Efficient Motion Estimation Accelerator for H.264/AVC

Abstract

This paper presents efficient memory reuse and sub-pixel interpolation algorithms for ME (motion estimation)/MC (motion compensation) of H.264/AVC. ME is a computationally intensive task and its number of memory accesses is quite high. Moreover, sub-pixel interpolation requires many additions and multiplications. The proposed memory reuse algorithm utilizes the position similarity of the MVp (motion vector prediction) among neighboring Sub-MBs (macro blocks). It can reduce a large number of memory accesses and can save power consumption by sharing SR (search range) of the current block In addition, simplifying weights of sub-pixel interpolation can eliminate multiplications. Therefore, the proposed memory reuse and sub-pixel interpolation algorithms can be employed for low power video compression. In addition, the proposed ME accelerator uses the proposed algorithms. Performance comparisons show a significant improvement compared with existing ME accelerators.