基于小波分解与动态密集空洞卷积的双分支神经网络水印算法

doi:10.3969/j.issn.1671-1122.2025.05.014

摘要/Abstract

摘要：

基于深度学习的数字水印算法主要倾向于向载体图像的中高频区域嵌入水印信息，只使用卷积神经网络将水印信息映射到载体图像的特征空间中，忽略了频域算法的作用。文章提出一种基于小波分解与动态密集空洞卷积的双分支神经网络水印算法，通过使用小波分解，更好地引导水印信息的嵌入和提取。该算法运用离散小波变换处理载体图像与水印图像，将其分解为高频信息和低频信息，再使用神经网络进行针对性学习，使用动态密集空洞卷积在减少神经网络层数的情况下，扩大感受野，增强捕捉全局信息的能力，也能避免使用过多的池化层影响重建图像的质量。实验表明，该算法拥有良好的不可见性和鲁棒性。

关键词: 双分支神经网络, 离散小波变换, 空洞卷积, 数字水印

Abstract:

Deep learning based digital watermarking algorithms mainly tended to embed watermark information into the mid to high frequency regions of the carrier image. They only used convolutional neural networks to map the watermark information to the feature space of the carrier image and ignored the role of frequency domain algorithms. The article proposed a dual branch neural network watermarking algorithm based on wavelet decomposition and dynamic dense dilated convolution. By using wavelet decomposition, it better guided the embedding and extraction of watermark information. This algorithm used discrete wavelet transform to process carrier images and watermark images, decomposed them into high-frequency and low-frequency information, and then used neural networks for targeted learning. Dynamic dense dilated convolution was used to expand the receptive field, enhanced the ability to capture global information while reducing the number of neural network layers. It could also avoid using too many pooling layers that affected the quality of reconstructed images. The experiment show that the algorithm has good invisibility and robustness.

Key words: dual branch neural network, discrete wavelet transform, hollow convolution, digital watermark

中图分类号:

TP309

李敬有, 席晓天, 魏荣乐, 张光妲. 基于小波分解与动态密集空洞卷积的双分支神经网络水印算法[J]. 信息网络安全, 2025, 25(5): 828-839.

LI Jingyou, XI Xiaotian, WEI Rongle, ZHANG Guangda. Double Branch Neural Network Watermarking Algorithm Based on Wavelet Decomposition and Dynamic Dense Dilated Convolution[J]. Netinfo Security, 2025, 25(5): 828-839.

