Complexity reduction method for overlapped block motion compensation based on spatio-temporal correlation

Owing to the dependency between the motion vectors of spatially neighboring blocks for overlapped block motion compensation (OBMC), the optimization of the motion vectors requires iterative search. The computational cost, stemming from optimizing the motion vectors, is one of the major obstacles for practical applications. In the proposed algorithm, the iterative motion search is applied only to the selected blocks to alleviate the computational complexity. To choose these blocks properly, we present an adaptive and robust classification method by exploiting the spatio-temporal correlation of neighboring blocks, i.e., variation of pel intensity continuity of motion field, and monotony of error surface. The proposed algorithm reduces the computational load by at least 55% and up to 80%, while retaining the PSNR performance comparable with the conventional OBMC algorithm.