Adaptive block sub-sampling algorithm for motion-estimation on SIMD processors

Block-motion vector estimation is an integral part of commercial video compression standards. This paper describes an adaptive technique to select predefined sub-sampled block patterns to reduce the computation complexity of block-matching criterion (BM) in motion estimation (MH). The search-window texture complexity and 8×8 sub-block texture complexity are used to select the sub-sampling pattern for each sub-block. The proposed technique is suitable for both four motion-vectors (4MV) and one motion-vector (1MV) calculation of 16×16 macro-block (MB). The sub-sampling patterns are designed to completely exploit the 32-bit and 64-bit single instruction multiple data (SIMD) architecture available on media processors and general purpose processors (GPPs). The simulation results show that significant speed improvement in ME module can be gained with better PSNR compared to fixed block sub-sampling techniques.