Power Spectrum Compliant QIM Watermarking for Autoregressive Host Signals

The autoregressive (AR) model is widely used in modeling image, speech and EEG signals. Using this model as the model for the host signal, we have devised a watermarking algorithm which is compliant with the power spectrum condition. This is achieved by embedding the quantization watermark in the residual signal of the AR model, both for dither modulation (DM) watermarking and spread-transform dither modulation (STDM) watermarking. This paper also analyzes the decoding performance. An analytic result is obtained, which describes the relationship between the decoding error rate and the signal to noise ratio, model parameters and the length of the vector. This analysis result is verified through numerical experiments. Using this analysis result, a designer of the watermarking system can determine the design parameters based on the specification of the given system performance index.