Coverless Steganography for Face Recognition Based on Diffusion Model

As a highly recognizable biometric face recognition technology, it has been widely used in many identity verification systems. In order to enhance the protection of personal privacy and ensure the safe transmission and sharing of sensitive information without affecting the user experience, this paper proposes an innovative coverless steganography framework for face recognition images based on diffusion model. The framework firstly extracts face features and generates masks containing these features. Then, combined with conditional diffusion model and text key, a deterministic Denoising Diffusion Implicit Model (DDIM) is used to sample coverless steganography images. Secret images can also be recovered in high quality with DDIM Inversion technology. A large number of experiments show that compared with the existing methods, this approach has markedly enhanced the quality of steganographic and restored images. The face recognition rate of the restored image is more than 96%, which can effectively replace the original image for face recognition. The detection accuracy of this method is 55.25% on the steganographic detection tool, which is closer to random guessing and can resist steganographic analysis. It ensures the higher security of hidden images and solves the limitation of existing methods in protecting the privacy of face images. Moreover, it is shown how to achieve controlled local steganography with a custom mask, which enhances the controllability and flexibility of the method. In conclusion, the proposed method outperforms traditional steganography in security, controllability and robustness, and provides an effective technical scheme for steganography protection of face recognition images without additional training.