Enhanced Fourier–Mellin domain watermarking for social networking platforms

Robustness to common hybrid distortions is a crucial requirement for effective watermarking, particularly on social networking platforms (SNPs). Images on SNPs undergo complex attacks initiated by both platforms and users, involving diverse distortion operations. However, there are few image watermarking schemes designed to handle such hybrid attacks effectively. Existing schemes, especially those based on the Fourier-Mellin domain, often struggle due to their susceptibility to single attacks. For instance, the ring watermark structure in the frequency domain is prone to distortion, leading to difficulties in mapping watermark information and causing streak diffraction phenomena in the image. Additionally, these schemes lack robustness against large-size image downsampling and image flipping attacks on SNPs. To address these limitations, this paper introduces an enhanced robust watermarking framework tailored for SNPs. The framework comprises three key modules: a module to stabilize the ring watermark structure, an adaptive embedding strength and range module, and a sliding window and flip state detection module. These modules, coupled with log-polar mapping (LPM) in the Fourier-Mellin domain, effectively mitigate the lack of robustness to specific attacks, resulting in comprehensive robustness for the entire framework. Numerous experiments demonstrate that our proposed scheme outperforms other state-of-the-art (SOTA) works in handling hybrid distortions on SNPs.