Zero-watermark of audio based on multi-scale entropy vectors

The properties of average short-term entropy of audio sequence are discussed in this paper. It is discovered that the changes of average short-term entropy on different scales depict the audio's statistical features well, which can be used to generate zero-watermark of an audio clip. A method of calculating normalized entropy vectors is presented in this paper. To generate the zero-watermark of an audio sample sequence, the multi-scale entropy vectors are calculated on each frame of the audio. The difference between the mean of all the components of each multi-scale entropy vector and the given threshold determines the category of the frame, which leads to a binary sequence representing statistical features of the audio. After selecting randomly some bits from the sequence and doing XOR operation with the binary watermark image specified, a binary image used to decrypt can be obtained. The experimental results show that the zero-watermark algorithm proposed in this paper has more remarkable resistance to various kinds of attack than common algorithms.