音频盲取证中一种基于EM的重采样检测方法  

A method of detecting re-sampling based on expectation maximization applied in audio blind forensics

作　　者：陈雁翔[1] 吴玺[1] 

机构地区：[1]合肥工业大学计算机与信息学院情感计算与先进智能机器安徽省重点实验室,安徽合肥230009

出　　处：《电路与系统学报》2012年第4期118-123,共6页Journal of Circuits and Systems

基　　金：国家自然科学基金(60672120);安徽省自然科学基金(11040606M127);国家高技术研究发展计划(863计划;2012AA011103)

摘　　要：盲取证指针对篡改信号无需添加任何附加信息就可鉴别出信号的真伪,而音频篡改中篡改者经常利用重采样以达到更好的篡改效果,因此重采样检测作为音频盲取证的重要组成部分得到了高度的重视。对信号的重采样会引入相关性,这种相关性是周期出现的,本文通过基于期望最大化的检测方法揭示这种相关性,并通过判断这种相关性是否呈现周期性达到检测目的。在检测流程中采取了奇异防止、低频段去除、归一化三阶原点矩等措施,达到了更好的检测效果。实验验证了该方法对于各种插值函数的鲁棒性,以及不同重采样率下和音频拼接篡改时检测的有效性。Blind forensics can discriminate such tampered signal wl＊thout any additional information. During the audio tampering tampers often use re-sampling to achieve better results, thus re-sampling detection obtained the high value as an important aspect of blind forensics. Re-sampling will introduce correlation and this relationship is often periodic. We use method based on Expectation Maximization to reveal the correlation and reach the purpose of detection by checking whether the correlation is periodic. To achieve better effect, we also take the steps of singularity prevention, low-band elimination and normalized 3-order origin moment. The experiment results show the robustness of interpolating functions, and the validity of detection under different re-sampling rates as well as during splice tampering.

关 键 词：重采样检测 盲取证 期望最大化 相关性 

分 类 号：TN912.34[电子电信—通信与信息系统]