Robust feature point matching based on geometric consistency and affine invariant spatial constraint

Feature point matching is essential in computer vision. In this paper, we propose a robust feature point matching framework in which we first obtain a set of refined matches from ranked initial-matches based on a restricted affine invariant spatial constraint, and then compute a global geometrical transformation from the refined matches. After that, we recall the missing correct matches meeting the geometric consistency and spatial constraint. Compared with existing methods, the proposed framework can yield much more correct correspondences, which will be definitely helpful to further tasks. Experimental results demonstrate the advantage of the proposed method.