文摘
Scaled Invariant Feature Transform (SIFT) is the state-of-the-art local image descriptor for its invariance to image translation, rotation, scaling, and change in illumination. However, its matching precision is not satisfactory in many situations. In this paper, we proposed a more precise matching algorithm—BKR-SIFT. We apply the Best-Bin-First (BBF) algorithm to achieve rough matching firstly. Then, the Kullback-Leibler (KL) divergence similarity score is used as the coarse pruning algorithm. Finally, we apply the Random Sample Consensus (RANSAC) algorithm to refine the matched features furtherly. Experimental results show that our proposed algorithm can reach a higher matching precision with approximately the same time compared to the SIFT.