A Detective Method Based On Support Vector Machine For Hiding Information In Voice

The technology that hiding secret messages in the digital media carriers is called steganography. In recent years, the proportion of voice in communication business has a great increase. Its popularity makes the voice became into an excellent digital carrier. However, some covert communications which carries illegal, malicious secret information will threat information security and cause a great damage. In that case, studying analytical methods of voice communication with carrying secret information will be important. It can maintain the communications security, prevent the spread of malicious information, protect the national security in the maximum guarantee and so on. In this paper, the main contributions to the steganalysis algorithms of information hiding in voice communication can be summarized as follow:(1) The algorithms of voice steganography and steganalysis method have been investigateFirstly, depending on different types of digital carriers, steganography and steganalysis research situation on voice is summarized. What is more, LSB matching and QIM steganalysis algorithms are analysed in detail. As well, a steganalysis system framework on voice and the main testing tools is given.(2) A steganalysis method for Least signify-cant bit(LSB) matching steganography based on multi-order Markov featureLSB matching steganography does not change the voice sampling signal values, but will change the irregularity of noise. Based on that, this paper designs a method to get noise sequence based on5/3wavelet de-noising and the difference of voice sampling sequence. Secondly, this paper models the correlation of adjacent sampling noise signals. Its first-order and second-order Markov correlation model are given later. According to the Markov correlation model, multi-order Markov feature vector is constructed. Finally, a detector based on SVM is designed to experiment. The experimental result shows that, the length of10seconds voice with10%embedding rate can reach69.75%recall rate and90.88%precision ratio, the detection accuracy of which can reach more than81.37%.(3) A steganalysis method for quantization indec modulation(QIM) steganography based on feature dimension reductionBy analyzing the statistic of compress voice files, we know that QIM steganography will change the distribution of quantization codeword C. Based on this difference, the weight distribution characteristic of LPC filter is calculated and the important LPC filter set is collected. Thirdly, weight features are fused according to the distance of LPC filter vector and dimension-reduction feature vector for stega-nalysis is constructed. Finally, a detector based on SVM is designed to experiment. The experimental result shows that, when the compress voice length is more than2s, the detective accuracy can reach80%.