Topological clustering and its application for discarding wide-baseline mismatches
Yongtao Wang · Optical Engineering · 2008
We present a novel scheme for discarding wide-baseline mismatches. Based on a general two-frame wide-baseline matching model, the proposed algorithm first generates match clusters that are topologically invariable between frames, and then discards mismatches from clusters. Experimental results demonstrate that our algorithm can effectively extract high-precision scale-invariant feature transform (SIFT) matches from low-precision initial SIFT matches for wide-baseline image pairs. Furthermore, the algorithm always performs best or close to best in the comparison, indicating that it is more robust than other methods for discarding wide-baseline mismatches.