Motion-compensated 3-D subband coding with multiresolution representation of motion parameters

This paper concentrates on the aspects of motion estimation and representation in the framework of motion-compensated 3-D subband coding. The motion vector field (MVF) is estimated hierarchically, based on a description obtained from a decimated field of support points. Displacements in between these points are interpolated. The motion parameters are encoded by use of a 3-D Laplacian pyramid structure, which can exploit both spatial and temporal redundancies in the motion vector field. The scheme can cope with non-translational and spatially/temporally variant motion, and supports a multiresolution description of the MVF. The representation of both information components-image and motion-has a hierarchical structure allowing efficient error protection as well as compatibility towards finer resolution and higher quality. More than that, with the improved motion estimation and representation procedures, noticeable enhancements were obtained in the encoding of full-motion video at low rates