Research On Robust Digital Image Watermarking With High Visual Quality

With the development of the Internet,a large number of hardware and software applications have created the need for image reproduction and copyright protection.A possible way to address this need is to use digital watermarking.The idea of digital image watermarking is to hide the information in the image,so that human eyes can not perceive the modification of the image,but the computer can read the hidden information from the image.After more than 20 years of development in digital watermarking technology,people initially focused on how to design a robust digital watermarking method.Now more and more application scenarios need to pay attention to the quality of the carrier image after the watermark is embedded.The trade-off in digital watermarking has been paid more and more attention.Robust digital watermarking algorithms have many application scenarios.The development of deep learning technology has promoted the advancement of various fields of computer vision.In the field of digital watermarking,deep learning technology has also attracted more and more attention.The first part of this paper studies the traditional robust digital watermarking technology,and proposes a robust digital image watermarking algorithm based on DCT.In this algorithm,each watermark bit is related to multiple image blocks,even if part of the image block is lost or modified,it will not affect the extraction of watermark information.Convolutional error correction codes and CRC are used to improve the success rate of watermark extraction,and the capacity of watermark embedding is improved by grouping image blocks.By classifying image blocks,all image blocks are divided into high-complexity blocks,medium-complexity blocks and low-complexity blocks.We make each type of block be modified to different degrees during the watermark embedding process to improve the visual quality of the watermarked image.At the same time,for the coefficients of different frequencies of the DCT coefficients,different embedding strengths are used to weigh the visual quality of the watermarked image and the robustness of the watermarking algorithm.The second part of this paper designs an adaptive watermarking method based on deep learning.Most watermarking algorithms based on deep learning need to train a separate model for different embedding capacities and images of different sizes,which is very troublesome in practical application scenarios.Moreover,using the previous model to embed the watermark,the smooth area and texture area of the image will be greatly changed.However,the change in the smooth area is easily perceivable by the human eye.The adaptive watermarking method proposed in this paper adopts a fully convolutional network structure,and uses the texture information of the carrier image for watermark information of different lengths to rearrange the watermark information into a watermark binary matrix of the same size as the carrier image.And embedding the template that marks the position of the watermark together with the watermark information ensures the correctness of the watermark extraction.Moreover,after the watermark is embedded,the smooth area of the carrier image changes very slightly compared to the texture area.Good experimental results have been obtained in terms of the visual quality of the image and the bit error rate of the watermark.