Abstract:

Steganography by texture image synthesis was a kind of common carrier synthesis steganography method. However, the texture image did not have semantic features, so this kind of method was easy to attract the attention of attackers after multiple transmission. The generative adversarial networks used game strategy to make the generator confront the discriminator. In theory, the generator with the best training could make the distribution of the generated samples the same as the real data. In ideal conditions, using GAN to realize synthesis steganography could construct a natural image. One of the problems of the image synthesis steganography based on GAN was that it couldn’t control the content of the generated image. To solve this problem, this paper proposed an image steganography method based on conditional generative adversarial networks. In this method, the combination of random noise and condition information was used as the representation of hidden space to train the generator, so that the generated image was controlled by condition information. The combination of generated image and condition information was used as the representation of probability space to train the extractor, so that the extracted noise was consistent with the driving noise. The experimental results showed that this method could complete the function of generating the image and extracting the message. The outstanding feature was that it could use the condition information to control the content of the image. At the same time, the image quality and accuracy rate of the message extraction were close to the method in the comparison.

Key words: generative adversarial networks, conditional generative adversarial networks, image steganography