A novel geometry based matching method for images with multi objects

A novel geometry based matching approach is proposed for images with multi objects. In the proposed method, shape and topology ordered graphic is developed to model both shape and structural information for images. Vertexes are used to represent shape attributes, while edges are employed to define structural attributes between objects. The similarity between images is measured with two consecutive steps: vertex matching and edge matching. Vertex matching by comparing shape descriptors is exploited to generate probability distribution map of similarity between vertexes. 2-neighborhood structural matching is employed as a further selective criterion to find correspondences. Finally, Experiments are provided to validate the proposed method.