Multi-resolution Steganalysis of JPEG Images Using the Contrast of DCT Coefficients

This paper proposes a novel JPEG steganalytic scheme, which combines the designing of feature called as the Contrast of Discrete Cosine Transform (DCT) coefficients, multi-resolution decomposition. Unlike the conventional steganalytic schemes, without using the co-occurrence to show directly the correlation between the DCT coefficients, the original JPEG image is decompressed into spatial domain and, applying the Fast Discrete Curvelet Transform (FDCT), the spatial version is resolved into several multi-resolution subimages. After transforming all the spatial subimages into JPEG image, the contrast of two DCT coefficients is turned into an angle and the l2 norm is used as the weight of the angle. Therefore, the new feature is seen as a union of the first-order statistics of weighted angle. On modern stegnographic algorithm JPEG Universal Wavelet Relative Distortion (J-UNIWARD), we demonstrate the proposed scheme achieves more accurate detection than Cartesian-calibrated feature (CFstar) and is superior to the Cartesian-calibrated JPEG Rich Model (CC-JRM) at lower embedding rate.