Overlapping local image descriptor

Abstract

A novel local image feature description method, which originates from the frameworks of classical SIFT (Scale Invariant Feature Transform) and SURF (Speeded up Robust Features), is presented in this paper. The design of our descriptor innovatively adopts sub-regions overlapped each other in feature description. Coordinated with overlapping sub-regions, we analyze different cell normalization methods. In addition, the optical adjustment is introduced to eliminate the impact of weather when it is applied in outdoor scenes. Since better robustness to capture salient information from a canonical image patch is what we try to seek out, experimental evaluations show that the method improves the performance of descriptor.