A novel video steganography against local optimality-based steganalysis schemes

Abstract

Current state-of-the-art motion vector (MV)-based video steganography modifies MVs with some designed rules and the rules usually contradict the principles of the video codec, for example, the local optimality of MVs in rate-distortion sense. By checking the local optimality of MVs, prevalent steganalytic methods can detect most MV-based steganographic schemes. In this paper, a novel multiple reference frame motion estimation (MRF-ME)-based steganographic scheme to guarantee the local optimality of a modified MV is proposed. Firstly, a pre-selected stage is developed to collect the local optimal cover MVs, which provided preliminary local optimal information for the steganography system. Secondly, to eliminate the distortion drift introduced by embedding operation, the non-reference frames have a higher priority for embedding messages. Furthermore, we designed an efficient re-search method to find the local optimal candidate MVs in the MRF with a fixed search window. Finally, the candidate MV with the closest rate-distortion cost to the cover MV will be selected as the modified MV. To verify the security of the proposed scheme, experiments are carried out on several prominent MV-based steganalysis and steganography in various QP. The comparative results show the proposed method can resist the current best steganalysis while achieving good video compression performance.