Progressive mesh-based motion estimation using partial refinement

A technique for performing progressive mesh-based motion estimation in a layered fashion is presented. Motion compensation based on image warping provides a block prediction free of block artifacts. The smooth prediction can be used to identify motion-active regions by comparing with the reference frame and generate a partial denser mesh, thus forming layers of mesh. This approach provides a hierarchical partial refinement according to motion activity without additional cost. Experimental results indicate that the technique shows improvement over a single-layered uniform mesh and advantages over block-based techniques, particularly in scalable and very low bitrate video coding.