Research On Audio Steganography And Steganalysis

Steganography is an art and science of hiding information into digital files, such as audio, image and video, to achieve covert communication, while steganalysis is the attack against steganography, detecting the covert communication. Both steganography and steganalysis are of great importance in intelligence and national security, while their prominent position in the field of information security pushes researchers to conduct in-depth exploration. Therefore, I dedicate myself to research on audio steganalysis and steganography in this paper.The main work and innovations of this paper are as follows:(1) A steganographic algorithm for MIDI files is proposed. Secret messages are embedded into both time parameters and velocity parameters to increase embedding capacity, while specific position is adaptively selected to reduce the embedding impact on audio quality. Steganographic encoding technique is applied to reduce the changes to statistical characteristics of the raw data, enhancing the algorithm's capability to resist steganalysis.(2) A steganographic algorithm for AAC files is proposed. Prior to embedding, secret messages are encrypted with 3DES technique. Then the encrypted data are embedded into AAC bit stream. SHA-1 algorithm is employed to select AAC frames for embedding. Data embedding takes place at the heart of the AAC encoder, and each bit is encoded as the parity bit of the number of bits used for Huffman and differential coding of an AAC frame.(3) A blind audio steganalysis based on feature selection is proposed. Four kinds of features are extracted respectively, including Mel frequency cepstrum coefficients, second-order derivative spectrum characteristics, audio quality measure and linear prediction error. To avoid dimensionality curse and reduce computational complexity, F-score strategy is employed to obtain the optimal feature vector. Experiment results show that the proposed steganalysis algorithm could effectively detect four steganography, including DSSS, QIM, ECHO and LSB, which makes the proposed methodology capable of being regarded as a blind steganalysis, and especially useful when the steganalyzers are without the knowledge of the steganographic scheme employed in data embedding.