Energy and Entropy Based Features for WAV Audio Steganalysis

Digital steganalysis techniques attempt to detect hidden information in digital media. The rising interest in steganalysis is attributed to the growing number of steganography algorithms and the threats they represent. This article presents a combined maximum entropy energy approach for audio steganalysis. First, the audio signal is divided into four energy-based regions: noise, low, medium and high; then entropy is computed from each region. Finally, a support vector machine is applied to the collected features for discovering the hidden data in audio signals. Active speech level algorithm is used to capture energy fluctuation in audio streams. The paper shows that the extracted features from separate energy-based regions of the signals have significantly improved detection accuracy of hidden messages. Our work includes comparisons with current state-of-the-art audio steganalysis techniques. The experimental results show that our method achieves up to 96.7% correct for an embedding rate of 25% or above of stegosignals produced by S-tools4, Steghide and Hide4PGP while using a much smaller feature set.