Affine schemes in mesh-based video motion compensation

We evaluate performances of different implementations of affine transform in motion compensation architecture based on a hierarchical adaptive structured mesh. The architecture predicts the next video frame using the reference frame, mesh code and mesh node motion vectors. It achieves significant reduction in describing the mesh topology by coding the splitting in recursive triangulation of the initial coarse geometry. It uses a memory serialization unit and one simple warping unit to map the hierarchical structure. The affine unit warps the texture of a patch at any level of hierarchical mesh independently. We compare Shift-After-Difference (SAD), LookUp Table (LUT) and naive implementations of the affine unit. Computing affine transform using multiplication-free SAD algorithm significantly reduces the complexity of the architecture. We establish from simulation results that our multiplication-free SAD affine computation requires far lesser power and area than other schemes. We discuss the limitations and advantages of either motion compensation schemes. Performance analysis shows that this scheme is suitable for video applications like MPEG and VRML.