Wavelet-FCWAN: Fast and Covert Watermarking Attack Network in Wavelet Domain

The protection of digital image content has become an important topic of scientific research with the continuous development and updates of Internet and multimedia technology. In the past few years, various watermarking algorithms with good robustness and imperceptibility have been proposed, but the development of watermarking attack techniques has stagnated. In this paper, we have attempted to use deep learning to develop a new watermarking attack scheme and present the Fast and Covert Watermarking Attack Network in Wavelet Domain (Wavelet-FCWAN). The watermarking attack scheme employs noise filling as a preprocessing step for the watermarked image, performs wavelet transform operation on the preprocessed watermarked image, and inputs the wavelet transformed sub-image into Wavelet-FCWAN along with the noise level map in parallel. The network can be trained quickly and produces a better watermarking attack effect while ensuring the visual quality and detail retention of the attacked image. The experiments show that Wavelet-FCWAN demonstrates the superiority of its watermarking attack effect by comparing different attack strategies, and it can produce varying degrees of attack effects on various image watermarking algorithms with high levels of universality and imperceptibility.