CAJ | 학술논문

提出了一种 G．729A 自适应码本分组基音调制信息隐藏的检测算法．对语音码流的分析发现，通过基音预测进行信息隐藏将改变相邻语音帧中基音周期估计值的共生特性．通过量化这种共生特性，并经过 PCA（Principal Component Analysis，主成分分析）降维获得对隐写检测敏感的特征向量．最后基于特征向量和 SVM（Support Vector Ma-chine，支持向量机）构建隐写检测器．对不同语音样本数据集的检测表明，当语音长度在2s 及以上时，检测正确率均超过96％．此检测算法是一种有效的压缩域信息隐藏检测方法．
제출료일충 G．729A 자괄응마본분조기음조제신식은장적검측산법．대어음마류적분석발현，통과기음예측진행신식은장장개변상린어음정중기음주기고계치적공생특성．통과양화저충공생특성，병경과 PCA（Principal Component Analysis，주성분분석）강유획득대은사검측민감적특정향량．최후기우특정향량화 SVM（Support Vector Ma-chine，지지향량궤）구건은사검측기．대불동어음양본수거집적검측표명，당어음장도재2s 급이상시，검측정학솔균초과96％．차검측산법시일충유효적압축역신식은장검측방법．
A detection algorithm of pitch modulation information hiding in G.729A low bit-rate speech codec is proposed. The analysis of speech stream showed that pith modulation information hiding would change the pitch symbiotic characteristics of adjacent speech frames.We designed a model to quantify these pitch symbiotic characteristics for steganalysis.However,the dimen-sion of quantitative feature vector of pitch symbiotic characteristics was too high,so PCA (Principal Component Analysis)was em-ployed to reduce the dimension of the feature vector.Finally,we built a pitch modulation information hiding detector based on the dimension reduced feature vector and SVM(Support Vector Machine)classifier.Experiments on different speech datasets show that the proposed steganalysis algorithm is very effective:the accuracy is more than 96% when speech length equals to or is greater than 2s.So this paper gives an effective method for compression domain information hiding detection.