Abstract: Research of image watermarking algorithms is an important issue in the field of multimedia technology. By comparing and combing the current two leading watermarking algorithms, an adaptive multiplicative watermarking algorithm is proposed. Motivated by Barni’s “pixel-wise masking” model and redundant wavelet domain masking effect modeling approach, the nonsubsampled Contourlet transform domain masking effect model is established. With generalized Gaussian distribution model and Cauchy distribution model to describe the statistical properties of the nonsubsampled Contourlet transform coefficients, the watermark detection problem is then addressed as a composite hypothesis testing. Theoretical analysis leads to two locally optimal nonlinear blind detectors for adaptive multiplicative watermarks as well as adaptive methods for determining detection thresholds. The experiment results show that the nonsubsampled Contourlet transform domain masking model achieves good visual invisibility, and watermarks can be accurately detected without using the original images and the adaptive embedding strength factors. In addition, Cauchy distribution based blind detector is found superior to the blind detector based on the generalized Gaussian distribution both in terms of detection performance and detection efficiency.

Key words: nonsubsampled Contourlet transform, generalized Gaussian distribution, symmetric α-stable distribution, Cauchy distribution, adaptive multiplicative watermarks, locally optimal nonlinear detector, blind detector