低速率语音码流中的码元替换信息隐藏检测

Steganalysis for Steganography Based on Code Units Replacement in Low Bit Rate Audio Streams

低速率语音码流中的码元替换信息隐藏选取编码压缩语音中的特定码元作为载体,可实现高嵌入率和较高隐蔽性的信息隐藏.本文提出了码元关联网络模型,既可以从空间角度描述并量化同一语音帧内不同码元的取值分布特性,又能从时间角度描述并量化不同语音帧内同一码元的取值关联特性。进一步,基于码元关联网络构建高性能分类器,最终可高效地实现针对码元替换信息隐藏的隐写检测。在不同语音数据库来源和不同语音长度条件下的实验结果证明了本文方法的检测准确率优于现有的隐写检测算法,实现了快速有效地码元替换信息隐藏检测。

Steganography based on code units replacement in low bit rate audio streams, which selects certain code units in the encoded and compressed speech as the cover, can hide information in the speech with large embedding capacity. This paper proposes three dif- ferent structure Bayesian networks specially to detect this type of steganography. We define the variable nodes in the Bayesian networks according to the distribution characteristics of the values of the changed code units by steganography, and establish three different net- work structures considering the periodicity of speech. The networks learn the parameters combing with data to efficiently classify the speech samples （embedded or not） at last. Experiment results under different speech source conditions and different speech length conditions prove that our steganalysis method outperforms current steganalysis method. And our method can effectively and quickly de- tect steganography based on code units replacement in low bit rate audio streams.