Semantic communication,as an emerging paradigm,enhances communication efficiency by directly transmitting the semantic meaning of information to meet diverse application requirements.It is also referred to as task-oriented communication.Video semantic coding,as a critical component of semantic communication,has gradually become a research hotspot.This paper reviews the research progress in video semantic coding,categorizing existing methods into two types:deep learning-driven coding frameworks and traditional-intelligence collaborative coding frameworks.First,deep learning-driven coding frameworks are introduced,including feature stream coding,video stream coding,and human-machine collaborative coding.These approaches rely on fully neural network-based architectures and achieve performance improvements through end-to-end optimization but face challenges due to their high computational resource demands,which limit practical deployment.Next,traditional-intelligence collaborative coding frameworks are discussed,combining the strengths of traditional coding techniques and artificial intelligence to enhance system performance and flexibility.Finally,the challenges faced by video semantic coding are summarized,and future development directions are outlined.
维普
万方数据