基于均值方差评价系数与多分类器的AMR语音隐写分析方法  

Steganalysis of AMR Speech Based on Mean Variance Evaluation Coefficient and Multiple Classifiers

作　　者：吴彦鹏[1] 吴倩 陈明辉[1] 张辉极[1] WU Yan-peng WU qian CHEN Ming-hui ZHANG Hui-ji(Xiamen Meiya Pico Information Co. , Xiamen 361008, China Cyber Security Department, Beijing 100006, China)

机构地区：[1]厦门市美亚柏科信息股份有限公司,厦门361008 [2]公安部网络安全保卫局,北京100006

出　　处：《计算机科学》2016年第B12期209-218,共10页Computer Science

摘　　要：信息隐藏技术在智能手机上具有很强的隐蔽性和可操作性，对传统的手机取证技术提出了更高的要求和挑战。自适应多速率语音编码是智能手机中常用的语音编码，通过分析其编码过程和信息隐藏的特性可设计多种特征用于隐写分析。为进一步提高隐写分析的检测效果，均值方差评价系数被用于支持向量机的一对一分类器集合中各分类器的特征选择。依据评价结果，各分类器可挑选不同的特征子集进行训练，训练后的分类器集合不仅可用于检测信息隐藏，还能识别出隐藏所使用的方法类别，较现有方法，其在分类准确性上也有较大提高。Steganography technology issues new and great challenge to traditional mobile forensics technology because of its strong concealment and convenience on smartphone. Adaptive multi-rate （AMR） codec is the most frequentlyused speech codec on smartphone. After analysis of the principle of AMR codec and corresponding steganography approaches, some new feature has been present for steganalysis, Then, mean variance evaluation coefficient was employed for feature selection on differentone vs. one classification of support vector machine （SVM） for steganalysis. With the evaluation result, each classifier can choose different feature subset for training. The trained combinedclassifiers can not only distinguish original and steganographic objects but also figure out what kind of steganographic approach was used on the speech. The experimental result shows that the trained classifierscan achieves much higher accuracy than the state of arts works.

关 键 词：隐写分析 支持向量机 自适应多速率语音编码 一对一分类 

分 类 号：TP391.4[自动化与计算机技术—计算机应用技术]