Video steganalysis of LSB based motion vector steganography

This paper proposes a novel flatness measure for video steganalysis targeting LSB based motion vector steganography. The proposed method has introduced two major improvements. Firstly, unlike previous approaches, it takes into account the anchor frame and current frame distances and directions, which significantly affect the correlation strength of adjacent motion vectors. Secondly, it defines a cover model that does not require a training based machine learning system. Large amounts of videos including broad range of motion patterns have been used in order to approximate the real detection rate of the proposed method. Test results show that proposed algorithm successfully classifies cover and stego videos.