Steganalysis Method For G.723.1 Speech Based On Quantization Index Modulation

Information hiding technology has been developing rapidly since it was put forward.Steganography,as an important branch of information hiding,is playing a more and more important role in network communication.Steganography can be utilized by criminals,which will cause serious security risks.Steganalysis can block illegal information transmission by detecting whether the carrier contains secret information,so as to resist the illegal use of steganography and guarantee communication security.G.723.1 speech codec is used as the carrier to detect steganography based on quantization index modulation(QIM).In vector quantization process,the index values' distribution changes after QIM-based steganography,the changes include the probability of index occurrence and the mutual influence of adjacent index values.The distribution probability matrix and transition probability matrix are conducted to quantify the distribution characteristics of index values as feature vectors,and the principal component analysis is used to reduce the feature vectors' dimension,the feature vectors with lower dimension are still sensitive to QIM,which is proved by experiments.The extracted feature vectors are input to support vector machine classifier,and experiments on different types of speech samples are carried out,a large number of samples are used to train the classifier to detect QIM-based steganography.The experiments are aimed at the accuracy and reliability of the proposed method.For the two kinds of feature vectors,TPR is higher than 90% and FPR is less than 8%,which proves the proposed method has high accuracy;the overall detection rate is above 90%,which proves the proposed method is reliable.Experiments on speech samples with different embedding rate and different length are conducted.For both feature vectors,when the speech length is 3 seconds or more,TPR exceeds 85%,and the detection about distribution probability matrix is more sensitive to speech length;TPR becomes higher with the increase of embedding rate.