WMHMT-IWD: Weibull Mixtures-HMT Based Image Watermark Detector

Balancing imperceptibility, data payload and robustness is a crucial and unsolved issue in the area of digital watermarking. It has been found that the method based on statistical modeling can alleviate the contradiction among them. On the basis of this, we suggest a novel image watermark detector based on Weibull Mixtures-HMT model in this paper. The Otsu-Canny edge detection method is used to select the high entropy blocks in the target subband obtained by non-subsampled shearlet transform (NSST), and fast polar complex exponential transform (FPCET) is computed on the target blocks to obtain the NSST–FPCET magnitude domain. The digital watermark is inserted into the NSST–FPCET magnitude domain by a multiplicative method. In the statistical modeling step, the NSST–FPCET magnitudes are modeled by Weibull Mixtures-HMT model, where Weibull Mixture model describes the distribution characteristic of the NSST–FPCET magnitudes while the HMT model describes the dependency between the NSST–FPCET magnitudes. Parameters of the proposed model are estimated by an efficient variance reduced stochastic expectation maximization method. Using the locally most powerful test, we finally develop an image watermark detector based on Weibull Mixtures-HMT model. Massive experiments demonstrate that the designed scheme has an impressive performance in detecting the presence of the watermark, and it can achieve a better balance among the data payload, imperceptibility and robustness.