Steganography-based facial re-enactment using generative adversarial networks

This paper presents a technique for hiding secret messages in images while transferring them over a network using steganography. The preprocessed standard datasets create steganographic datasets for facial re-enactment purposes. The facial re-enactment GAN (FRe-GAN) technique and qualitative and quantitative results have been presented over various datasets. A comparative study has been conducted that showcase the drawbacks of existing literature and motivated their work. We propose a steganography-based GAN model and used benchmark datasets such as Flickr-Faces-HQ (FFHQ), IMPA-FACE3D, FaceForensics++, and CelebFaces Attributes (CelebA) facial datasets in the experimentation. We have derived a Generative Adversarial Networks-based approach to face re-synthesis and re-enactment that adjusts for facial expressions and pose. The face blending network is used to blend two faces seamlessly. We have compared the proposed approach with existing state-of-the-art systems and show that our method achieves qualitatively and quantitatively better results.

Automated summary

This paper presents a technique for hiding secret messages in images using steganography during network transfer. The effectiveness of the facial re-enactment GAN (FRe-GAN) technique is shown through qualitative and quantitative results on various datasets. A comparative study is conducted to highlight the limitations of existing literature and motivate the proposed approach. The proposed steganography-based GAN model is experimented using benchmark facial datasets, and achieves superior qualitatively and quantitatively results compared to state-of-the-art systems.