High Payload Adaptive Audio Watermarking based on Cepstral Feature Modification

In this paper, we propose two blind adaptive audio watermarking schemes based on complex cepstrum transform (CCT) domain features. In first scheme (SchemeI), each audio segment is divided into two subsets having approximately same statistical mean value using down-sampling method. Since the human auditory system (HAS) is not much sensitive to the minute change of the wavelet high-frequency components, first level discrete wavelet transform (DWT) detail coefficients subbands of both subsets are used for embedding. Watermark is embedded by changing slightly the difference between the mean values (Mdiff ) of insignificant CCT coefficients of these subsets in order to guarantee minimal perceptual distortion. The offset value by which mean is modified is made adaptive to the local energies of the audio frames in order to increase the audio quality further. In order to enhance the payload capacity, we propose an alternate audio watermarking scheme (Scheme-II) where watermark is embedded by deciding the transition of Mdiff value from one frame to another frame in two successive frames. In contrast to previous works, instead of embedding one bit per frame, Scheme-II can embed three bits per two frames. Thus 33.33% increase in embedding capacity is achieved. As amplitude scaling in time domain does not affect selected insignificant CCT coefficients, strong invariance towards amplitude scaling attacks is also proved theoretically. Experimental results reveal that the proposed watermarking schemes maintain high audio quality and are simultaneously robust to general attacks like MP3 compression, amplitude scaling, filtering, re-sampling, re-quantization etc.