基于权值分配的隐写分析算法

SRM算法是目前隐写分析中广泛使用的方法，但未能有效检测自适应隐写算法。为提高针对自适应隐写算法的检测率，该文通过改进SRM算法，利用不同区域的像素对隐写检测贡献的差异性，提出了一种基于权值分配的隐写分析算法。理论证明了权值分配能够提高隐写检测特征的分类能力，并设计了一种基于权值分配的特征提取框架。首先依据像素失真代价确定优先像素集，之后设计合理的权值函数对不同区域的像素噪声残差分配权值，最后提取四阶共生矩阵作为隐写检测特征。实验结果表明，在检测以HILL为代表的自适应隐写算法时，与SRM和PSRM检测算法相比，所提算法的平均错误率分别降低了2.09%和1.53%，说明能够有效实施针对自适应隐写算法的检测。

A Steganalysis Method Based on Weight Allocation

SRM (Spatial Rich Model) is one of the most widely used steganalysis methods but it fails to detect content adaptive steganography. In order to enhance the detection rate of those algorithms, this paper improves it and presents a novel steganalysis method based on weight allocation, considering the diversity of contribution those pixels make located in different region. It is proved theoretically that weight allocation can be used to enhance the classification ability of steganalysis features and then a framework for extracting features is designed. Firstly, preferential pixel set is determined according to embedding costs. Secondly, in order to assign weights to pixel noise residuals in different regions, a proper function is described. Finally, the forth-order co-occurrence is extracted as the steganalysis feature. Experimental results show that in comparison to SRM and PSRM(Projection Spatial Rich Model), the average error rate of the proposed method decreased by 2.09% and 1.53% when detecting HILL, therefore demonstrating the detection validity of content adaptive steganography.