面向无线视频通信的视频软编码技术研究

The visual communication systems nowadays are mostly built upon the Shannon’s separation theorem that source coding and channel coding can be designed separately without sacrificing the optimality in performance. With the prevalence of wireless networks and mobile devices in recent years, the weakness of the traditional coding and transmission framework becomes evident due to the “cliff effect” of digital coding systems. In this project, we aim to investigate a new and efficient soft video coding framework for the video broadcast application in wireless and mobile scenarios. Firstly, we investigate spatial-temporal decorrelation transform and transform-domain signal energy modeling in the soft video coding framework, focusing on analyzing the performance of different transforms and evaluating their individual performance gain. Secondly, we study the mechanism to conduct intra-prediction and inter-prediction in the soft video coding framework, and investigate hybrid digital-analog coding, embedded multi-layer quantization, and the SK dimension-reduction mapping in order to achieve soft video coding with higher efficiency. Finally, based on the characteristics of human vision system, we study the way to represent and code the perceptually sensitive visual information in video, and design a perception-friendly soft video coding scheme. Based on all these techniques, we construct a robust and efficient solution for video communication in wireless and mobile environment.

现有的视频通信系统大多以香农的分离编码理论为基础，在快速动态变化的无线环境下、以及在差异化的多用户条件下，传统编码传输框架由于“悬崖效应”而面临着很大挑战。针对移动无线环境下的视频广播等典型应用，本项目围绕视频的高效软编码传输开展研究。具体包括：首先，研究视频软编码中的空域-时域去相关变换、变换域信号建模和能量分配，分析不同信号变换的性能及其增益，构建无线视频的优化软编码基本框架；其次，研究视频软编码中的高效帧内/帧间预测、数字模拟混合编码、嵌入式分层量化等核心问题，提升软编码的性能；第三，针对人眼视觉系统的特性，研究视觉敏感信息的高效表示和编码方式，实现感知优化的视频软编码。最终形成一种无线环境下鲁棒而高效的视频软编码传输方案。

现有的视频通信系统大多以香农的分离信源信道编码理论为基础，在快速动态变化的无线通信环境下、以及在差异化的多接收用户条件下，传统编码框架由于“悬崖效应”而面临着很大挑战。针对移动无线环境下的视频广播等典型应用，本项目围绕视频的高效软编码传输开展研究。具体包括：首先，研究了软编码中的视觉信号的空域-时域去相关变换、变换域视觉信号的能量分布建模和传输能量分配，分析不同信号变换的性能增益，从而构建了无线视频广播的软编码优化基本框架；其次，研究了视频软编码中的高效预测、数模混合编码、嵌入量化等几个对传输性能具有关键影响的核心问题，提升了软编码的性能；第三，针对人眼视觉特性研究了视觉感知优化的视觉信号软编码；最终形成了一种面向无线环境下视频通信的鲁棒而高效的视频软编码传输方案。本项目的研究工作在视频图像领域的高水平国际期刊上发表论文11篇，国际会议上发表论文14篇。项目支持了7名博士生和1名硕士生的培养，其中4名博士生已通过答辩获得博士学位。一项研究工作获得了CVPR2019 UG2+ Workshop最佳论文奖。