CRoSS: Diffusion Model Makes Controllable, Robust and Secure Image Steganography

• DOI: 10.48550/arxiv.2305.16936
• Publication Date: 2023-01-01
• Authors: Jiwen Yu; Xuanyu Zhang; Youmin Xu; Jian Zhang
• Keywords: Robustness; Steganalysis

Current image steganography techniques are mainly focused on cover-based methods, which commonly have the risk of leaking secret images and poor robustness against degraded container images. Inspired by recent developments in diffusion models, we discovered that two properties ability to achieve translation between without training, noisy data, can be used improve security natural tasks. For choice model, selected Stable Diffusion, a type conditional fully utilized latest tools from open-source communities, such as LoRAs ControlNets, controllability diversity In summary, propose novel framework, named Controllable, Robust Secure Image Steganography (CRoSS), has significant advantages controllability, robustness, compared methods. These benefits obtained additional training. To our knowledge, this is first work introduce models field steganography. experimental section, conducted detailed experiments demonstrate proposed CRoSS framework security.