多视点多描述视频编码的研究

Error resilient ability is very important to multi-view video coding. However, there is still a lack of the suitable solution for this issue at present. To address this problem, this proposal presents a robust multi-view video coding solution from the viewpoint of multiple description coding. First, a polyphase transform and mode duplication based multiple description coding scheme is proposed. More specifically, the horizontal and vertical down-samples are performed on each frame of the input multi-view video sequence, forming four sub-sequences, which are then paired into two descriptions in the quincunx way. Only one of the four sub-sequences is chosen to perform mode decision by a JMVC encoder. Other sub-sequences directly adopt its mode decision results and conduct a prediction coding based on its predictive vectors. Therefore, the computational complexity is greatly reduced. Moreover, each description only needs to transmit the mode information and prediction vectors of one sub-sequence, so that the required bit rate is effectively decreased. Second, a two-loop coding structure, including a side-reconstruction-distortion encoder and an above-mentioned multi-view multiple description encoder, is introduced to automatically control the redundancy according to the channel conditions, achieving a good trade-off between the coding efficiency and error resilient ability. Third, a super-resolution method is introduced to improve the multi-view video quality reconstructed from the side decoder. Finally, a frame difference projection based error concealment method is proposed to rebuild the lost packets and lost frames, improving transmission quality over none ideal multiple description channels. In summary, this proposal will conduct a high efficiency, low computational complexity, and high reliability multi-description coding solution for multi-view video transmitting over the Internet and mobile channels.

容错能力对多视点视频编码至关重要，但目前还没有成熟的解决方案。本项目从多描述编码的角度研究鲁棒的多视点视频编码，提出基于空间下采样与模式复制的多视点多描述视频编码方案。首先对多视点视频的每一帧进行水平垂直下采样，形成四个子序列，将其两两组合产生两个描述。编码时只对一个子序列进行模式选择，其他子序列复制它的模式，在其预测向量基础上进行预测编码，大大降低了计算复杂度。每个描述只需传一个子序列的模式和预测向量，降低了所需码率。在上述多描述编码器基础上加入边解码重建误差编码器，形成可根据信道特性自适应调整冗余度、实现编码效率和容错能力联合优化的双环多视点多描述视频编码结构。提出边解码视频图像的超分辨率重建算法，提高视频重建质量。提出帧差投影算法恢复丢失分组和丢失帧，提高非理想多描述信道下的视频重建质量。研究成果将为互联网和移动信道提供一个高效、低复杂度、高可靠性的多视点多描述视频编码新方案。

我们从多描述编码的角度研究鲁棒的多视点视频编码，提出一种基于空间下采样与模式复制的多视点多描述视频编码，为互联网和移动信道提供一个高效、低复杂度、高可靠性的多视点视频编码新方案。研究成果已经发表在国际学术会议2013 Data Compression Conference (DCC 2013)。在此基础上提出一种基于数据复用的多描述编码方案并应用于平面视频和多视点视频编码，研究成果发表在2013 International Conference (ICIP 2013),并申请了一个专利。为进一步提高编解码效率，我们对深度编码、立体视频分割以及感兴趣区（人脸）检测与跟踪技术进行研究，在SCI期刊《Journal of Visual Communication and Image Representation》和EI期刊《Journal of Applied Science and Engineering》各发表1篇论文，在《信号处理》期刊发表2篇论文。项目执行期间，共培养1名博士研究生，6名硕士研究生（其中2人已毕业），圆满地完成了规定的任务。