A Fast Recursive Algorithm for Gradient-Based Global Motion Estimation in Sparsely Sampled Field

This paper proposes a new approach for global motion estimation using recursive algorithm in the sparsely sampled field, as well as we process parametric estimation in framework of one stage not in proposed pyramid structure. Firstly, we divide the image into blocks and obtain the highest gradient magnitude in each block to form a sparsely sampled field. Then, we derive a new recursive gradient-based algorithm for global motion estimation in sparsely sampled field. The low pass filtering is for eliminating noise of original images before the estimation processes. Finally, we propose one stage framework for the parametric refinement without the proposed hierarchical configuration. The simulation results show the comparisons of performance between our method and others.