Robust feature matching for geospatial images via an affine‐invariant coordinate system
Jiayuan Li, Qingwu Hu, Mingyao Ai · The Photogrammetric Record · 2017
Abstract Feature matching is a crucial stage for many photogrammetric and remote sensing applications. This paper proposes a novel and robust feature‐matching method based on a normalised barycentric coordinate system (NBCS), which is superior to a Cartesian system for this task. A scale‐invariant feature transform (SIFT) is performed to provide initial matches containing both correct matches (inliers) and false matches (outliers), with a focus on model estimation from contaminated observations (matches with outliers). An affine‐invariant coordinate system calledNBCSis defined based on ratios of areas. The two feature points of a correct match have the same coordinates underNBCSwhile false correspondences do not. This principle is adapted into a hypothesise‐and‐verify framework. The proposed method is robust, efficient and effective. Extensive experiments on real geospatial image pairs show that it significantly outperforms six other state‐of‐the‐art approaches. The source code and datasets used in this paper have been made public.