A fast motion search for warping prediction based on node classification

We propose a fast motion search algorithm for warping prediction (WP). Because of the spatial dependency between motion vectors of neighboring node points carrying motion information, optimization of motion search requires an iterative search. The computational load stemming from the iterative search is one of the major obstacles for practical usage of WP. In the proposed fast search a subset of nodes chosen in advance is only considered in iteration. By restricting the number of nodes on which iterative search is performed, the proposed method can speed up the overall motion search procedure significantly. For a proper choice of nodes, a simple and effective classification scheme has also been developed. Experimental results have shown that the proposed fast search can reduce the number of iterations considerably while maintaining performance as close as the full search.