基于可逆神经网络的小波域抗屏摄数字水印

打开文本图片集

中图分类号：TP929 文献标志码：A

Screen-shooting resilient digital watermarking based on invertible neural network in wavelet domain

CHENG Senmao1， GUO Daidou’，LIFengyong²， HAN Yanfang1，QIN Chuan1 （1.SchoolofOptical-ElectricalandComputerEnginering，UniversityofanghaifoienceandThologangha03， China;2.CollgeofComputer ScienceandTechnology，Shanghai UniversityofElectricPower，Shanghai 2Ol306，China）

Abstract： Multimedia security in screen-shooting channel transmisson remains a major challenge in the digital watermarking research. In view of the problems of differences in light intensities and sampling distortions during the screen-shooting process， watermarking schemes designed based on digital channels are not suitable for screen-shooting channels. Therefore， a screen-shooting resilient watermarking scheme based on invertible neural networks in the wavelet domain was proposed to address the issue of "cross-media robustness". First， the watermark message and the original image were preprocessed using a preprocessing network based on the U-Net structure to generate the residual image. Next， the residual image was subjected to discrete wavelet transform with the original image， and the invertible neural network was used to embed and extract watermarks in the wavelet domain. Finally， a noise pool was integrated into the model training process to enhance the robustness against screenshooting noise atacks. Simulation results demonstrate that the proposed scheme generates watermarked images with better visual quality and achieves high accuracy of watermark extraction at diffrent distances， angles， and light intensities of screen-shooting.

Keywords： robust watermarking; screen-shooting; invertible neural network; wavelet domain

随着智能便携摄影设备拥有越来越高的拍照性能和普及率，拍照成为了日常生活中最简单高效的信息传递方式之一。（剩余15623字）