Image Steganography based on Style Transfer

Traditional embedding-based image steganography embeds information by modifying the pixels of a natural image, which inevitably results in modification traces. We propose an image steganography method based on style migration. The method performs secret information hiding during the process of style transfer of an image, and uses the feature transformations brought about by the style migration to mask the steganography behavior. Meanwhile, the style-specific content-aware loss function is used to preserve the style-related content features to provide space for the embedding of secret information. Compared to other steganographic approaches utilizing style transfer, the method proposed in this paper significantly enhances the steganographic capacity for embedding secret information in stylized images. Furthermore, the special design of the network structure ensures the accuracy during extraction. Finally, the effectiveness of the method proposed in this paper is verified through a large number of experiments.