MPEG-4 synthetic video object prediction using embedded memory

Synthetic video object tools in MPEG-4 use a mesh topology representation that expresses the dynamic of the video object using motion of control points (i.e. nodes). It uses affine transformation to predict the video object from the deformation of the mesh topology. This paper presents SIMD (single instruction stream multiple data stream) architecture that prototypes an embedded memory for video motion compensation using affine transformation as defined by MPEG-4. The affine transformation computations are distributed on the parallel processing elements to improve the performance and reduce the power consumption. In addition, the embedded memory is optimized to be suitable with the parallel implementation. The proposed architecture is prototyped, simulated and synthesized for 0.18 μm CMOS technology using TSMC standard cells.