Digital Forensics and Watermarking: 18th International Workshop, IWDW 2019, Chengdu, China, November 2–4, 2019, Revised Selected Papers

Steganography has been widely used to conceal secret information in multimedia content. Using generative adversarial networks (GAN) where two subnetworks compete against each other, steganography can learn good distortion measurement. Nevertheless, the convergence speed of GAN is usually slow, and the performance of GAN-based steganography has large room to improve. In this paper, we propose a new GAN-based spatial steganographic scheme. The proposed learning framework consists of two parts: a steganographic generator and a steganalytic discriminator. The former generates stego images, and the latter evaluates their steganography security. Different from existing GAN-based steganography, we reconstruct the generator by combining multiple feature maps, and then expand the maximum number of feature channels to 256. The reconstruction generator can effectively generate a sophisticated probability map, which is used to calculate optimal distortion measurement and further provides a better guidance for adaptive information embedding. Comprehensive experimental results show that, with the same discriminant network, the anti-steganalysis performance of our method is better than that of ASDL-GAN scheme and Yang’s scheme.