Feature Fusion for Registering Wide-baseline Urban Images

This paper proposes a novel automatic image registration method designed for urban scenes containing tall 3D objects under significant viewpoint changes. Our method is based on the fusion of two kinds of features: the interest regions and the edge groups, which we believe are more suitable than conventional local textures for such scenes. Many existing methods face matching difficulty either because of the lack of appropriate textures in urban scenes or due to the difficulty in modeling textures for non-planar objects under complex 3D viewpoint changes, especially for boundary regions. Throughout our registration process, the two “feature tracks” interactively support and constrain each other, producing a large number of initial point-to-point correspondences, which are later refined utilizing global context information. The proposed approach has been tested using urban aerial photos of diverse viewing conditions. Experimental results for challenging inputs have demonstrated its effectiveness.