Robust Stepwise Correspondence Refinement for Low-Altitude Remote Sensing Image Registration

Abstract

Low-altitude aerial photography using small unmanned aerial vehicles (UAVs) often exists image viewpoint changes as human, nature, and equipment factors. The various viewpoint changes bring a series of problems to the subsequent applications of low-altitude remote sensing images. In this letter, a robust neighborhood structure-invariant descriptor is proposed to solve the scaling, rotation, deformation, and their mixture problems existing in the images. To estimate a robust correspondence under severe mismatches, a dynamic strategy is designed. Using the designed descriptor, a classifier is trained to distinguish between true matches and mismatches. Our method compares with six state-of-the-art methods on 55 small UAV images and performs a feature-invariant capability in horizontal/vertical rotation, scaling, mixture, and extreme situations, and gives the best performance in feature matching and image registration.