Research Of Audio Steganalysis Method

Steganalysis is a countermeasure technique for steganography, its target is to find out the suspicious carriers who take along secret message, further intercept, detect and recover the secret message. It has performed a significant role in communication, internet and information security. Our research is focused on the audio steganalysis based on the theories of feature mining, pattern recognition and machine learning. The main contributions of this dissertation are summarized as follows:1. Based on the audio amplitude distribution change caused by secret message embedding in different bit-planes, a method of secret message length estimation is presented. This method can not only discriminate the existence of secret message, but also recognize the embedding bit-plane and estimate the amount of secret message. The proposed method performs high detection rate on both the sequential and randomly scattered embedding methods.2. By the analysis of the maximal cumulative cross correlation between PN sequence and test audio under various PN sequence length, an effective detector for spread-spectrum steganography is designed. Then, employing Genetic Algorithm (GA) to solve the optimization problem of two variables, which are the length and composition of the estimated sequence, the Spread-Spectrum (SS) sequence used by transmitter can be estimated. Finally, the sequentialy embedded secret message can be blindly recovered.3. After in-depth exploring the changing rule of the Sliding Window Cepstrum (SWC), we extract the sensitive statistical feature named Cepstrum Peak Location Aggregation Rate (CPLAR) in cepstrum domain to realize the echo detecting system. Then, we proposed an active steganalysis approach for extracting hidden messages from the stego audio by estimation of several key parameters including delay offsets, segment length and embedding positions.4. A universal audio steganalysis method based on PCA (Principle Component Analysis) and SVM (Support Vectors Machine) is proposed in this paper. The optimum feature combination chosen by PCA technique is employed to train the SVM, and then the obtained SVM classifier is used to detect the existence of secret message in test audios. The experimental results have demonstrated the validity of the proposed method.The research of universal audio steganalysis method is the hot issue in the future. How to find out more distinguishable and stable features by exploring the basical model of information hiding, improve the detecting and recovering performance, and enhance the self-learning ability of the proposed analysis system is our further work.