【作者】 王翠平；

【导师】 郭立；

【作者基本信息】 中国科学技术大学 ， 电路与系统， 2011， 博士

【摘要】 研究隐写分析对监视非法信息、防止机密信息泄露,保障国家安全和维持社会的稳定有着重要的意义。隐写分析的核心问题是选取检测率高,适用性广的特征向量。以往的文献缺乏对频域统计特征的深入研究,不能有效地对频域隐写方法进行检测。本文研究了音频DCT系数的统计分布以及隐写对DCT域统计特征的影响,在此基础上设计了两种新颖的隐写分析方法。为了提高检测率和检测尽可能多的隐写方法,提出了一种基于融合的隐写分析方法。本文的主要工作和创新点如下:1)针对音频DCT系数统计分布研究的不足,对音频DCT系数的统计分布进行了深入研究,为音频DCT系数建立了一个由广义高斯分布和α稳定分布组成的混合模型。研究结果表明,该模型比广义高斯分布和α稳定分布更接近于音频DCT系数的统计分布。2)提出了一种基于DCT直方图统计矩的音频隐写分析方法,采用的特征是DCT系数直方图的高阶统计量,实验结果表明该算法获得了较好的检测率。3)根据隐写会改变音频统计特征的假设,提出一种基于统计分布模型参数统计矩的音频隐写分析方法。实验数据表明,相比于同类算法,检测率有了较大的提高。4)针对现有隐写分析方法实用性不强的问题,基于融合思想提出了一种通用的隐写分析方法。该分析方法能够继承已有隐写分析方法的优点,提高隐写分析算法的检测性能。更多还原

【Abstract】 The research of steganalysis is significant to prevent disclosure of confidential information, monitor illegal information, protect national security and maintain social stability. Designing high detecting accuracy and wide applibality feature vectors is the core of steganalysis. Intensive research on statistical analysis of frequency domain in the literature is lack, so that the steanography which insert the secret message in frenquency domain can not be detected. In this paper an attempt has been made to analysis statistical distribution of audio DCT coefficients and the effect of steganography on the statistical nature. Based on the study, two novel steganalysis is designed. In order to improve the detection accuracy and enlarge the range of detected steganography, a universal steganalysis based on fusion technology is proposed.The main work and innovation are as follows:1) Aiming at the scarcity of statistical distribution of audio DCT coefficients, the statistical distribution of audio DCT coefficients is deliberated. A mixed model which is a linearly weighted average of generalized Gaussian distribution and alpha stable distribution is proposed. The experimental results show that the mixed model is near to the true distribution of DCT coefficients than generalized Gaussian distribution and alpha stable distribution.2) An audio steganalysis based on statistical moments of DCT histogram is proposed. The statistical moments of the histogram in DCT domain and the statistical moments of the histogram of the wavelet coefficients of every level in frequency domain are calculated as the features. Experimental results show that the proposed technique gain high detection accuracy.3) On the Assumption that steganography change the statistics of audio, a steganalysis based on the statistics of the parameters of statistical distribution is proposed. Experimental results show that compared to similar algorithms, detection rate has improved.4) For the defect of practicability of existing steganalysis, a universal steganalysis based on fusion theory is proposed. This method can inherit the advantages of exiting steganalysis algorithm, and then improve the detection performance.更多还原