Audio Steganalysis Based On Statistical Analysis

The research of steganalysis is significant to prevent disclosure of confidential information, monitor illegal information, protect national security and maintain social stability. Designing high detecting accuracy and wide applibality feature vectors is the core of steganalysis. Intensive research on statistical analysis of frequency domain in the literature is lack, so that the steanography which insert the secret message in frenquency domain can not be detected. In this paper an attempt has been made to analysis statistical distribution of audio DCT coefficients and the effect of steganography on the statistical nature. Based on the study, two novel steganalysis is designed. In order to improve the detection accuracy and enlarge the range of detected steganography, a universal steganalysis based on fusion technology is proposed.The main work and innovation are as follows:1) Aiming at the scarcity of statistical distribution of audio DCT coefficients, the statistical distribution of audio DCT coefficients is deliberated. A mixed model which is a linearly weighted average of generalized Gaussian distribution and alpha stable distribution is proposed. The experimental results show that the mixed model is near to the true distribution of DCT coefficients than generalized Gaussian distribution and alpha stable distribution.2) An audio steganalysis based on statistical moments of DCT histogram is proposed. The statistical moments of the histogram in DCT domain and the statistical moments of the histogram of the wavelet coefficients of every level in frequency domain are calculated as the features. Experimental results show that the proposed technique gain high detection accuracy.3) On the Assumption that steganography change the statistics of audio, a steganalysis based on the statistics of the parameters of statistical distribution is proposed. Experimental results show that compared to similar algorithms, detection rate has improved.4) For the defect of practicability of existing steganalysis, a universal steganalysis based on fusion theory is proposed. This method can inherit the advantages of exiting steganalysis algorithm, and then improve the detection performance.