LAUSR

Many innovation modules introduced by High Efficiency Video Coding (HEVC) are remaining unexplored in the steganography domain. In this paper, a novel steganalytic approach is proposed against the PU partition mode-based steganographic methods which makes use of some innovative features in HEVC. Firstly, the influence of multilevel information embedding on the group proportion and the group proportion difference before and after recompression is analyzed. Then the statistical distribution of these two aspects are modeled as two different feature sets, and these two feature sets are combined for the final feature design to generate a 24-dimensional classification feature. Finally, a feature optimization is applied to reduce the 24-dimensional steganalysis feature to a 6-dimensional feature. It is demonstrated in experiment results that the proposed features have achieved a more accurate detection rate than current steganalysis methods under various circumstances, especially in the low embedding level situation.