提出了用于彩色图像通用隐写分析的3类统计特征。针对隐写所导致的图像DCT系数分布、空域像素值平坦性以及颜色空间一致性的改变,从系数分布模型背离程度、空域相关性和不同颜色分量间的相关性3个方面设计能够体现原始载体和含密图像差异的特征,提取了10维特征向量。这些特征对隐写行为较为敏感、涵盖面广,且维数低、计算方便。在这些特征的基础上,可利用支持向量机分析彩色JPEG图像是否含有秘密信息。实验结果表明,基于这些特征的通用隐写分析方法可以准确、高效地检测以Jstge、F5或MB隐写方法嵌入的秘密信息,辨别隐写方法种类,具有很好的通用性。
In this paper we propose three types of features which can be used to realize efficient universal steganalysis on color images. After measuring the deviation of DCT coefficient distribution, the correlation of smoothness in spatial domain, and the correlation between different components, a 10-dimentional feature vector is produced for each color image. Those features are sensitive to steganographie process and the feature vector, having a low dimension, is easy to calculate. Then, a SVM classifier is built to distinguish the stego-images from the original images. Experiments show that the universal steganalytic approach can effectively detect the presence of secret message hidden by Jsteg, F5 or MB steganography, and can identify the steganographic techniques used.