Image Matching with Multi-order Features

We propose a novel image matching method that may incorporate first-, second- and third-order features. These features are defined by a feature point, an edge linking two feature points and a triangle connecting three feature points, respectively. Taking them as vertices, the matching method constructs a weighted bipartite graph for computing the maximum weight matching solution by the Kuhn-Munkres algorithm. In second- and third-order cases, we design a Hungarian decoder to get the final matching between feature points. Experimental results show that the method can achieve relatively good performances on video sequences, even similar or superior to some of the state-of-the-art.