【作者】 苗艳华；

【导师】 于明；

【作者基本信息】 河北工业大学 ， 通信与信息系统， 2007， 硕士

【摘要】 传统的视频编码标准均采用基于块的空间域运动估计技术来减少视频帧的时间冗余。由于小波变换在静态图像编码应用中的优良特性,基于小波变换的运动估计技术也成为了近来研究的热点。然而,离散小波变换具有移变性,这使得在小波域里难以获得精确的运动估计与补偿效果。冗余小波变换克服了小波变换的移变性,在冗余小波域进行运动估计可以达到很好的效果,但是算法的计算复杂度过高。本文主要研究基于冗余小波变换的运动估计算法。由于在冗余小波域进行运动估计计算复杂度高,因此,本文在分析视频序列的运动特性和已有的运动估计算法的基础上,利用冗余小波变换的特性,提出了一种基于冗余小波变换的快速多分辨率运动估计算法。此方法首先在冗余小波域提取潜在运动块并对图像块进行运动剧烈程度划分,在此基础上对不同类型的图像块采取不同的搜索策略,从而有效减少了运动估计过程中不必要的搜索。另外,本文将可变块多分辨率运动估计的思想引入冗余小波域,运动估计的过程首先在低频子带进行,之后在高频子带以较小的搜索范围细化运动矢量。此方法在估计精度和计算复杂度上取得了较好均衡,解决了现有的基于冗余小波变换的运动估计算法中存在的计算复杂度过高的问题。试验结果表明,本文方法较经典的基于冗余小波变换的运动估计算法在估计精度以及计算复杂度方面都具有优势。更多还原

【Abstract】 Block-based motion estimation in the spatial domain is widely employed in modern video compression systems. However, given the promising performance of wavelet-based still-image compression algorithms, now one hot spot in the research of video codec is the motion estimation algorithm based on the Discrete Wavelet Transform (DWT). However, the fact that the usual critically sampled DWT is shift variant greatly hinders the motion estimation process when deployed in the wavelet domain. Because the Redundant Discrete Wavelet Transform (RDWT) is shift invariant, the motion estimation algorithms in the redundant wavelet domain have good performance, but very high computational complexity.This paper mainly researches on the RDWT-based motion estimation algorithms. Because of the high computational complexity of such algorithms, so based on the analysis of the motion characters of video sequences and the existing motion estimation algorithms, utilizing the characteristic of RDWT, this paper proposes a fast multi-resolution motion estimation algorithm based on the RDWT. Firstly, this method by identifying the potential motion blocks and dividing the blocks into different degrees of motion in the RDWT domain takes different search schemes to blocks of different types, so the search area is decreased efficiently. In addition, this paper introduces the multi-resolution motion estimation into the redundant wavelet domain. Motion estimation first happens in the low-band, and then the motion vectors are refined in the high-band with a smaller search area. This motion estimation algorithm gets a tradeoff between the precision and the computational complexity, and overcomes the high computational complexity of the existing RDWT-based motion estimation algorithms.The experimental results prove that this approach has a superior performance compared with the traditional method in terms of the precision and the computational complexity.更多还原