Markov bidirectional transfer matrix for detecting LSB speech steganography with low embedding rates

Abstract

Steganalysis with low embedding rates is still a challenge in the field of information hiding. Speech signals are typically processed by wavelet packet decomposition, which is capable of depicting the details of signals with high accuracy. A steganography detection algorithm based on the Markov bidirectional transition matrix (MBTM) of the wavelet packet coefficient (WPC) of the second-order derivative-based speech signal is proposed. On basis of the MBTM feature, which can better express the correlation of WPC, a Support Vector Machine (SVM) classifier is trained by a large number of Least Significant Bit (LSB) hidden data with embedding rates of 1%, 3%, 5%, 8%,10%, 30%, 50%, and 80%. LSB matching steganalysis of speech signals with low embedding rates is achieved. The experimental results show that the proposed method has obvious superiorities in steganalysis with low embedding rates compared with the classic method using histogram moment features in the frequency domain (HMIFD) of the second-order derivative-based WPC and the second-order derivative-based Mel-frequency cepstral coefficients (MFCC). Especially when the embedding rate is only 3%, the accuracy rate improves by 17.8%, reaching 68.5%, in comparison with the method using HMIFD features of the second derivative WPC. The detection accuracy improves as the embedding rate increases.