Deep joint source channel coding model based on key information extraction for wireless image transmission

With the advancement of wireless communication technology, real-time image transmission faces challenges such as limited channel bandwidth, noise interference, and high computational resource requirements. Traditional image transmission methods, which rely on separate source and channel coding, struggle to balance transmission efficiency and image quality under varying channel conditions. To address these challenges, this paper presents an optimized Deep Joint Source-Channel Coding (JSCC) model that integrates key information extraction and background matching for high-resolution wireless image transmission. Our method leverages the Mediapipe library for accurate human detection and the Rembg library for efficient background removal. By extracting essential features and matching backgrounds from a pre-established library, we reduce the amount of data transmitted, thereby decreasing computational load and improving image quality. The proposed approach also includes resolution detection and compression to further optimize transmission under limited computing resources. Experimental results demonstrate that our method outperforms direct JSCC transmission, particularly under low SNR conditions, achieving higher Peak Signal-To-Noise Ratio (PSNR) and better visual quality. This approach offers a practical solution for resource-constrained applications in mobile communications and multimedia services.