Steganalysis of perturbed quantization steganography based on the enhanced histogram features

In this paper, the enhanced histogram features are proposed for detecting perturbed quantization (PQ) steganography applied to double-compression JPEG image. Firstly, the principle of PQ steganography is analyzed and the special positions for feature extraction are determined. Secondly, the changes of the global, local and dual histogram features are analyzed for PQ embedding, and then these histogram features are extracted from the DCT coefficients at the special positions. Thirdly, to improve the effectiveness and diversity of steganalysis feature, the three kinds of histogram features are also extracted from DCT coefficients difference. Lastly, all the histogram features are calibrated and combined as the enhanced histogram features, and the ensemble classifier is employed to obtain detection results. The experimental results show the proposed feature can improve the detection accuracy for PQ and PQt; for PQe, it can obtain approximate detection accuracy with Cartesian-calibrated JPEG rich model (CC-JRM), but the feature dimensionality is far below CC-JRM.