Steganalysis for Color JPEG Images Based on Ensemble Proportion Training

Feng-yong Li,Xin-peng Zhang,Jiang Yu
DOI: https://doi.org/10.3724/SP.J.1146.2013.00443
2014-01-01
Abstract:A new steganalytic scheme of color JPEG images is proposed based on YCbCr color space. The features of the proposed scheme include intra-channel features and inter-channel features. The intra-channel features are formed by Markov features, extended DCT features and co-occurrence matrices features and capture effectively the dependency among DCT coefficients in Y channel. The inter-channel features are extracted in difference planes between channels, which can effectively capture the dependency between channels. In the classification process, the intra-channel and inter-channel features are respectively used to train sub-classifiers. By adjusting the proportion of two kinds of sub-classifier, the optimal decisions are synthesized by using majority voting. Experimental results show that proposed scheme is applicable to low embedding color JPEG images and the performance outperforms some state-of-the-art feature sets.
What problem does this paper attempt to address?