Adaptive subsampling and motion feature based fast H.264 motion estimation

The latest video coding standard H.264/AVC can provide excellent performance compared with previous standards. However, the introduction of new techniques in motion estimation also brings about heavy burden of computation complexity on the video coding system. In hardware field, the subsampling technique is widely used to reduce hardware cost and release computation burden of ME engine. However, direct subsampling on video sequences will cause serious quality degradation. This paper proposes a hardware friendly fast ME algorithm which combines two complexity reduction schemes. Firstly, the pixel difference based adaptive subsampling scheme select different subsampling patterns according to the feature of current MB. Secondly, in order to fasten the ME procedure, a search position based 3-layer search range adjustment scheme is proposed to remove redundant search points. Experimental results show that about 74.17% to 90.26% ME Time can be reduced while the video quality loss is trivial. Moreover, since the two schemes are both friendly to hardware implementation, the proposed fast ME algorithm can be implanted into realtime hardwired video coding system.