Keypoint descriptor fusion with Dempster-Shafer Theory

Abstract

Keypoint matching is the task of accurately nding the location of a scene point in two images. Many keypoint descriptors have been proposed in the literature aiming at providing robustness against scale, translation and rotation transformations, each having advantages and disadvantages. This paper proposes a novel approach to fuse the information from multiple keypoint descriptors using Dempster-Shafer Theory of evidence [1], which has proven particularly e cient in combining sources of information providing incomplete, imprecise, biased, and con ictive knowledge. The matching results of each descriptor are transformed into an evidence distribution on which a con dence factor is computed making use of its entropy. Then, the evidence distributions are fused using Dempster-Shafer Theory (DST), considering its con dence. As result of the fusion, a new evidence distribution that improves the result of the best descriptor is obtained. Our method has been tested with SIFT, SURF, ORB, BRISK and FREAK descriptors using all possible combinations of them. Results on the Oxford keypoint dataset [2] shows that the proposed approach obtains an improvement of up to 10% compared to the best one (FREAK).