结合集成比例训练的彩色JPEG图像隐写分析

该文提出一种YCbCr颜色空间的彩色JPEG图像隐写分析方法。该方法中的特征包括通道内特征和通道间特征，首先从Y通道提取Markov特征，扩展DCT特征以及共生矩阵特征构成通道内特征集合，通道内特征可以有效捕捉到Y通道内DCT系数之间的相关性；然后对Y通道进行下采样，从采样平面与CbCr平面相互之间的差分平面上提取特征构成通道间特征集合，通道间特征可以捕捉到两两通道之间的相关性。由于通道内特征和通道间特征在分类性能上有着较大差别，在分类阶段由通道内特征和通道间特征分别训练子分类器，通过调整两类子分类器的比例，使用多数投票方式来合成集成判决结果，最终获得最佳的检测性能。实验结果表明，该方法不仅适合小嵌入率的彩色JPEG图像，而且在性能上优于已有的JPEG图像隐写分析方法。

Steganalysis for Color JPEG Images Based on Ensemble Proportion Training

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.