Mp3 is a very popular audio format and hence it can be a good host forcarrying hidden messages. In fact, different steganography methods have beenproposed for mp3. But, only steganalysis of mp3stego has been conducted in theliterature. In this paper we mention some limitations of mp3stego and arguethat Undermp3cover (Ump3c) does not have them. Then, steganalysis of Ump3c isconducted based on measuring mutual information between global gain and otherfields of mp3 bit stream. Furthermore, we show that different mp3 encoders havequite different performances and therefore we propose a novel multi-layerarchitecture for its steganalysis. In this manner, the first layer detects theencoder and the second layer performs the steganalysis job. One of advantagesof this architecture is that feature extraction and feature selection can beoptimized per encoder. We show this architecture outperforms the conventionalsingle-layer methods. Comparing results of the proposed method with other worksshow an improvement of 20.4% in accuracy of multi-encoder scenario.
展开▼
