Low Resolution Coverless Image Steganography for High Resolution Images using Generative Adversarial Networks

Abstract

Abstract Traditional image steganography techniques modify cover images to hide information, making them vulnerable to detection. Coverless image steganography eliminates this dependency but often struggles to maintain high reconstruction fidelity while ensuring robust concealment. This study proposes a novel coverless image steganography method using Generative Adversarial Networks (GANs) to address these limitations. GANs excel at learning complex data distributions and generating realistic stego images that e!ectively encode high-resolution secret images, even after aggressive downscaling. By downscaling secret images prior to embedding, our approach achieves a fourfold increase in embedding capacity while reducing computational burden and maintaining competitive reconstruction quality as measured by SSIM, PSNR, and MAE metrics. Extensive experiments demonstrate superior performance in embedding capacity, reconstruction fidelity, and concealment robustness compared to traditional multi-image steganography techniques. This research provides an e!ective solution for secure high-resolution image embedding, advancing coverless image steganography.