Research on Image Semantic Transmission Scheme Based on Edge Information Assistance

Semantic communication technology can effectively reduce data transmission volume and improve transmission efficiency by replacing traditional symbolic information with semantic information. It has become one of the most promising research directions. At the same time, more and more research work has begun to explore the method of semantic communication realization, the development of deep learning technology provides a feasible solution for the realization of semantic communication, and this paper proposes an image semantic transmission scheme based on edge information assistance. When there is a significant deviation in structural and color information during semantic image restoration, the transmitter uses a panoramic segmentation neural network to extract semantic information, which is then combined with color information. Additionally, the Canny algorithm is applied to extract edge information from the original image to assist in image restoration. At the receiver, a diffusion probability model based on conditional information is employed, using edge information as a condition to guide the generation of semantic information. Finally, Polar code is introduced into the transmission process to improve the reliability of information transfer. The proposed scheme is compared with traditional image compression methods—JPEG, JPEG2000, and WebP—using SSIM and FID evaluation metrics. Results show that the proposed scheme not only achieves greater data compression but also maintains superior performance.