Highly applicable and imperceptible watermark attack network

Chunpeng Wang, Pengfei Tian, Zhiqiu Xia, Qi Li, Jian Li, Ziqi Wei, Ting Luo, Bin Ma

Abstract: Highlights•CNN and U-Net merge to create an imperceptible watermark attack network.•We expand the watermark attack network’s receptive field with a contextual module and improve convergence with residual network enhancement.•Incorporating noise-based attacks, we evaluate denoising attack effectiveness through varying noise levels.•Our algorithm outperforms in attacking spatial, transform domain, and other types of watermarking algorithms.