【作者】 范晨；

【导师】 唐昆； 崔慧娟；

【作者基本信息】 清华大学 ， 通信与信息系统， 2003， 博士

【摘要】 由于视频业务误码敏感度高并且有实时性要求，这与无线信道相对较窄的带宽和较高误码率形成了尖锐的矛盾。因此，要在无线信道上开展视频业务，必须在保证视频编码压缩效率的同时，提高视频编码的容错性能。由于该问题极具挑战性，因此成为近年来通信领域内的一个研究热点。针对主流的DCT变换结合运动补偿的视频编码框架，论文从提高其容错性能入手，同时兼顾压缩效率和运算复杂度的要求，重点进行了以下几个方面的研究。为了抑制视频编码在空间方向的误码扩散，提出了针对变长熵编码的信源信道联合卷积编码算法。该算法借助系统卷积码的结构，在不增加编码开销的前提下，实现了对变长熵编码码流的逐比特的非等重保护，从而有效抑制了误码扩散。他人这方面的工作主要集中在通过增加冗余提高变长码本身的容错性能，和提高变长码译码算法的鲁棒性上。因此，该算法为解决变长码误码扩散问题提供了全新的思路，具有重要的理论意义。同时，为了克服视频编码在时间方向的误差扩散，提出了基于棋盘插值的多描述视频编码框架。在此框架内时间方向的误差扩散可以通过两个步骤得到有效的控制。第一步是通过增加冗余来降低可能出现的编解码器不匹配的程度，第二步是通过反馈来消除残余误差的积累效应。实验结果证实，在误比特率为1×10－4时，通过以上两个步骤，时间方向的误差扩散得到了非常有效的控制。另外，论文对图像遭遇误码之后的修复算法进行了重点研究，用于提高出错后重建图像的主观视觉效果。对于分块DCT编码，提出了一个基于掩模插值的图像块修复算法。考虑到人眼对于物体边缘信息比较敏感，该算法利用相邻图像块之间的相关性，对受损图像块的边缘位置与走向进行估计，然后选择合适的方向进行插值修复。与运算复杂度相当的同类算法相比，该算法的修复图像质量无论在客观评价准则和主观评价准则上都有一定的提高。针对小波变换编码，论文提出了自适应DWT技术，它能够在一定程度上抑制高频子带丢失之后重建图像中物体边缘附近的波纹效应。最后，论文还研究了针对恶劣无线信道的链路层技术，包括自适应纠错与帧同步，并提出了基于最大后验概率的帧同步判决算法，大大提高了存在滑码时<WP=6>同步判决的可靠度，从而显著改善了链路层的吞吐率。这部分内容与前面的视频容错编码算法共同构成了一个完整的高鲁棒性的无线视频传输解决方案。论文的部分成果已经应用于军工项目“多功能终端”之中，首次在军用无线综合业务数字网上实现了语音、视频和数据的多业务综合通信，能够在高误码信道上提供稳健可靠的通信质量。该设备已通过正样鉴定，开始投入生产。更多还原

【Abstract】 The compressed video data is extremely vulnerable to bit errors, and requires real time deliver, which conflicts to the low bandwidth and high error rate of wireless channels. Therefore, to realize real time transmission of video traffic over wireless channels, the error resilience of video coding must be improved, without significant compromises in video coding efficiency. This problem becomes a hot spot of research in communication area in recent years, since this task is quite challenging. For the state-of-the-art video coding algorithm using DCT and motion compensation, this thesis aims to enhance its error resilience, while preserves its coding efficiency and computation complexity. To restrain the spatial error propagation of video coding, a joint source/channel convolutional coding algorithm for variable-length code (VLC) is proposed. Using the structure of a systematic convolutional code, this algorithm achieves bit-by-bit unequal error protection for VLC-coded bitstream, and thus restrains the error propagation effectively. The work of other researchers on this topic mainly focuses on improving the error resilience of VLC through inserting redundancy, or enhancing the robustness of VLC decoding algorithm. Therefore, this algorithm provides a novel approach to solve the error propagation of VLC, which has theoretical significance. At the same time, to overcome the temporal error propagation of video coding, a chessboard-interpolation-based multiple description video coding algorithm is proposed. In this framework, the temporal error propagation could be eliminated in two steps. In the first step, redundancy is inserted to reduce the possible mismatch between the encoder and the decoder; consequently, in the second step, the residual error drift is eliminated through the encoder imitating the behaviors of the decoder. Simulation results demonstrate that the temporal error propagation is effectively controlled through the above two steps, when bit error rate is 1e-4.<WP=8>In addition, the error concealment algorithms for images corrupted by bit errors, are also discussed, to enhance the subjective visual effects when erroneous image is reconstructed. For block-based DCT coding, an error concealment algorithm based on masked shift interpolation is proposed. The algorithm estimates the positions and directions of edges in a corrupted image block, and then interpolates the image block along proper directions, because edges are more important to human visual effects. Compared with other error concealment algorithm with comparable computation complexity, this algorithm has a better recovered image quality, both in subjective and objective criterion. Moreover, for wavelet-based image coding, a novel technique named adaptive discrete wavelet transform is introduced, which is able to alleviate the ringing artifact near edges in the reconstructed image, in case some high frequency subbands being discarded. Finally, the techniques in the link layer for error-prone wireless channels, including adaptive error-correcting and frame synchronization, are discussed. A frame synchronization algorithm based on Maximum A Posteriori is presented, which significantly enhances the reliability of synchronization in case that slips exist in the communication link, and hence the throughput of the link layer is obviously improved. This work together with previous work concerning error resilient video coding, form a complete solution for robust video transmission over wireless channels. Parts of the achievements have been applied to the military project "Multi-function terminal", which realizes integrated real time communication of voice, video and data, over military wireless integrated digital network, for the first time. The device can provides high quality and reliable communication over error-prone channels. The device has passed the final appraisal and begun to be produced.更多还原