FR-Watermarking: A Fusion Framework for Face-Based Digital Watermarking

FR-Watermarking: A Fusion Framework for Face-Based Digital Watermarking

연구 분야: Verification

학회: Chinese Conference on Pattern Recognition and Computer Vision (PRCV)

The expansion of We Media has significantly enhanced individual content sharing, while also intensifying the challenges of copyright protection. Various biometric watermarking techniques, including traditional and Deep Neural Networks (DNN)-based schemes, have been proposed to assert ownership over digital properties. In this paper, we introduce a Face Recognition-Watermarking (FR-watermarking) framework, which integrates Facenet model with a traditional watermarking block. Here, the watermark block employs Discrete Wavelet Transform and Singular Value Decomposition (DWT-SVD) alongside hybrid encryption to enhance security and robustness against attacks. Additionally, joint training is implemented to further improve robustness. Evaluation on the Labeled Faces in the Wild (LFW) dataset demonstrates that our framework achieves promising results in terms of invisibility and robustness. Specifically, it achieves an average Peak Signal-to-Noise Ratio (PSNR) exceeding 38 dB and an average Structural Similarity Index (SSIM) over 0.99. Moreover, the verification performance of the extracted watermarks shows significant improvements in robustness, with accuracy (ACC) and validation rate (VAR) scores of approximately 0.98 and 0.86 when subjected to various distortions, surpassing that of the baseline Facenet model.