Digital Audio Watermarking by Learning in Wavelet Domain

Most of the watermark (WM) decoding schemes use correlation-based methods because of their simplicity. Generally, a decision threshold specified semi-automatically is used at the decoding site. The main problem of the correlation-based decoders is the existence of undesirable correlation between the embedded signal and the host signal that makes the decision threshold specification harder, especially in noisy channels. In this paper, WM decoding is modeled as a pattern recognition problem, thus eliminates the threshold specification problem by learning the embedded data in wavelet domain followed by a nonlinear classification. Furthermore, the encoding performance is improved by perceptual control of Watermark-to-Signal-Ratio (WSR) without disturbing imperceptibility. When the WSR is higher than -30 dB, the decoding and detection performances of the developed system are greater than 99% and 98%, respectively. System false alarm ratios remain less than 2%