Low-Complexity Global Motion for AVC and HEVC Coders

Abstract

This paper presents a low-complexity approach to deriving Region Of Interest vectors for an image based on fast sorting and histogram binning indexed by the components of the vector. Using the ROI vector list as guidance, a motion estimation algorithm can go through the motion vector field for the frame and align vectors with high global cost to the ROI vectors. This technique reduces global cost by incorporating potential unrealized benefit of non-causal vectors. Incorporating global benefit allows higher ? values to be used when Lagrangian approaches are used to determine motion vector cost, and therefore images can be coded with lower motion vector costs while still maintaining motion alignment on object boundaries.