【作者】 刘国金；

【导师】 杨小帆；

【作者基本信息】 重庆大学 ， 计算机软件与理论， 2003， 硕士

【摘要】 视频压缩及传输由于具有广泛的应用前景，成为国际研究的热点。它是HDTV、DVD、视频会议、Internet视频及其它相关应用的核心技术。目前，由于现有标准的局限性，国际上提出了嵌入式小波变换编码方案，与传统的DCT变换相比，小波变换具有时频域局部化特性和多分辨性质，且小波是对整个图像进行变换，避免了DCT变换的方块效应，更适合可伸缩的编码。随着用于图像压缩的EZW、SPIHT和EBCOT模型及用于视频压缩的3D-EZW、3D-SPIHT和3D-ESCOT模型的出现，基于小波变换的嵌入编码已成为图像视频压缩研究的主流。嵌入编码具有率伸缩特性，很适合在带宽随时延变化的互联网上使用。 目前，嵌入编码都是基于两种类型：一种是分层的嵌入编码，主要是利用小波不同子带间系数的偏序性和自相似性，将不同子带间的系数形成一棵棵树结构，通过将整棵树量化为零来减少压缩比。另一种是基于上下文模型，主要是利用小波系数同子带内相邻系数的相关性，组织高效的上下文，来显著地提高压缩效果。受EZW、SPIHT、EBCOT模型及3D-EZW、3D-SPIHT、3D-ESCOT模型的启发，本文综合利用小波系数不同子带间的自相似性、偏序性和同子带内相邻系数的相关性，提出了一种新的三维小波嵌入编码上下文模型3D-PEM。实验结果表明：利用这种模型获得的PSNR高于3D-ESCOT模型。同时也具有3D-ESCOT模型的优良性质。在此基础上，本文又提出了基于自适应运动补偿的三维小波模型MC-3DPEM，利用运动估计和补偿消除视频间的运动，对补偿后的亮度、色度分量先进行空间域的小波变换，再利用亮度系数、色度系数之间的相关性，将三个分量的系数形成一个变换系数矩阵，对变换系数矩阵进行时间域的小波变换，用先前的模型对变换后的三维小波系数编码。实验结果表明：相同码率情况下，基于运动补偿的MC-3DPEM模型获得的PSNR高于3D-PEM模型，同时视频三个分量的混合编码比每个分量单独编码时具有更好的PSNR。更多还原

【Abstract】 Video compression has long been one of the main topics of compression coding. Since it is critical to HDTV, DVD, Video conference, etc. In order to avoid the faults of international standards, a number of embedded wavelet coefficients coding models have been proposed. Due to its multiresolution, time/frequency localization and lower time complexity, wavelet is suitable for scalable video coding, which has been applied to coding algorithms such as EZW, SPIHT and EBCOT. Embedded coding schemes are SNR scalable, and, even farther, which are essential in transmission of image and video on Internet.There have been two type of embedded coding. The first is based on partial order and self-similarity between subbands of wavelet coefficients, which organizes the wavelet coefficients across subbands into a zerotree, while the second is based on context models, using the correlation of neighboring wavelet coefficients. Inspired by typical coding schemes such as EZW, SPIHT, EBCOT, 3D-IEZW, 3D-SPIHT and 3D-ESCOT, a new context model of wavelet coefficients is proposed, which make use of partial order, self-similarity between subbands of wavelet coefficients and the correlation of neighboring wavelet coefficients. Experimental results show that the model can achieve higher PSNR than 3D-ESCOT .while at the same time keeping all the advantages of 3D-ESCOT. In order to further improve the compression performance, the model is modified with the help of motion compensation technique. Experimental results also show that at various conditions, MC-3DPEM is superior to 3D-PEM without motion compensation更多还原