Screen‐shooting resistant robust document watermarking in the Discrete Fourier Transform domain

Abstract

AbstractMetaverse's augmented reality (AR) function allows virtual information to be seamlessly superimposed onto real scenes through the camera of a head‐mounted device. However, this raises concerns about privacy protection and copyright authentication when transmitting cross‐media information. Additionally, there is a risk of secret information leakage due to screen candid shooting in the real world. Ensuring information security and copyright authentication in case of unauthorized screen capturing is crucial. To prevent information loss and interference from cross‐media transfer between screens and cameras, we implement digital watermarking for copyright protection. We have proposed an innovative framework for automatic document watermarking that can resist screen‐shooting. Our approach involves embedding a ring watermark in the document underlay. On the extraction side, the watermark extraction process is divided into three key steps: automatic location, automatic correction, and automatic extraction. First, the document image is located in the covert photography. Then, perspective correction is performed based on the text line features of the document. Finally, the watermark information is extracted by combining the ring watermark features. Our method is capable of automatically extracting watermarks from covert photography while considering aspects such as concealment, robustness, and visual quality. The watermark is embedded in the document underlay, which ensures good visual quality and does not affect the normal reading and editing. We also propose various embedding strength schemes that can adapt to different usage scenarios, providing resistance to screen‐shooting or screenshot attacks, as well as various noise attacks. Through extensive experiments, we have demonstrated the feasibility of the proposed automated framework and the robustness of the watermarking algorithm, as well as the superiority and broad application prospects of our method.