Multi-layer architecture for efficient steganalysis of Undermp3cover in multi-encoder scenario

Mp3 is a very popular audio format and hence it can be a good host for carrying hidden messages. In fact, different steganography methods have been proposed for mp3. But, only steganalysis of mp3stego has been conducted in the literature. In this paper we mention some limitations of mp3stego and argue that Undermp3cover (Ump3c) does not have them. Then, steganalysis of Ump3c is conducted based on measuring mutual information between global gain and other fields of mp3 bit stream. Furthermore, we show that different mp3 encoders have quite different performances and therefore we propose a novel multi-layer architecture for its steganalysis. In this manner, the first layer detects the encoder and the second layer performs the steganalysis job. One of advantages of this architecture is that feature extraction and feature selection can be optimized per encoder. We show this architecture outperforms the conventional single-layer methods. Comparing results of the proposed method with other works show an improvement of 20.4% in accuracy of multi-encoder scenario.