Generative Image Steganography Based on GANs

Abstract

According to the embedding method of secret information, steganography can be divided into: cover modification, selection and synthesis. In view of the problem that the cover modification will leave the modification trace, the cover selection is difficult and the load is too low, this paper proposes a generative image steganography scheme based on GANs, which combines with cover synthesis. Based on GAN, the scheme uses secret information as the driver and directly generates encrypted images for transmission, which can effectively resist the detection of steganalysis algorithms. The security of the scheme is based on the key of the encryption algorithm. Even if the attacker obtains the transmitted information, only the meaningless result will be obtained without the key. Experiments were carried out on the data set of CelebA, and the results verified the feasibility and security of the scheme.