A watermark-adaptive and resolution-adaptive image watermarking system using deep learning

摘要

A watermark and resolution adaptive image watermarking system using deep learning that a user can apply to various watermark data and host images of various resolutions without an additional learning process, comprising: performing pre-processing on an input host image preprocessing network for hosts; a preprocessing network for watermarks that performs preprocessing on the watermarks; a matching unit for matching the preprocessed host image with the preprocessed watermark; an embedding network that receives a matched result and outputs a watermarked image in which a watermark is embedded in a host image; an attack simulator for simulating an attack on the watermarking image and outputting an attacked watermarking image (hereinafter referred to as an attacked image); an extraction network that receives the attacked image and extracts a watermark; and an optimizer for optimizing parameters of each network using a loss function, wherein the network provides a configuration including at least one convolution block and an activation function. With the above system, in the preprocessing network, the invisibility of the host image can be secured by maintaining the resolution of the host image and increasing the resolution of the watermark image, and the invisibility and robustness can be adjusted by adjusting the intensity of the watermark information. can control the complementary relationship of