【作者】 王安红；

【导师】 赵耀；

【作者基本信息】 北京交通大学 ， 信号与信息处理， 2009， 博士

【摘要】 为了使低能耗视频设备可以在复杂的网络环境下,可靠地传输图像和视频,人们在设计图像视频编码器的时候,不仅需要考虑压缩性能,同时也需要兼顾视频编码算法编码端的复杂度和传输的可靠性。作为一种具有低复杂度编码和一定鲁棒性的编码框架,分布式视频编码（Distributed Video Coding,DVC）,也称Wyner-Ziv（WZ）视频编码,近年来已成为图像/视频编码研究的热点课题。本论文从五个方面深入研究了DVC的相关理沦和关键技术,主要内容和创新点如下:1.在基于LVQ（Lattice Vector Quantization）的DVC方案研究方面,首先提出了一种基于A₂格的像素域LVQ-WZ视频编码方案:对每个输入矢量运用一细一粗两种A₂格型矢量量化器量化为两个格点,将这两个格点的差值矢量用信道码编码,传送校验位到解码端;解码端首先运用信道解码和边信息恢复差值矢量和粗格点,然后用边信息对细格点进行量化重构。该方案编码简单而且具有较标量量化方案更好的压缩性能。在此基础上,本文又提出了一种小波域LVQ-WZ编码方案:小波变换进一步去除了像素间的相关性,因而小波域LVQ-WZ具有比像素域LVQ-WZ更好的率失真性能,同时,该方案对小波系数进行优化组合量化,并提出一种解码端运动补偿方法以使边信息逐步精细。2.在高效DVC方案研究方面,提出了两种方案:基于LQR Hash（Low QualityReference Hash）残差的DVC方案和基于帧内模式判断的混合DVC方案。在第一种方案中,LQR Hash利用了帧间相关性,因而比现有Hash具有更好的率失真性能。同时,利用小波域DVC进一步压缩LQR Hash和WZ帧之间的残差,有效提高了压缩性能;在第二种方案中,通过帧内模式判断来区分同一帧内不同空间位置的时间相关性,对时间相关性差的小波块用帧内编码,对相关性强的用WZ编码,实验结果表明,这种混合DVC取得了较好的压缩性能,同时编码复杂度基本不变。3.在可伸缩DVC方案研究方面,提出了质量可伸缩的DVC方案,方案利用SPIHT的嵌入式编码,将小波系数按照树结构重新组织为小波块,借助反馈信道,每个小波块的SPIHT树信息在解码端搜索更精确的边信息,提高整个系统性能。4.在鲁棒的DVC方案研究方面,提出了两种基于多描述编码的鲁棒DVC方案。方案一利用DCT域插值法将视频分开成两个描述,分别利用DVC压缩并传输到不同信道。当传输中出现丢包现象时,可用收到的描述来恢复丢失的描述,从而保证了丢包率较高时的鲁棒性;方案二采用奇偶帧分离法将视频分为两个子序列来形成两个描述。其中每个描述由两部分组成,第一部分对应子序列的零运动矢量H.264码流,第二部分是另一个子序列的WZ码流。作为一种冗余信息,WZ码流保证了丢失的描述可以被恢复。在此基础上,又提出一种残差两描述DVC来进一步减少冗余,并对冗余的数量进行了优化处理。5.构建了一个基于DVC的手机视频通信系统,实现了像素域DVC算法。更多还原

【Abstract】 With the emergence of the low-power video devices in network communication, some new challenges have been posed to the conventional communication system. These create the need for coding approaches combining low-complexity encoding, robustness and scalability. Distributed video coding (DVC, also called WZ video coding) has emerged as an attractive framework for low-complexity encoding and robust transmission over unreliable channels. This thesis aims to design the schemes of efficient DVC to satisfy network communication. The research focuses on five aspects.1. Research on DVC based on lattice vector quantization. Firstly, LVQ-WZ coding in pixel domain is proposed. Two LVQs, a coarse one and a fine one, are implemented to each 2-dimension input vector, and then the difference between them is channel encoded. Only the parity bits are sent to the decoder to recover the coarse lattice point with the help of side information. Finally, the re-quantization of fine lattice points is also refined by the side information. LVQ-WZ can achieve high efficiency and low-complexity encoding. Furthermore, LVQ scheme is extended to the wavelet domain and a LVQ-WZ in wavelet domain is proposed. Discrete wavelet transform removes the spatial redundancy so this scheme has better rate-distortion performance than pixel domain WZ. The scheme presents an optimized combination quantization to wavelet coefficients. And a gradual refinenment to side information with motion compensation is proposed.2. Research on efficient DVC. Two kinds of efficient DVC, the residual DVC based on LQR and the hybrid DVC with intra-coding mode, are proposed. In the former, LQR Hash, the low quality version of WZ frame compressed by zero-motion based H.264 encoding, is sent to the decoder to help the motion estimation at decoder. LQR Hash is efficient due to the exploration of temporal correlation between frames. Since there is correlation between LQR Hash and WZ frame, the scheme also removes such correlation by using the residual WZ coding to get gains. The other scheme is the hybrid DVC based on intra-mode decision. In view of the characteristics of different wavelet blocks, two kinds of coding modes are determined by an intra-coding mode, including intra-SPIHT and Slepian-Wolf SPIHT. The intra-SPIHT is applied to the block in the case of low correlation estimated; otherwise the Slepian-Wolf SPIHT is implemented. Experimental results show the proposed DVC schemes achieve better performance than the referenced with low-complexity encoding.3. Research on scalable DVC. A scalable DVC scheme is proposed. The basic idea is to use Slepian-Wolf theory to compress SPIHT information. Furthermore, a higher efficiency scalable DVC using block SW-SPIHT is proposed, where side information is refined with the help of feedback channel for each wavelet block. The scheme is scalabe in PSNR and has low-complexity encoding.4. Research on robust DVC. Two kinds of robust DVC schemes based on multiple description coding are proposed. In the first scheme, the video is interpolated and divided into multiple descriptions, and then a DVC codec is applied to each description and transmitted by different channel. When the packet-loss even with high packet-loss rate occurs, the lost can be recovered by the received. In the second scheme, the odd/even frame-splitting partitions a video into two subsequences to produce two descriptions. Each description consists of two parts. Part 1 is a zero-motion based H.264 bitstream of the corresponding subsequence and part 2 is a WZ bitstream of the other subsequence. As a redundancy, WZ bitstream guarantees the lost subsequence is recovered. A residual 2D-DVC is proposed to reduce such redundancy. The optimization scheme is applied to introduce the appropriate amount of redundancy. The experimental results show that the proposed schemes achieve better performance than the referenced especially when the motion is low. Moreover, our schemes maintain low-complexity encoding. Since the amount of redundancy will affect the performance, the redundancy optimization schemes are also presented in these MDDVCs.5. A DVC-based mobile communication system is constructed and a pixel DVC is realized in this system.更多还原