A Mismatch Elimination Method Based on Reverse Nearest Neighborhood and Influence Space

To tackle the problem of usually high feature mismatching rate for the scene containing similar structures and repeated textures,a mismatch elimination method based on the reverse nearest neighborhood and influence space is proposed under the framework of clustering-based analysis.More specifically,after obtaining a tentative noisy set of matching candidates,a new feature matching dataset is built by making use of the global information,which could better represent the local similarity,the geometric and the motion consistencies between matching pairs.Then a new salable radius clustering method is designed to determine the clusters by combining the reverse nearest neighbors and influence space,by which outliers are deleted from the clusters and mismatches are effectively discarded.With the standard dataset DTU and ancient Chinese architectural dataset,the experimental results show that,the average precision,recall rate and F score of our method can over 90%compared with other feature matching method.