基于学习的视频编码技术进展  

Advances in learning⁃based video coding technologies

作　　者：朱秀昌[1] 唐贵进[1] ZHU Xiuchang;TANG Guijin(Jiangsu Province Key Lab on Image Processing&Image Communication,Nanjing University of Posts and Telecommunications,Nanjing 210003,China)

机构地区：[1]南京邮电大学江苏省图像处理与图像通信重点实验室,江苏南京210003

出　　处：《南京邮电大学学报（自然科学版）》2022年第2期1-12,共12页Journal of Nanjing University of Posts and Telecommunications：Natural Science Edition

基　　金：国家自然科学基金面上项目(61071091);“信息与通信工程”江苏高校优势学科建设工程资助项目。

摘　　要：近来,深度神经网络为视频编码提供了一种新的有前途的解决方案,即基于学习的视频编码技术。文中着重回顾了两类基于学习的视频编码技术,一类是用神经网络替代或协助传统编码框架中某些“功能模块”的压缩,另一类是全神经网络实现的“端到端系统”压缩。分别给出了两类技术中一些近来有代表性的研究成果,展示了它们的优越之处、长足进展及发展潜力。在给出这类技术目前存在问题的同时,也简要展望了今后的研究方向。Deep neural networks have provided a new and promising solution for video coding,namely learning⁃based video coding techniques.This paper reviews the two types of learning⁃based video coding technologies:(1)methods that use neural networks to replace or assist the compression of certain functional modules in the traditional coding framework,and(2)the end⁃to⁃end system compression implemented by full neural networks.Some recent representative studies on the two types of technologies are summarized,and their advantages,great progresses and development potentials are elaborated.Finally,we present problems of these technologies,and identify the future research directions.

关 键 词：视频编码 神经网络 深度学习 编码工具 端到端系统 

分 类 号：TN919.8[电子电信—通信与信息系统]