【作者】 孔祥维； 李玲玲； 尤新刚；

【Author】 Kong Xiangwei~1,Li Lingling~1,and You Xingang~(1,2) 1(Information Security Research Center,Dalian University of Technology,Dalian,Liaoning 116024) 2(Beijing Institute of Electronic Technology and Application,Beijing 100091)

【机构】 大连理工大学信息安全中心； 北京电子技术应用研究所；

【摘要】 针对现有F5定量隐密分析方法仅考虑直方图收缩导致定量估计误差较大的问题,提出了一种新的基于多特征支持向量回归(SVR)的定量隐密分析算法.算法通过最大似然估计拟合DCT系数分布模型的参数,利用参数的相对变化量来反映嵌入机制对直方图的修改,并使用共生矩阵描述F5算法对图像像素间相关性的破坏.在确定特征与嵌入量之间的函数关系时,算法采用SVR在高维空间中进行拟合,从而估计待测图像的容量.实验表明,该算法可以对F5算法的嵌入量进行有效估计,而且估计效果要优于已有的算法.更多还原

【Abstract】 Focusing on the cause that the existing F5 quantitative steganalysis algorithm which uses the feature caused by the shrinkage of image histograms has low precision,a new algorithm using several features and SVR to estimate the capacity is proposed.The image features include thirteen model-based parameters which can be computed by maximum likelihood reflecting the changes of histograms caused by the embedding mechanism and twenty-five co-occurrence characters which are defined to describe the changes of DCT coefficients correlation.SVR is used to model the relationship between features and the message length in the high dimension,and then the embedding capacity can be predicted using the model.Experimental results show that the proposed algorithm can produce high precision for F5 embedding mechanism and performance better than the existing method.更多还原