Coding of independent pre-segmented arbitrary-shaped video objects, of the type envisaged for MPEG-4, conventionally require separate descriptions for shape and content. We describe a novel coding technique which combines shape and motion segmentation information. The motion estimator provides irregular-shaped areas containing uniform motion, and minimises the number of motion vectors required. The area descriptions are quad-tree related and these are integrated with the object shape to provide a unified structure. Temporal redundancy in the motion segmentation description is also minimised. Comparisons with MPEG-4 show a 15% improvement in coding cost, for the same quality prediction. |
|||