基于HCF统计特征的MIDI音频隐写分析

MIDI Audio Steganalysis via HCF-based Statistical Features

针对乐器数字接口(MIDI)音频三种常见的LSB隐写方法:最低位替换、最低位匹配和低两位替换隐写,为了提高隐写检测正确率,提出基于直方图特征函数(HCF)统计特征和支持向量机(SVM)的隐写分析方法,通过提取MIDI音频力度分量直方图特征函数域21维特征,用支持向量机训练分类器对MIDI音频进行分类。实验表明,当嵌入率大于10%情况下,此隐写分析方法对三种LSB隐写方法的平均分类正确率达90%以上。

In view of 3 kinds of LSB Steganography for MIDI audio,that is,LSB replacement,LSB matching and 2LSBs replacement,a method of steganalysis is proposed,in order to improve steganalysis accuracy based on the statistical features of the histogram characteristic function and SVM classifier.The 21-dimensional statistical features of the histogram characteristic function are extracted to classify original MIDI audio and stego MIDI audio.Experiments show that when embedded rate is more than 10%,steganalysis by the proposed method could detect MIDI audio with an average correct decision rate of above 90%.