Steganalysis of DCT-embedding based adaptive steganography and YASS

• Authors: Qingzhong Liu
• Venue: Proceedings of the thirteenth ACM multimedia workshop on Multimedia and security
• Year: 2011

Recently well-designed adaptive steganographic systems, including ±1 embedding in the DCT domain with optimized costs to achieve the minimal-distortion [8], have posed serious challenges to steganalyzers. Additionally, although the steganalysis of Yet Another Steganographic Scheme (YASS) was actively conducted, the detection of the YASS steganograms by a large B-block parameter has not been well explored. In this paper, we aim to detect the state-of-the-art adaptive steganographic system in DCT-embedding and to improve the steganalysis of YASS. To detect DCT-embedding based adaptive steganography, we design the features of differential neighboring joint density on the absolute array of DCT coefficients between the original JPEG images and the calibrated versions. To discriminate YASS steganograms from covers, the candidate blocks that are possibly used for embedding and the non-candidate block neighbors that are impossibly used for information hiding are identified first. The difference of the neighboring joint density between candidate blocks and the non-candidate neighbors is obtained. Support Vector Machine (SVM) and logistic regression classifiers are employed for classification. Experimental results show that our approach is very promising when detecting DCT-embedding based adaptive steganography. Compared to the steganalysis based on CC-PEV feature set, our method greatly improves the detection accuracy; the advantage is especially noticeable in the detection of the steganograms with low relative payload. In steganalysis of YASS, our approach is superior to a previous well-known steganalysis algorithm; our method remarkably improves the detection accuracy especially in the detection of the YASS steganograms that are produced with a large B-block size, which was not well addressed before.