SDMStega:Robust Steganography Based on Stable Diffusion Model and Spread Spectrum Technology

Generative image steganography has drawn significant attention due to its ability to embed secret information during the image generation process. However, existing methods still have limitations in terms of controllability, applicability, and robustness, restricting their practical applications. To address these issues, we propose the SDMStega, a practical and robust generative image steganography method based on a latent diffusion model. By introducing spread spectrum communication techniques, SDMStega enhances the concealment and robustness of the embedded information, ensuring that secret messages are more resistant to disfo and variants of attacks. Additionally, we incorporate the DDIM process to refine the steganographic latent representation, leveraging forward and reverse diffusion to improve the image quality while preserving the hidden information intactly. The overall process involves generating a cover image using a diffusion model, transforming the latent space via a VAE encoder, embedding the secret information, and applying DDIM-based denoising before decoding it back into the stega-image. Upon receiving the image, the recipient extracts the secret information through VAE encoding, DDIM reverse diffusion, and spread spectrum decoding. Experimental results demonstrate that SDMStega not only achieves high invisibility of generated content but also ensures efficient and reliable secret information retrieval. Furthermore, it maintains excellent steganographic performance under common JPEG compression and exhibits superior resilience against image processing operations and attacks, offering new possibilities for secure communication.