RFIS-FPI: Reversible Face Image Steganography Neural Network for Face Privacy Interactions

Face information is vulnerable to theft and misappropriation in Internet face interactions, compromising user privacy. Image hiding can covertly embed confidential images in cover images while allowing seamless retrieval by recipients, providing new ideas for face privacy interactions. In this study, we present a novel framework, RFIS-FPI, which uses a target-protected hybrid policy combined with a reversible neural network to protect face image interactions. To address the character fidelity problem, this study employs a diffusion generation network that aims to generate cover images while preserving the essential properties of secret face images. To improve the covertness of the secret image, a low-frequency wavelet loss restriction is imposed on the information related to the secret image hidden in the high-frequency wavelet subbands, which significantly improves the security of the hidden image. Experimental evaluations show that RFIS-FPI excels in both face protection properties and hidden image detection, outperforming other state-of-the-art methods both qualitatively and quantitatively.