基于流形学习的JPEG图像定量隐写分析算法

摘要

A quantitative steganalysis algorithm of JPEG images based on DCT coefficient statistical characteristics is pro-posed. Based on the research of DCT coefficient statistical model of JPEG images, the characteristic parameterα, which can reflect the embedding capacity change rule, is extracted in the proposed method. With the LIB-SVM classifier, a kind of manifold learning algorithm is applied to feature extraction from the characteristic parameters including α. By this means, the change ratio of DCT coefficients caused by steganography can be estimated. The results show that compared with other traditional quantitative analysis, the algorithm gains higher estimation accuracy and stability.%提出了一种基于DCT系数统计特性的JPEG图像定量隐写分析算法。该算法在对JPEG图像DCT系数的统计模型进行研究的基础上，提取了能够反映嵌入容量变化规律的特征参数α。以特征参数α为基础，提出了基于流形学习的特征提取算法，通过LIB-SVM分类器进行训练，估计隐写对DCT系数的更改比率。实验结果表明，与传统的定量分析算法相比，提出的算法具有更高的估计准确率和稳定性。