一种基于图像内容的隐写分析方法

数字图像已成为信息隐藏的一类重要载体，然而不同内容的图像对隐写分析呈现出不同的检测性能. 该文基于图像信源区域平稳马尔可夫特性，用方差度量图像的区域复杂度，运用四叉树分割方法对图像进行递归分解，提取对数字隐写较为敏感的平坦区域，并从这些区域提取游程长度直方图统计矩特征，最后结合支持向量机实现隐藏信息检测. 实验结果表明，依据图像内容分割出敏感区域进行隐写分析可有效改善检测性能.

Steganalysis Based on Image Contents

Digital images are important covers in information hiding. It has been shown that steganalytic methods perform differently on images with different contents. Based on regionally stationary Markov characteristics of natural images, a new steganalysis method based on image segmentation is proposed. Quad-tree
segmentation based on local Variance is used to locate areas sensitive to message embedding. Features based on run length histogram are then extracted and optimized. Finally, detection of steganography is done using support vector machine. Experimental results show that the proposed steganalysis method outperforms previous methods in detection accuracy.