Multilayer reference frame motion estimation for video coding

Abstract

This paper proposes a new algorithm, multilayer motion estimation, based on the usage of a partial distortion measures with the objective of minimizing the overwhelming computational complexity of motion estimation. In order to make the computation of the partial distortion efficient, we build intermediate structures, which we have named as layers, from the reference frame using an overlapped hierarchical quad-tree structure. We have also proposed three types of search patterns to perform motion estimation on three layers. The derivation of these search patterns exploits the characteristic of the center-biased motion vector distribution and that of less-intensive block distortion measurement of the layers. Experimental results on a wide variety of video sequences show that the proposed algorithm outperforms the other state-of-the-art fast block matching algorithms while maintaining the motion prediction quality very close to the full-search algorithm.