Research On Digital Image Watermarking Algorithm Based On Convolutional Neural Network

In recent years,deep learning and convolutional neural networks have become the focus of computer vision research.Digital watermarking researchers have introduced convolutional neural networks into the field of digital watermarking and achieved good results.Convolutional neural networks have incomparable advantages in extracting robust features of carrier images and balancing the contradiction between robustness and invisibility of watermarking algorithms.Two watermarking algorithms based on convolutional neural network are proposed in this paper,and the experimental results are outstanding.In addition,a watermark implementation system is designed based on two algorithms.The specific research content of this paper is as follows:(1)A zero-watermarking algorithm based on style transfer is proposed.Firstly,a new image with a certain style of the carrier image is generated by using the VGG network from the carrier image and the time-stamped watermark image.Then,Arnold transform is used to encrypt the generated stylized image and submit it to the third party for registration as a zero-watermark image.Finally,the RCNN neural network is designed to extract the watermark image and verify the copyright of the carrier image.Experimental results show that the proposed algorithm is more secure and robust than the existing zero-watermarking algorithm.(2)An embedded image watermarking algorithm based on U-Net network is proposed.The network framework of the algorithm consists of the following four modules:① Watermark image preprocessing module,which preprocesses the watermark image and makes it easy to be embedded in the carrier image;②Embed the watermark module,and embed the pre-processed watermark image into the carrier image;③Attack module,which simulates various attacks that the carrier image may suffer in real applications to improve the robustness of the algorithm;④Extract watermark module,extract watermark information from watermarking image.The experimental results show that the watermark image generated by this model has good fidelity and can be successfully extracted under several common attacks.(3)Based on the two algorithms,a digital image watermarking system is designed.The system includes three functions:zero watermark deneration and copyright verification,watermark embedding and extraction by embedded watermark algorithm,and image processing.