Research On Mechanism And Key Technologies Of Audio Steganalysis

Steganography is the art and science of hiding secret information into open digital media in a certain way, in order to hide the existence of secret information and realize covert communication. To the contrary, steganalysis is an attack against the steganography, in order to reveal the existence of secret information and even destroy the covert communication. Technology of steganalysis has important academic values and applications prospects, it helps to prevent the illegal application of steganography and guide the design of secure steganographic methods. At present, the majority of information hiding research takes image as a carrier, and steganalysis methods of audio are relatively scarce, moreover, methods of steganography and steganalysis which use Advanced Audio Coding (AAC) as the carrier are even rarely, research in this field is practical and challenging. Therefore, we are dedicated to research of audio steganalysis in this paper, using uncompressed and compressed audios as carriers, and try to explore practical methods of audio steganalysis from theory to practice.The main works and innovations in this paper are as follows:1. An audio steganalysis method based on wavelet packet and adaptive predictor is proposed, which is mainly used to detect the audio steganography method based on additive noise. The experiment results verify that, for the commonly used steganography methods of additive noise, even under low embedding strength or low embedding rate, high classification accuracy can be achieved.2. An audio steganalysis method to echo hiding based on statistical features of power cepstrums is proposed. This steganalysis method can not only detect the basic single echo kernel, but also apply to the improved echo kernels. Even under low attenuation coefficients, high classification accuracy can be achieved, and no matter how long the embedded segments of audio signals, this algorithm can also achieve high classification accuracy.3. A universal method of audio steganalysis based on features combination is presented, firstly a method of features extraction based on the short-time Fourier transform (STFT) is proposed. To further enhance the universality of the steganalysis, we combine the features based on STFT with the features based on audio quality metrics and the features based on linear prediction, and select the features. This steganalysis method is more universal, and it merges the advantages of different kinds of features, making the overall detection accuracy enhanced.4. Two kinds of steganographic methods taking AAC as carriers are proposed. They are steganographic method based on little data region of quantized MDCT coefficients and method based on escape sequences of Huffman coding, both methods embed secret information into the quantized MDCT coefficients to avoid the loss of secret data during quantization, and so the computational complexity is reduced. Performance tests of the AAC steganographic methods reveal that, both methods can obtain a high hidden data capacity, furthermore, they have good imperceptibility and can resist the steganalysis to some extent.5. A steganalysis method based on quantized coefficients of AAC is proposed. According to changes of quantized coefficients histograms caused by steganalysis, we fit the histograms with the generalized Gaussian distribution (GGD) model, and extract GGD parameters of quantized coefficients as features. In addition, according to effects of steganalysis on the quantized coefficients of inter-frames, some data of the Markov transition matrix are extracted, and the support vector machine (SVM) is implemented as a classifier. Experimental results show that the detection accuracy to the steganographic method of spread spectrum modulation on the quantized coefficients is high.