Research On Detection Technology Of Low Embedding Rate LSB Audio Steganography

Current detection technology for LSB audio steganography basically take advantage of the statistical characteristics of the carrier,and generally have good detection performance when the embedding rate is high.However,when the amount of embedded information is small,the current detection algorithm performs poorly.In this study,we design a new detection method to achieve reliable detection of LSB audio steganography with low embedding rate.The previous work that tried to design a classifier that directly modeled the specific content of audio was not good.The possible reason is that when the embedding rate is low,the difference between quantified vectors before and after steganography is weakened.Since the least significant bit of the carrier signal has a noise-like characteristic,and LSB audio steganography is to replace the least significant bit of the sample point with the message bit,LSB audio steganography is regarded as a kind of artificially added noise.The method based on local correlation and the method based on wavelet de-noising are used to estimate the noise sequence.Markov chain is widely used in the study of digital audio signals.It describes the correlation of adjacent sampling points in the form of conditional probability.Therefore,Markov model is used to extract Markov features from the estimated noise sequence and construct the first-order and second-order conditional probability transfer matrix.The probability matrix is transformed into a one-dimensional vector,which is used as the input vector of classifier training.This detection technology belongs to the detection technology based on feature change and is the current mainstream detection technology.In order to verify the detection performance of this detection method,a comparative experiment is conducted with the detection method based on feature changes and a new method of constructing classifier based on SPAM feature.Experiments show that this method can effectively improve the recognition accuracy of LSB audio steganography with low embedding rate.