OWR: Optimizing Watermark Robustness for Screen Recording
Zixuan Hu, Kun Hu, Zizhuo Wang, Ranran Pan, Xingjun Wang · 2024
The prevalence of mobile recording devices is on the rise, posing greater challenges for existing watermarking solutions in copyright authentication of images that have been illegally captured. However, current approaches have not accurately simulated real-world noise, thus obstructing the balance between robustness and imperceptibility. To address this problem, we propose a novel approach to screen noise simulation with the assistance of image pairs and pre-trained optical flow models. Specifically, we decompose the pixel distortion of screen recording into intensity distortion and positional distortion. Due to the simplicity and efficiency of the screen noise decomposition scheme, the noise simulated by our method is much closer to the real scene, which improves the robustness of the scheme and the visual quality of the images. Extensive experimental results demonstrate that our scheme significantly outperforms existing SOTA methods.