Robust watermarking against arbitrary scaling and cropping attacks

Digital watermarking technology has wide application prospects due to its excellent performance in copyright protection and traceability. However, there are still some gaps in robustness when it comes to arbitrary scaling and cropping attacks scenarios. In this paper, we propose a robust watermarking method against arbitrary scaling and cropping attacks, which maintains a high image visual quality and can extract the watermark from arbitrary position and size local region of an image. Specifically, the padding operation is used to process the arbitrary resolution image, ensuring that they can be fed into the algorithm. Then the watermark is embedded using the DFT-based polar coordinate mapping method, and finally the resolution is restored to obtain the watermarked image. At the watermark extraction, a machine learning-based scaling parameter estimation method is proposed, which can accurately predict the scaling parameters. At the same time, the padding operation is applied to the attacked image, which can maintain the watermark state in the watermarked image as much as possible and is more conducive to the extraction of the watermark. The experimental results demonstrate the effectiveness and adaptability of the proposed watermarking method, while maintaining the performance of state-of-the-art against arbitrary scaling and cropping attacks.