GAN-based spatial image steganography with cross feedback mechanism

Steganography has been widely used to realize covert communication with multimedia content. By simulating the competition between a generator and a discriminator in a generative adversarial network (GAN), good distortion measurement can be automatically learnt to help designing high undetectable steganography. Nevertheless, existing GAN-based steganography always uses linear convolutional neural networks to construct generators, where the modification information of pixel in deep layers cannot be real-time transmitted to the neurons of the shallow layers, resulting in a low undetectable performance due to insufficient network training. To address the problem, we propose a new GAN-based spatial steganographic scheme. Different from existing GAN-based steganography, we build multiple cross feedback channels between the contraction path and the expansion path of the generator, which allows that the down-sampling information is directly sent to the expansion layer through these feedback channels. Since the detailed information from deep layers can be captured effectively by referring to the information from cover image, the generator can finally learn a sophisticated probability map to guide high undetectable steganography. Comprehensive experimental results demonstrate that, with the same steganalyzers, our scheme is superior to existing GAN-based steganographic schemes in terms of anti-steganalysis capability.