Joint Source-Channel Coding for Image Super-Resolution Tasks in Semantic Communications

Image super-resolution (SR) technology is crucial for applications such as telemedicine. Performing image SR tasks using semantic communication allows the received information to be directly applied without the need for complex operations at the receiver. However, research still grapples with the challenges of constructing a joint source-channel coding (JSCC) method that effectively extracts semantic information from LR images for reconstructing HR images, as well as reducing the unfavorable impact of the wireless channel. In this paper, we propose a deep semantic information extraction (DSIE) module for the image SR task in semantic communications. This module ingeniously integrates the advantages of convolutional neural network (CNN) and Transformer architectures and incorporates signal-to-noise ratio (SNR) to dynamically adjust the output via the channel attention mechanism. Building on DSIE, we custom-design the JSCC scheme for image SR tasks in semantic communications. Simulation results confirm the effectiveness of our proposed DSIE and show that the proposed JSCC scheme surpasses the existing DNN-based JSCC schemes and the traditional separated-source channel coding schemes.