Research On Steganalysis For Compression Domain Of MP3

Informatin hiding is to embed the secret data in media(such as text, image, audio and video etc. ) that is difficult to perceive. The aim of steganalysis mainly detects the existence of secret data, estimate the length of embedded data and even recover it. For the popular compressed format audio (such as MP3, AAC etc.), the secret information in these audios based on compression domain is more difficult to detect and perceive because of the complex and highly compressed characteristic of audio compressed format. Hence, it is more challenge to make research on the steganalysis in compression domain.In this paper, the audio format-MP3 is mainly discussed. The major contribution is as follows:The features about Markov and Neighbor Joint density proposed by Qiao et.al.are modified from inter frame to inter and intra frames, the modified features could be effectively used to detect the secret information by MP3Stego tool and improve the detection rate.A new feature- Errors of Polynomial fitting is proposed and successfully applied to detect the common information hiding in compression domain of MP3. Meanwhile, the feature is also combined with Markov and Neighbor Joint Density to prove the effectiveness and improvement on detection rate.A software system for experiments is also conducted. It accomplishes a series of function including the MP3 decoding, feature extraction, features selection, SVM training and SVM predicting etc. All the experiments are based on VC++6.0 platform and test the effective method proposed in this paper.In conclusion, a new investigation on steganalysis of classic information hiding in compression domain of mp3 based on features extraction from MDCT coefficient is conducted. The experimental results show this effective detection of secret data in compression domain of MP3. More work on the improvement of detection rate and detection on secret data in other compressed-format audio needs to be furtherly done.