Block-level refinement of motion description in layered H.261 video

A layered refinement scheme for H.261 video is proposed, where the refinement layer consists of additional motion vector information rather than the traditional pixel/coefficient difference information. The motion vectors of the macroblocks within an H.261-coded frame are refined to give block motion vector information, thus providing a more accurate description of the motion present within the frame. We also propose a zonal block weighting scheme that targets refinement of perceptually important regions of each frame. In addition, we address the issue of the increase in computational complexity incurred by the additional block-level motion prediction by considering a motion prediction implementation using only one block-based motion predictor, i.e. block-based prediction only, with no explicit macroblock motion prediction, and present results illustrating the effects of such a simplification.