• Investigate and articulate the intrinsic relation between matrix pixels and the maximum singular value of SVD. • Adaptively adjust the quantization step based on the image scaling ratio to resist scaling attacks. • Use the rotational invariance of the largest connected domain to calculate the rotation angle. • Address pixel overflow to keep pixel values within the valid range after watermark embedding. With the explosive growth of digital multimedia content, ensuring the security and integrity of media data has become increasingly critical. Nevertheless, robustness against geometric attacks, specifically downscaling and rotation, remains a critical bottleneck for existing watermarking schemes. To address this, a robust watermarking scheme is proposed, which utilizes an adaptive quantization step and maximum connected domain to effectively resist geometric attacks. Specifically, by deeply exploring the intrinsic relationship between the largest singular value and image pixels, the downscaling ratio is accurately calculated, enabling adaptive adjustment of quantization steps to improve resistance against downscaling. Additionally, by exploring the rotation invariance of the maximum connected domain centroid, the rotation angle is calculated without the need for the host image, thereby correcting the rotation attack. The overflow problem after watermark embedding is solved. Experimental results are as follows: 1) The time complexity of embedding and extraction in this scheme is O ( n 2 ); 2) The average PSNR is above 40 dB and the average SSIM is above 0.97; 3) The average values of NC are above 0.95 under different attacks. Overall, the proposed scheme achieves superior performance, particularly against geometric attacks.
Chen et al. (Tue,) studied this question.