中文摘要：

由于MPEG-4AVC/H.264采用了可变块尺寸（variable block size）和率失真最优化（rate distortion optimization）两项技术,视频编码流程中复杂度最高的运动估计（motion estimation）模块变得更加复杂.另一显著的变化就是,所获取的运动信息与目标码率紧密相关.这给传统的快速转码技术带来了很大困难.该文首先提出了一种分层模型,按由粗到细的方式描述每个宏块（macroblock）的运动信息.基于这个分层模型,文中提出了一种通过预编码获取运动信息的算法,从而生成视频序列运动特性的完整描述.有了这种运动信息描述,在编码或是转码的过程中,编码器可以跳过运动估计过程,编码复杂度因此大大减少.为进一步加快编码速度,还提出了从运动信息描述中抽取最优运动信息的快速算法.实验结果验证了所提算法的有效性,在大大降低编码复杂度的同时,编码性能与最优的MPEG-4AVC/H.264非常接近.

英文摘要：

Since variable block size and rate-distortion optimization techniques have been adopted in MPEG-4 AVC/H. 264, motion data is computed with high intensity and the obtained data is strongly related to the target bitrate. This paper proposes a hierarchical model to describe the motion of each macroblock （MB） from coarsely to finely. Meanwhile, a pre-encoding algorithm is developed to calculate these data in advance. They can be stored as user data in a media file or a stream. The overhead of coding these motion data is very small for raw sequence or a high bitrate stream. As a result, it does not need motion estimation anymore in encoding or transcoding process. This paper also proposes a method to readily extract a specific motion from the model for efficiently coding at either a high bitrate or a low bitrate. Experimental results have shown that extracted motion can achieve similar performance as the originally optimal MPEG-4 AVC/ H. 264.