高效率视频编码中联合非纹理模型的率失真优化编码技术研究

Rate distortion opimization (RDO) plays a critical role in the hybrid video coding. It aims at minimizing the distortion under a constraint on the rate.In new generation of video coding standard (High Efficiency Video Coding, HEVC), many advanced coding tools provide more accurate predictions coding and thereby improve the compression efficiency. Meanwhile, signaling of the side information for non-texutre data becomes more significantly important. Therefore, the rate distortion analysis and optimization strategies are different from the traditional macroblock-based techniques. Firstly, a joint non-texture and texture rate distortion modeling approach is derived by integrating the texture and non-texture rate distortion models. The new joint non-texture and texture rate distortion models provide the optimization Lagrange multiplier, which can adaptively optimized from the input video content and quantization parameter, to improve the overall coding efficency. Secondly, in order to reduce the complexity of RDO coding, some fast optimization algorithms in the rate-distortion sense, including fast coding unit size decision, fast mode selection, and constrained optimization techniques, are proposed to speed up the coding process for practical applications. Finally, based on the derivation of joint rate disortion model and non-texture structure complexity, an accurate rate control scheme is proposed for H.265/HEVC. Research results of this project will be integrated into HEVC HM reference and X.265 platforms for analysis and comparison.

率失真最优化技术是混合视频编码框架中的重要组成部分，旨在码率受限的条件下达到最小的失真。高效率视频编码采用了很多先进的编码工具，大大的提高了预测编码精度并改善了压缩性能。由于新编码特性的引入，用于传输边信息的非纹理数据变得越来越重要，并造成了率失真最优化处理的差异，传统基于宏块的率失真优化方法不再适用。首先，本项目通过综合非纹理率模型和纹理率失真模型，提出了一种联合率失真建模方法。基于该方法，提出拉格朗日乘子的最优化选择策略，该策略可根据视频内容和量化参数自适应选择乘子，改善整个编码率失真性能。其次，本项目为保持率失真性能，降低率失真最优化编码的计算复杂度，提出快速编码单元选择、快速模式判决和简化率失真最优化函数等方法。最后，基于联合率失真模型和联合内容复杂度，提出一种精确H.265/HEVC码率控制方法。本项目研究成果将集成至HEVC参考平台HM和X.265中进行分析对比验证。

为满足高清/超高清视频编码的需求，新一代混合视频编码框架H.265/HEVC引入更多的编码参数，以获得更精确的预测和更高编码效率。预测得到的残差纹理信号适合使用率失真优化技术编码；而编码选择的指数增加导致了非纹理信息量的增加，对残差纹理率失真代价的影响不断增加。简单忽略非纹理信息的影响进行率失真优化选择处理已达到了率失真编码增益上限。本项目研究非纹理信息与纹理率失真之间的关系，针对非纹理信息的差异性问题，结合已编码信息，采用主成份分析方法，提出了一种描述非纹理信息的时间复杂度模型；通过统一尺度方式构造联合率失真模型，提出了一种基于时空视频内容复杂度的自适应拉格朗日参数选择率失真优化策略。基于以上联合率失真模型，在保持编码性能不降低的情况下提出了降低计算复杂度优化方法，实现了一套端到端的移动终端视频实时通信系统。针对无线视频传输中视频编码器计算资源和无线传输资源之间的匹配问题，提出一种面向无线视频数据负载增量的接入资源分配方法，并应用于智慧旅游、教育和水下激光通信等码率控制应用场景中。共发表论文10篇，获得2项授权发明专利，获得海南省科学技术进步奖三等奖。