基于特征排名的图像隐写分析算法

为提高用于隐写分析的集成分类器的检测精度,提出一种基于特征排名的隐写分析算法。首先计算每维检测特征的互信息得分并根据得分高低将特征进行排名,然后设置分界点将特征分为重要特征区域与普通特征区域,依据设定的抽样比例从两个区域随机抽取特征组成不同的特征子空间并训练集成分类器。最后使用集成分类器进行分类。实验结果表明,针对使用nsF5及S-UNIWARD算法进行隐写的频域及空域图像,本算法较传统分类器在检测错误率方面分别平均下降约0.006 5和0.006 2,具有较好的检测效果。针对频域与空域中两种不同的隐写算法,与传统的集成分类器相比,该算法具有更高的检测精度。

Image steganalysis algorithm based on feature ranking

In order to enhance the detection rate of ensemble classifier, an algorithm based on feature ranking is proposed. First, the original feature is sorted according mutual information score. Then the sorted feature is divided into important feature part and common feature part according to the divide point. The feature subset is formed by selecting features randomly in each part according to the given sampling rate. Experimental results show that, our method detects two format images after embedding nsF5 and S-UNIWARD, the false detection is lower than classical ensemble classifier 0.0065 to.jpeg images, and the false detection is lower than 0.0062 to.bmp images. Compared with typical ensemble classifier, the proposed method is more effective than the different stego algorithms in frequency domain and spatial domain.