A block matching algorithm for deriving quality-tunable motion estimation architecture
Bharat Garg, G. K. Sharma · 2016
Energy efficient designs are prime requirements for portable multimedia devices where compression is done for efficient data transmission and store. The video compression requires huge computations in which maximum power is consumed within motion estimator (ME). This paper presents a new VLSI architecture of ME that provides the tradeoff between quality and energy. We propose a novel triangular search block matching (TSBM) algorithm that reduces the significant amount of computations within motion estimator. The proposed TSBM algorithm computes only a few initial search points and conserves power significantly by exploiting the high correlation between consecutive frames. The sub-sampling and threshold techniques are also employed to further reduce the complexity of each search point. The proposed architecture provides improved energy efficiency with acceptable output quality. Simulation results show that proposed algorithm reduces computation complexity by 60.9% and 39.9% over the Three-step and Diamond search algorithms respectively. The sub-sampling technique reduces the sum of absolute difference computation complexity by 50%, and threshold provides further 3.2% reduction with 0.02dB PSNR degradation. Further, the proposed ME architecture provides 77.8% energy saving over the exiting architectures. Keywords: Motion estimation, Block matching search, Video processing.