Sign steganography revisited with robust domain selection

Abstract

To enable covert communication in the real world, robust steganography has attracted widespread attention. In this paper, a novel DCT sign-based robust steganography with robust domain selection is proposed, named Sign Steganography Revisited (SSR). The prior arts ensure the robustness by selecting robust cover elements, while the data embedding might make the selected elements invalid. This inspired SSR to select invariant stego elements before and after compression; meanwhile, error correction coding is removed to improve undetectability. Specifically, all the embedding domains are represented by the 8×8 DCT block, and the robust embedding domains with invariant stego elements are further selected. Therefore, the shared location information of selected domains is remarkably reduced. Besides, the error-free extraction is proved theoretically under compression within a specific range. Moreover, the undetectability of SSR is improved by correcting the distortion function. Experimental results show the effectiveness of the proposed SSR in lossy channels; the practicability of SSR is also verified on social network platforms.