Concave regularisation-based non-rigid feature point matching algorithm

A new non-rigid feature point matching algorithm is proposed, where the non-rigid global deformation is preserved by dividing the object into several overlapping local regions and only regularising the deformations of the local regions. A key property of the proposed algorithm is that the objective function is concave. By combining with the total unimodularity of the constraints, the proposed algorithms can be proved to directly yield integer valued solutions. Experimental results demonstrate the better robustness of the method over several state-of-the-art methods.