A Motion Compensation Method Using Adaptive Interpolation Filters Based on Variable Shape Masks

Abstract

Motion-compensated prediction based on precise motion vectors, typically with half-pel or quarter-pel accuracy, is commonly used in current video coding schemes. This paper proposes a motion-compensated prediction method which can compensate precise motions using two-dimensional adaptive interpolation filters. In this method, plural number of non-separable two-dimensional interpolation filters are designed for each frame, and an appropriate combination of the filter and a motion vector with integer-pel accuracy is determined in each macroblock composed 16 x 16 pels. Moreover, a shape of the filter mask is adaptively changed to reduce prediction errors with smaller amount of side-information required for filter coefficients. Simulation results indicate that the proposed method provides better coding performance than the conventional methods using motion vectors with fractional-pel accuracy.