Semantic Communication Based on Content Style Adaptation for Image Transmission

Mengshu Song, Nan Ma, Yiming Liu, Linzi Shen, Chen Dong

Abstract: The development of semantic communication has promoted the research on image transmission. Content and style are two crucial characteristics of image information. Existing studies have explored image information transmission and recovery methods in semantic communication systems, but do not consider both image style information and content information simultaneously. To address this issue, based on deep joint source-channel coding (JSCC) structure, a novel system named semantic communication based on content style adaption (CSASC) for image transmission is proposed in this paper. Specifically, a novel semantic information extraction method are designed to extract the semantic content information and semantic style information of images, meanwhile an image content-style adaptive reconstruction method is proposed by fusing semantic content and semantic style features to improve the image reconstruction accuracy. In addition, the impact of the weighting of semantic style and content information on image reconstruction quality is analyzed. Results demonstrate that the proposed method has better recovery results compared with existing image semantic communication methods.