Plug-and-Hide: Provable and Adjustable Diffusion Generative Steganography

• DOI: 10.48550/arxiv.2409.04878
• Publication Date: 2024-09-07
• Authors: Jiahao Zhu; Zixuan Chen; Lingxiao Yang; Xiaohua Xie; Yi Zhou
• Keywords: FOS: Computer and information sciences; Computer Science - Cryptography and Security; Cryptography and Security (cs.CR)

Generative Steganography (GS) is a novel technique that utilizes generative models to conceal messages without relying on cover images. Contemporary GS algorithms leverage the powerful capabilities of Diffusion Models (DMs) create high-fidelity stego However, these algorithms, while yielding relatively satisfactory generation outcomes and message extraction accuracy, significantly alter modifications initial Gaussian noise DMs, thereby compromising steganographic security. In this paper, we rethink trade-off among image quality, security, accuracy within (DGS) settings. Our findings reveal normality DMs crucial factors can offer theoretically grounded guidance for DGS design. Based insight, propose Provable Adjustable Message Mapping (PA-B2G) approach. It can, one hand, guarantee reversible encoding bit from arbitrary distributions into standard DMs. On other its adjustability provides more natural fine-grained way trade off accuracy. By integrating PA-B2G with probability flow ordinary differential equation, establish an invertible mapping between secret be seamlessly incorporated most mainstream such as Stable Diffusion, necessitating additional training or fine-tuning. Comprehensive experiments corroborate our theoretical insights regarding in settings demonstrate effectiveness algorithm producing high-quality images preserving desired levels security