基于区域相关性的JPEG图像隐写盲检测方法

Regional Correlation Based Blind Detection of JPEG Image Steganography

针对JPEG压缩算法,分析了DCT系数的分布特点和统计特征以及块内和块间系数的相关性,提出一种基于区域相关性的JPEG图像隐写盲检测算法。采用宏观校准和微观校准估计出载体图像,以Markov模型描述块内和块间DCT系数的相关性,提取概率转移矩阵的差值作为检测特征,通过支持向量机进行分类,实现了对DCT域多种隐写算法的检测。混合图像库的实验结果表明,该方法的检测性能优于典型的隐写分析算法。

Targeting at current compression algorithms for JPEG images, the distribution and statistical features as well as intra-block and inter-block correlations of Discrete Cosine Transform（DCT） coefficients are analyzed in depth. A blind steganalysis approach for JPEG images based on region correlation is proposed. First, macroscopic and microscopic calibrations are combined to estimate cover images. Then, intra-block and inter-block correlations of DCT coefficients are described according to the Markov model. Finally, the differences of transition probability matrices are extracted as characteristics for detection, which are classified by a support vector machine. Hence steganographic algorithms for JPEG images can be determined. Experiments on a mixed image database reveal that the proposed steganalyzer outperforms existing typical steganalyzers.