Generative adversarial network image watermarking model based on dense block structure
Jianghua Li, Huibing Wang · 2024
Aiming at the situation that the image features extracted for fusing watermark information are not rich and robust enough in the generative adversarial network based image watermarking model, resulting in weak robustness, a generative adversarial network image watermarking model based on the dense block structure is proposed. The reuse of shallow features by using dense block strengthens feature propagation and highlights important features through channel attention to learn more representative and robust feature images for fusing watermarking information; in order to attenuate the image distortion caused by information embedding, different embedding strengths are set for different texture regions of the image. The experimental results show that with the model proposed in this paper, the visual quality of watermarked images is better and has better robustness to common image attacks.