Reduction of memory accesses in motion estimation by block-data reuse

A new algorithmic approach to reduce the number of frame memory accesses in video coding system is proposed. The approach is based on observation that a large portion of accesses to the frame memory is spent for re-writing multiple blocks of data whose content remains constant between the adjacent frames. Due to data-driven mechanism of block characterization and selective block processing, the method can reduce the amount of frame memory accesses by as much as 79%, while providing high picture quality. An implementation scheme is presented.