Research on Watermark Embedding of Fax Channel Images Based on U-Net Network

This paper introduces a novel watermarking algorithm based on the U-Net Network to manage and trace the source of fax document leaks. The goal is to enable the embedding and extraction of watermark information in fax text images. To adapt the U-Net structure to the unique characteristics of text images transmitted through fax channels and cross-media channels, we have enhanced it. During joint training of the encoder and decoder, we have incorporated several functions, including the introduction of Gaussian noise, JPEG compression, motion and bokeh blur, and shot correction. These enhancements aim to improve both the visual quality of the images and the accuracy and robustness of watermark extraction. The results demonstrate that this approach achieves an image similarity of 0.977 and an image signal-to-noise ratio of 34.96. These outcomes surpass those of three other image watermarking algorithms in terms of image visual quality, watermark embedding capacity, and robustness.