Self-coding color image robust watermark processing method based on visual perception

Abstract

The invention discloses a self-encoding color image robust watermark processing method based on visual perception, which comprises the following steps of: training a watermark embedding module in an AGWNet by using a carrier image training set and an original watermark image training set to obtain structure loss and visual perception loss of a watermark-containing image and a carrier image; inputting the watermark-containing image into a noise attack module in the AGWNet to generate an attacked watermark image; inputting the attacked watermark image into a watermark extraction module in the AGWNet for training to obtain cross entropy loss of the extracted watermark and the original watermark image; calculating the overall loss according to the structure loss, the visual perception loss and the cross entropy loss, and obtaining a trained AGWNet when the overall loss is smaller than a threshold value in the iteration of the overall loss; and embedding or extracting the watermark by using the trained AGWNet. According to the method, the watermark robustness can be improved while the quality of the watermark-containing image is ensured.

Description

technical field [0001] The invention relates to a robust watermark processing method for self-encoding color images based on visual perception, and belongs to the technical field of digital image watermarking. Background technique [0002] The quality of the watermarked image, that is, the invisibility of the watermark, and the accuracy of extracting the watermark under attack, that is, the robustness of the watermark to attacks, are two important indicators for evaluating the quality of the watermarking method. Since 2017, many robust watermarking methods for color images based on deep learning have emerged. In 2018, Zhu et al. proposed to first introduce the autoencoder structure into the field of robust color image watermarking, creatively regard the watermark embedding and extraction process as the reconstruction of image information, and propose an end-to-end training method, which cleverly allows the neural network to automatically Adaptively find a balance of invisib...

AI Technical Summary

Problems solved by technology

However, this method only relies on the feature map generated by another network as a guide for embedding, and does not fully consider the robustness of each pixel to the encoded information.

[0004] Zhang et al. proposed to use the gradient of the carrier reconstruction loss to generate the attention mask, and filter the suitable embedding position in advance according to the generated mask. Therefore, the watermark embedding capacity of this scheme is limited and it is difficult to deal with two-dimensional meaningful watermark information. In the scheme The maximum embedding capacity shown only supports 90 bits. At the same time, this method is slightly lacking in network interpretability

[0005] Based on the above analysis, the existing end-to-end training robust watermarking algorithm is difficult to give competitive experimental results in terms of invisibility and robustness, the reason is that the following two problems are not well solved: (1 ) does not add prior knowledge guidance in the watermark embedding stage, assigns different embedding strengths to different positions of the carrier, and the watermark embedding method lacks interpretability; (2) does not solve the problem of long-distance feature loss when reconstructing images in the embedding and extraction stages, It is difficult for the network to achieve an efficient balance when finding the optimal point of dual tasks

Images