Bidirectional weight graph transformation matching algorithm
Song Wang, Xin Guo, Xiaomin Mu, Lin Qi · 2014
A stable and accurate point matching algorithm named Bidirectional Weight Graph Transformation Matching (BWGTM) is proposed in this paper. The algorithm starts with a set of correspondences which contain a variable number of erroneous correspondences, or outliers, in addition to a fixed number of true correspondences (inliers). For each feature point and its K nearest neighbors (KNN), there are two set of graphs to be generated. Depending on the co-angular distances between edges that connect a feature point to its KNN in graphs, the vertices with maximum distance will be founded and deemed as candidate outliers. Considering that some inliers whose KNN consist of outliers are regarded as candidate outliers, a recovery strategy utilizes the addition of fresh vertices to regain these inliers. Experimental results demonstrate the superior performance of this algorithm under various conditions for images.