Fusing PDTDFB magnitude and relative phase modeling for geometrical correction-based image watermarking

Abstract

Robustness, which refers to the ability that watermark signal survives various attacks, like additive noise, compression, rotation, etc., has played extremely important role in the multiple applications of digital watermarking. As one of the most difficult kinds of signal processing operations for a digital watermark to survive, geometric distortions have become a central problem in image watermarking research. Therefore, developing a greatly robust digital image watermarking approach, which can withstand geometrical distortions, remains a quite challenging work. In this paper, a new geometrical correction-based image watermarking approach using PDTDFB magnitude and relative phase modeling is proposed. This approach consists of digital watermark embedding, geometric distortions correction, and digital watermark extraction. In the watermark embedding process, PDTDFB (Pyramidal dual-tree directional filter bank) decomposition is performed on the original host image, followed by the low-pass subband partitioning. Watermark bit is inserted into low-pass subband block by modifying (quantization index modulation, QIM) the set of low-pass PDTDFB coefficients. In the geometric correction, the PDTDFB magnitude and relative phase are modeled by using Weibull distribution and Vonn distribution, respectively. Utilizing the compact statistical model parameters, the LS-SVR (Least squares support vector regression) correction is performed to estimate the geometrical distortions parameters. After LS-SVR geometrical correction, the watermark bits are extracted from the watermarked low-pass subband by employing the inverse QIM. Experimental results confirm that, under various well-known practical attacks, including common signal processing operations and geometrical distortions, the proposed approach performs well compared to conventional image watermarking methods.