SAW: A Robust Watermarking based on Visual Structure Measurement and Spatial-Channel Attention
Chuanning Wang, Dongming Zhang, Jing Zhang, Xueqiang Zou · 2023
Deep image watermarking still faces the image quality degeneration of the watermarked and recovered watermark. A robust watermarking method, SAW is proposed to embed watermarks in an imperceivable way and improve the recovered watermark. It uses visual structural measurement, which provides better image quality supervision, to guide the watermark embedding process. In the watermark recovery, SAW introduces the spatial-channel attention module, which focuses on watermark extraction. Compared with UDH, the experimental results show that the method remains robust and obtains noticeable image quality improvement of both the watermarked image and the watermark.