Detection of MP3Stego exploiting recompression calibration-based feature

MP3Stego is a typical steganographic tool for MP3 audio. Once the cover audio is unavailable, it is hard to distinguish between background noise and steganographic distortion. In this work, the MP3Stego algorithm has been analyzed from a warden’s perspective. It is observed that the number of bits in the bit reservoir will be disturbed when the secret message is embedded. In addition, a reliable estimation of cover audio is obtained by the proposed recompression calibration. The calibrated features are classified with support vector machine technique. Experimental results show that the proposed scheme is effective and gets good performance, especially when the embedding rate is not less than 0.01 %. The results also shows that the proposed scheme can achieve lower false positive rate comparing to the existing algorithms.