Bilateral Grid Statistics Combined with BRISK for Robust Matching

Feature-based matching in image registration methods has always been a research hotspot. Matching using traditional robust features, while accurate, is often time consuming. How to improve the precision by simple descriptors for robust matching has become a new direction in feature matching research. Grid-based Motion Statistic is a fast and effective algorithm, but it lacks matching accuracy and does not perform well in large deformations. In this paper, we present a method combined bilateral grid statistics and BRISK algorithm on the basis of GMS. The original large number of messy matches are scored by the bilateral grid statistics, the correct and false matches are distinguished, and more accurate results are got. Compared with GMS, our method has higher correct rate, wider range of matching, better robustness than SIFT and SURF, and greatly reduces computation and time-consuming operation.