of all,we define a distance measure meathod which called inner-distance,we show that the innerdistance is insensitive to shape articulation and it is often more discriminative t-han the Euclidean distance for complex shape, then the Euclidean distance of shape context is directly replaced by the inner-distance.at last,we use dynamic programing based method for shape matching. Experimental results show that the proposed method can achieve shape matching with articulation.2. Considering the skeleton of shape, this paper proposes a shape matching method which based on critical nodes using path similarity. Firstly we define the critical nodes(junction nodes and end nodes),then take mergence of junctio n nodes to obtain skeleton structure graph with stability.Compute the skeleton path between junction nodes and end nodes; as for skeleton path, the skeleton feature invariant is made to represent the skeleton junction nodes. According to the descriptor of junction nodes is represented base on partial skeleton path in this thesis,it is local descriptor, so the proposed method can effectively recog-nize the object with articulation and partial occlusion. Experimental results sho-w that the method can achieve shape matching within articulation and partial occlusion.