Image Matching Using Mutual k-Nearest Neighbor Graph

摘要: Though weighted voting matching is one of most successful image methods, each candidate correspondence receives score from all other candidates, which can not apparently distinguish correct matches and incorrect using scores. In this paper, a new method based on mutual k-nearest neighbor (k-nn) graph proposed. Firstly, the k-nn constructed according to similarity between correspondences. Then, only its k nearest neighbors. Finally, scores, correspondences are computed by greedy ranking technique. Experimental results demonstrate effectiveness proposed method.