Abstract: Based on the Discrete Cosine Transform (DCT) coefficient histogram difference sequences of JPEG images, a new steganalytic method was proposed to realize the steganalysis of JPEG MB2. DCT coefficient histogram difference sequences based on JPEG images were defined as a measure of the correlation between JPEG images and its Cauchy fitting distributed model, and then the measure was used to construct the classification feature sequences to discriminate the stego-image from the carrier-image. The Hilbert-Huang transform decomposed the sequences with empirical mode and constructed Hilbert-oriented characteristic vectors. The experimental results demonstrate the correct rate to detect the hiding message achieves 84.48% with Support Vector Machines (SVM) classifier. The approach is also applicable for JPEG MB1 steganalysis.

Key words: steganalysis, JPEG MB2 stegonagraphy, histogram sequences, Hilbert-Huang Transform (HHT), sequential analysis

摘要： 基于JPEG图像的离散余弦变换(DCT)系数直方图差值序列，提出一种新的检测JPEG MB2隐写术的方法。该算法定义JPEG图像的DCT系数一阶直方图与其Cauchy拟合分布模型的系数直方图之间的差值序列作为相关性度量，在此基础上构造区分隐写图像和载体图像的分类特征序列。在特征分析中，通过Hilbert-Huang变换对特征序列进行经验模式分解，构建了基于Hilbert谱的特征向量。实验结果表明，基于该特征向量的支持向量机(SVM)对MB2隐写算法检测平均准确率达到84.48%，该方法也适用于JPEG MB1的检测。

关键词: 隐写分析, JPEG MB2隐写, 直方图序列, Hilbert-Huang变换, 随机序列分析