信息安全研究2024，Vol.10Issue(9)：849-855,7.DOI:10.12379/j.issn.2096-1057.2024.09.09
基于注意力机制和护照层嵌入的图像处理模型水印方法
Image Processing Model Watermarking Method Based on Attention Mechanism and Passport Layer Embedding
陈先意 1周浩 1刘腾骏 1闫雷鸣1
作者信息
- 1. 南京信息工程大学计算机学院 南京 210044
- 折叠
摘要
Abstract
With the wide application of deep neural networks in the field of artificial intelligence,the copyright protection of deep neural networks has received extensive attention.However,so far,most of the methods for model copyright protection focus on detection or classification tasks,and are difficult to be directly applied to image processing networks.To this end,this paper proposes an image processing model copyright protection framework combining attention mechanism and passport layer embedding.Firstly,the channel and spatial attention network are used in the watermark embedding network to locate the human eye insensitive area in the image,which improves the robustness and imperceptibility of the watermark.Secondly,the passport layer watermark is inserted after the convolution layer of the target model to improve the ability to resist the ambiguity attacks.Finally,the combination loss is designed to guide the convergence direction of the model in combination with structural consistency and passport layer factors.Experimental results on super-resolution and semantic segmentation models show that the watermark extraction rate of this method is more than 98％,and it has good robustness to surrogate attack and ambiguity attack.关键词
深度学习/模型水印/版权保护/注意力机制/护照层Key words
deep learning/model watermark/copyright protection/attention mechanism/passport layer分类
信息技术与安全科学引用本文复制引用
陈先意,周浩,刘腾骏,闫雷鸣..基于注意力机制和护照层嵌入的图像处理模型水印方法[J].信息安全研究,2024,10(9):849-855,7.
