Destroying robust steganography in online social networks

Abstract

In recent years, some robust steganography approaches for covert transmission in online social networks (OSNs) have been proposed. Some existing image algorithms, such as watermarking removal methods, have been developed to remove the hidden messages that might exist in an uploaded image. However, these methods may result in poor image quality after processing, which is not practical for OSN platforms. In this paper, we propose a general method for destroying robust steganography in OSNs that is able to preserve the image quality. In particular, we design a deep learning network including an attack module and an optimization module, and a new loss function is proposed to consider both. The experimental results indicate that the proposed method can be universally applied tocounter multiple robust steganography algorithms. Furthermore, the resulting processed image can be good quality, and the time cost is sufficiently low to meet most real-time requirements.