Wavelet Domain Steganalysis Based on Predictability Analysis and Magnitude Prediction

The accuracy of magnitude prediction is crucial for steganalysis schemes that use high order statistics in wavelet domain. The steganalysis performance can be improved by avoiding large prediction errors. In this paper, a statistical steganalysis algorithm is proposed based on predictability analysis and magnitude prediction of wavelet coefficients, which improves the steganalysis sensitivity by identifying potential locations with bad predictability. The weighting factors of the predictor, as well as the magnitude predictability, are derived from the local correlations of wavelet coefficients. Finally, stego images are distinguished from cover ones by analyzing the statistical properties of these prediction errors. Experiments show the performance enhancement due to bad points removing, and the proposed method is proved to be highly effective for the wavelet domain steganography.