An iterative overlapped motion compensation method based on segmentation of a reference image

摘要

This paper proposes a novel region-based motion compensation (MC) method for efficient video coding. The method segments the reference image instead of the current one and appropriately selects the regions to be motion-compensated one by one. Therefore we need not encode region shapes and can freely control a coding rate for the MC. In order to reduce unnatural artifacts due to intensity discontinuities in a predicted image, a kind of overlapped MC techniques is applied to boundaries of the selected regions. Simulation results indicate that the proposed method outperforms the conventional block-based MC and produces visually better predicted images.