Improvement of error concealment technique for H.264 scalable video coding

Today’s scalable video coding technology provides a solution to multimedia content delivery over heterogeneous network to various kinds of terminals. In this thesis, the main focus is on the application of spatial scalability, which enables the delivery of a single multimedia content to users with devices supporting low video resolution to high resolution. However, the ubiquitous delivery network still uses a packet-based approach which cannot necessarily offer a guaranteed Quality of Service (QoS). Some losses may still be corrupt the video bitstream. For a highly compressed H.264 video, a single frame loss may have a dramatic effect on the visual quality of the affected frame as well as its succeeding frames. Error concealment techniques are applied at the decoder side in order to estimate the lost information. Some works on error concealment for the H.264 video coding standard have been presented before, but their performance is found to be inadequate at times. In this thesis, we propose an improved error concealment scheme. We propose a motion estimation method which takes a weighted average of the motion vectors in a neighborhood of macroblocks from the previously decoded frame. Furthermore, the reconstructed frame is enhanced by filtering it with a hybrid median filter. The proposed algorithm was found to have an average gain up to 0.8 dB over the state of the art BL_Skip which was also reflected the better quality in the subjective evaluation. Further investigation also shows that the proposed algorithm is able to suppress the effects of error propagation on the frames succeeding the lost frame.当前可伸缩视频编码（scalable video coding）技术可为异构网络环境下面向多类型终端的多媒体内容传输提供可行解决方案。本论文的核心研究内容为空间可伸缩性的应用，该技术可实现单份多媒体内容向配备不同分辨率终端的用户传输，适配从低分辨率到高分辨率的各类显示需求。然而，当前普及的传输网络仍采用基于分组的传输模式，无法确保服务质量（Quality of Service, QoS）得到稳定保障。数据包丢失仍可能对视频比特流造成损坏。针对高压缩比的H.264视频，单帧数据丢失会对受损帧及其后续帧的主观视觉质量产生显著负面影响。为预估并恢复丢失的视频数据，解码器端通常会采用错误隐藏技术。此前已有诸多针对H.264视频编码标准的错误隐藏技术研究，但相关方案的性能在部分场景下仍存在不足。为此，本论文提出一种改进型错误隐藏方案：该方案提出一种运动估计方法，对已解码参考帧中宏块（macroblock）邻域内的运动向量进行加权平均运算；此外，本方案还通过混合中值滤波器（hybrid median filter）对重建帧进行滤波增强处理。实验结果表明，相较于当前主流的BL_Skip方案，本算法的平均峰值信噪比增益最高可达0.8dB，主观评价结果同样印证了其更优的画质表现。进一步研究显示，本算法还可有效抑制丢失帧后续帧上的错误扩散效应。