基于小波纹理分析的隐写分析研究(英文)

Steganalysis Techniques Based on Wavelet Texture Analysis

提出一种基于模型的隐写分析技术.图像的小波子带分解系数纹理可以建模为两参数广义高斯分布.同时采用极大似然方法进行两个参数的估计.分析发现隐写将改变图像的纹理特性,从而可以从子带分布模型参数的变化中判断是否隐藏信息.采用神经网络、支持向量机,回归分析和费歇尔判别分析分别进行比较验证.试验结果表明方法的有效性,同时对各分类器的性能进行了评价.

A novel steganalysis technique based on model is presented. The key element of the method is wavelet coefficients in each sub-band of wavelet transform are modeled as a Generalized Gaussian distribution (GGD) with two parameters. These two parameters of each subband coefficients are obtained by the maximum-likehood estimator. It appears that these parameters are a good measure of image features and can be used to discriminate stego-images from cover images. Neural network, SVM (support vector machine), regression analysis are adopted to train these parameters to get the inherent characteristic of cover and stego images. Experimental results show that the algorithm is comparable to previously existing techniques. And this method is a general steganalysis method which is applicable for the detection of data hiding and watermarking techniques.