图/表 19

图1

图2

图3

图4

图5

图6

表1

图7

表2

表3

表4

表5

表6

表7

表8

表9

表10

表11

表12

参考文献 21

[1]	CHENG Bin, ZHANG Bo. Application of Robust Digital Watermarking Technology in Image Copyright Disputes[EB/OL]. (2021-04-01)[2025-02-20]. https://dx.doi.org/10.1088/1742-6596/1883/1/012124.
[2]	SOUNDRAPANDIYAN R, RAJENDIRAN K, GURUNATHAN A, et al. Analysis of DWT-DCT Watermarking Algorithm on Digital Medical Imaging[EB/OL]. [2025-02-20]. https://pubmed.ncbi.nlm.nih.gov/38162418/.
[3]	CAO Fang, GUO Daidou, WANG Tianjun, et al. Universal Screen-Shooting Robust Image Watermarking with Channel-Attention in DCT Domain[EB/OL]. (2023-10-06)[2025-02-20]. https://www.sciencedirect.com/science/article/abs/pii/S0957417423025642?via%3Dihub.
[4]	BALUJA S. Hiding Images in Plain Sight: Deep Steganography[J]. Advances in Neural Information Processing Systems, 2017, 30: 2066-2076.
[5]	ZHU Jiren, KAPLAN R, JOHNSON J, et al. HIDDEN: Hiding Data with Deep Networks[C]// Springer. Proceedings of the European Conference on Computer Vision (ECCV). Heidelberg: Springer, 2018: 657-672.
[6]	TANCIK M, MILDENHALL B, NG R. StegaStamp: Invisible Hyperlinks in Physical Photographs[C]// IEEE. 2020 IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR). New York: IEEE, 2020: 2117-2126.
[7]	ZHU Liuhao, ZHAO Yi, FANG Yixiang, et al. A Novel Robust Digital Image Watermarking Scheme Based on Attention U-Net++ Structure[J]. The Visual Computer, 2024, 40(12): 8791-8807.
[8]	ZHANG Chaoning, BENZ P, KARJAUV A, et al. UDH: Universal Deep Hiding for Steganography, Watermarking, and Light Field Messaging[J]. Advances in Neural Information Processing Systems, 2020, 33: 10223-10234.
[9]	JING Junpeng, DENG Xin, XU Mai, et al. HiNet: Deep Image Hiding by Invertible Network[C]// IEEE. 2021 IEEE/CVF International Conference on Computer Vision (ICCV). New York: IEEE, 2021: 4713-4722.
[10]	GIRI K J, QUADRI S M K, BASHIR R, et al. DWT Based Color Image Watermarking: A Review[J]. Multimedia Tools and Applications, 2020, 79(43): 32881-32895.
[11]	CHANG Shiyu, ZHANG Yang, HAN Wei, et al. Dilated Recurrent Neural Networks[J]. Advances in Neural Information Processing Systems, 2017, 30: 76-86.
[12]	YU F, KOLTUN V. Multi-Scale Context Aggregation by Dilated Convolutions[EB/OL]. (2016-04-30)[2025-02-20]. https://arxiv.org/abs/1511.07122.
[13]	WANG Panqu, CHEN Pengfei, YUAN Ye, et al. Understanding Convolution for Semantic Segmentation[C]// IEEE. 2018 IEEE Winter Conference on Applications of Computer Vision (WACV). New York: IEEE, 2018: 1451-1460.
[14]	SZEGEDY C, LIU Wei, JIA Yangqing, et al. Going Deeper with Convolutions[C]// IEEE. 2015 IEEE Conference on Computer Vision and Pattern Recognition (CVPR). New York: IEEE, 2015: 1-9.
[15]	HAN Yizeng, HUANG Gao, SONG Shiji, et al. Dynamic Neural Networks: A Survey[J]. IEEE Transactions on Pattern Analysis and Machine Intelligence, 2021, 44(11): 7436-7456.
[16]	ALMOHAMMAD A, GHINEA G. Stego Image Quality and the Reliability of PSNR[C]// IEEE. 2010 2nd International Conference on Image Processing Theory, Tools and Applications. New York: IEEE, 2010: 215-220.
[17]	WANG Zhou, BOVIK A C, SHEIKH H R, et al. Image Quality Assessment: From Error Visibility to Structural Similarity[J]. IEEE Transactions on Image Processing, 2004, 13(4): 600-612. doi: 10.1109/tip.2003.819861 pmid: 15376593
[18]	LUO Xiyang, ZHAN Ruohan, CHANG Huiwen, et al. Distortion Agnostic Deep Watermarking[C]// IEEE. 2020 IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR). New York: IEEE, 2020: 13548-13557.
[19]	DING Weiping, MING Yurui, CAO Zehong, et al. A Generalized Deep Neural Network Approach for Digital Watermarking Analysis[J]. IEEE Transactions on Emerging Topics in Computational Intelligence, 2021, 6(3): 613-627.
[20]	BALUJA S. Hiding Images within Images[J]. IEEE Transactions on Pattern Analysis and Machine Intelligence, 2019, 42(7): 1685-1697.
[21]	DUAN Xintao, WU Guoming, LI Chun, et al. DUIANet: A Double Layer U-Net Image Hiding Method Based on Improved Inception Module and Attention Mechanism[EB/OL]. (2023-12-23)[2025-02-20]. https://doi.org/10.1016/j.jvcir.2023.104035.

噪声类型	编码图像${{I}_{en}}$	噪声图像${{I}_{no}}$	噪声图像和含水印图像差异放大20倍后的成像$20\left\| {{I}_{en}}-{{I}_{no}} \right\|$
像素替换
高斯滤波
近似JPEG
旋转
平移
椒盐噪声
中值滤波

原始载体图像${{I}_{co}}$	原始水印图像${{W}_{in}}$	提取出的水印图像${{W}_{out}}$	图像的残差成像\|${{W}_{in}}-~{{W}_{out}}$\|	图像的残差成像放大10倍后的成像10\|${{W}_{in}}-~{{W}_{out}}$\|

	PSNR	SSIM
平均值	39.49	0.9900
最大值	41.58	0.9908
最小值	37.79	0.9891

原始载体图像${{I}_{co}}$	融合图像${{I}_{en}}$	图像的残差成像\|${{I}_{co}}-~{{I}_{en}}$\|	残差放大10倍后的成像10\|${{I}_{co}}-~{{I}_{en}}$\|

	PSNR	SSIM
平均值	41.34	0.9907
最大值	43.22	0.9912
最小值	39.56	0.9900