基于主成分分析和RBF网络的泛盲掩密分析方案

A Universal Blind Steganalysis Based on Principal Component Analysis and RBF Network

针对Farid泛盲掩密分析方法所选的图像分类特征数目多而且具有相关性的缺陷,采用主成分分析技术对特征进行去相关性的预处理,并基于RBF网络提出了新的掩密分析方案。该方案不但大大降低了用于分类的图像特征的维数,从而提高了掩密分析速度,而且提高了掩密分析的检测性能。分别利用该方案和Farid的方案对JSteg等软件掩密后的图像进行检测,比较实验结果表明,对于不同长度的嵌入消息,该算法具有更好的检测性能。

On the basis of the limitation of Farid's universal blind steganalysis that image features for the classification are too many and have correlations, a new universal blind steganalysis scheme is presented. Decorrelation preprocessing using principal component analysis on the image statistics features and steganalysis classifier using RBF network are proposed. The scheme not only reduces the dimension of the feature vector enormously so that the speed of steganalysis detection is increased, but also the performance of the detection is improved as well. Applying this scheme and Farid's scheme to detecting the stego images produced by JSteg, EZStego, and S_Tools respectively, the comparison of these simulation results shows that this scheme is quite efficient for different size of messages embedded.