Covert Communication via InfoGAN-Based Image Generation-Steganography

A covert communication technique based on image generation by InfoGAN, which harnesses InfoGAN's capability of reconstructing information, is investigated. In the proposed InfoGAN framework, secrecy data is embedded into randomly generated images through the generator model at the transmitter and a classifier model deployed at the receiver can decode the embedded information bits. It achieves covert transmission of the secrecy data while maintaining a high level of security and stealth. By leveraging the adversarial nature of InfoGAN, the method ensures that the hidden information is embedded into the images without compromising the visual quality, making it difficult for observers to notice the presence of the embedded data. Through experimental results, it is demonstrated that the proposed method not only successfully hides and restores the secrecy information, but also offers clear advantages in security and stealth. The generated images containing secrecy data are visually indistinguishable from images in the data set, ensuring that the hidden information is hard to detect. Furthermore, the method introduces an additional robust security layer through the adversarial learning framework of InfoGan, making it highly resilient to potential attacks.