StegGAN: hiding image within image using conditional generative adversarial networks

Steganography is the art of hiding a secret message in another innocuous-looking image (or any digital media). Statistical imperceptibility is one of the major concerns for conventional steganography. In recent times, deep learning-based schemes have shown remarkable success in hiding an image within an image. However, a majority of these approaches suffer from the visual artifacts in the embedded and extracted images. In this paper, we have proposed a conditional generative adversarial network-based architecture for hiding an image within an image. The proposed method ensures the visual quality, statistical un-detectability as well as a noise-free extraction by incorporating the perceptual loss function and adversarial training. The proposed framework is tested on various datasets, and results have shown notable improvement ( ∼ 1 dB ) over existing methods. An ablation study is presented at the end of this paper to demonstrate the contributions of the various modules of the proposed architecture. Code is available at https://github.com/brijeshiitg/StegGAN .