基于核主元统计的音频隐写分析

Audio steganalysis based on kernel principal component statistics

选取隐写前后音频载体信号小波分解第二级细节分量进行分帧,利用核主元分析原理提取每帧的不重要核主元,进行形态学变换后,以其相邻两帧汉明距离的奇数阶中心矩作为特征向量,用支持向量机分类,隐写分析全局检测率可超过95%。

Selects the second level detail coefficients from original and stego audio carrier wavelet decomposition to divide as frames,and makes use of the principle of kernel principal component analysis to extract least kernel principal components of every frame.Following morphology transformation,odd order center moment of Hamming distance of its two neighbouring frames is used as eigenvector,and classed by support vector machine.The synthetical test accuracy of steganalysis can be attained over 95%.