视频会议环境下面向语义通信的高鲁棒视频重建方法  

High Robust Video Reconstruction Method for the Semantic Communication in the Video Conference Environment

作　　者：郭子睿 孙鲁楠 周英男 杨洋[1] 杜忠田 郭彩丽[1] GUO Zirui;SUN Lunan;ZHOU Yingnan;YANG Yang;DU Zhongtian;GUO Caili(Beijing Laboratory of Advanced Information Networks,Beijing University of Posts and Telecommunications,Beijing 100876,China;China Telecom Digital Intelligence Technology Co.,Ltd.,Beijing 100035,China)

机构地区：[1]北京邮电大学北京先进信息网络实验室,北京100876 [2]中电信数智科技有限公司,北京100035

出　　处：《移动通信》2024年第2期83-89,110,共8页Mobile Communications

基　　金：北京市自然科学基金“基于语义率失真的语义通信理论与系统架构研究”(L222043);国家自然科学基金“基于知识增强信息瓶颈的语义编码理论与方法研究”(62371070)。

摘　　要：在传统的视频会议场景中,如果用户网络带宽不足,就会出现严重的时延,卡顿现象,从而导致用户体验较差。与此同时,随着深度学习等技术的发展,目前已经出现了效果逼真的视频重建方法。现有的基于深度学习的视频重建方法可以很好地解决传统视频会议技术的带宽不足问题,其基于语义通信有广阔的应用前景,然而,当前视频重建方法在面部大幅扭动情况下重建效果差。针对这一挑战,提出了一种视频会议环境下面向语义通信的高鲁棒视频重建方法。首先,改进了现有的视频重建算法,引入来自于原始视频的压缩语义特征构建了新的面向语义的视频重建模型,其次,针对视频会议场景,提升了重建分辨率并设计了完整的视频重建方法。最后,实验和仿真结果表明,所提方法相比于传统视频会议方案降低了约三分之二的带宽,同时所提方法的重建效果,相比于当前基于深度学习的视频重建效果更具有鲁棒性,验证了所提视频重建方法的性能优势。In traditional video conferencing scenarios,if the user's network bandwidth is insufficient,serious latency and lag phenomena willoccur,leading to poor user experience.Meanwhile,with the continuous development of technologies such as deep learning,a series of efficient and realistic video reconstruction methods have emerged,providing new possibilities for solving the problem of insufficient bandwidth in traditional video conferencing techniques.These video reconstruction methods based on deep learning have shown great potential in semantic communication,however,the current challenge is that the reconstruction results of these methods are still not satisfactory when the face is greatly twisted.To overcome this problem,this article proposes an innovative video reconstruction method specifically designed for semantic communication in video conferencing environments.Firstly,an improvement was made to the existing video reconstruction algorithm by introducing compressed semantic features from the original video,and a new semantic communication oriented video reconstruction model was constructed.Secondly,in response to the special needs of video conferencing scenarios,the reconstruction resolution has been improved and a complete video reconstruction strategy has been designed.The experimental evaluation results show that the proposed method reduces bandwidth requirements by about two-thirds compared to traditional video conferencing solutions.More importantly,this method outperforms current deep learning based video reconstruction methods in the face of significant facial contortions,thus verifying its significant advantage in improving video conferencing performance.Overall,this study not only addresses the urgent issue of bandwidth constraints,but also provides an innovative solution to improve the quality and robustness of video conferencing experiences.With the continuous development of the digital field,the video reconstruction method designed for semantic communication is expected to provide new solutions for vi

关 键 词：语义通信 视频会议 视频重建 深度学习 

分 类 号：TN925[电子电信—通信与信息系统]