Hierarchical Block-motion Estmation Using Linear Model-based Postprocessing

Abstract

A two-hierarchy block-motion estimation method is proposed which uses a postprocessor based on a linear model of the block-motion vector field. Postprocessing in each hierarchy uses the linear model to correct false motion vectors estimated by blockmatching. By depending on a weighting parameter and window size for the correction, the global or local motion is effectively generated at each hierarchy.