On the effect of approximate-computing in motion estimation

Motion estimation is the key engine of modern video coding systems. Despite being an effective technique to exploit temporal redundancy, it features high computational complexity, being also a power-hungry block. Stemming from an optimized architecture for the computation of the Sum of Absolute Differences (SADs) in the context of the High Efficiency Video Coding (HEVC) standard, this work analyses the effects and the advantages offered by approximated adders. Experimental results show that the adoption of approximate adders can effectively save power while introducing a small error, in particular when dealing with larger Prediction Units.