Please wait a minute...
浙江大学学报(理学版)  2021, Vol. 48 Issue (3): 261-269    DOI: 10.3785/j.issn.1008-9497.2021.03.001
图像处理算法     
基于快速神经网络架构搜索的鲁棒图像水印网络算法
王小超1, 张雷1, 余元强2, 胡坤3, 胡建平4
1.天津工业大学 数学科学学院,天津 300387
2.天津大学 智能与计算学部,天津 300350
3.中国科学院大学,北京 100049
4.东北电力大学 理学院,吉林吉林 132012
Robust image watermarking network algorithm based on effective neural architecture search
WANG Xiaochao1, ZHANG Lei1, YU Yuanqiang2, HU Kun3, HU Jianpiang4
1.School of Mathematical Science,Tiangong University, Tianjin 300387, China
2.College of Intelligence and Computing, Tianjin University, Tianjin 300350, China
3.University of Chinese Academy of Science, Beijing 100049, China
4.School of Science, Northeast Electric Power University, Jilin 132012, Jilin Province, China
 全文: PDF(3130 KB)   HTML  
摘要: 为解决深度学习在图像水印算法中计算量大且模型冗余的问题,提高图像水印算法在抵抗噪声、旋转和剪裁等攻击时的鲁棒性,提出基于快速神经网络架构搜索(neural architecture search,NAS)的鲁棒图像水印网络算法。通过多项式分布学习快速神经网络架构搜索算法,在预设的搜索空间中搜索最优网络结构,进行图像水印的高效嵌入与鲁棒提取。首先,将子网络中线性连接的全卷积层设置为独立的神经单元结构,并参数化表示结构单元内节点的连接,预先设定结构单元内每个神经元操作的搜索空间;其次,在完成一个批次的数据集训练后,依据神经元操作中的被采样次数和平均损失函数值动态更新概率;最后,重新训练搜索完成的网络。水印网络模型的参数量较原始网络模型缩减了92%以上,大大缩短了模型训练时间。由于搜索得到的网络结构更为紧凑,本文算法具有较高的时间性能和较好的实验效果,在隐藏图像时,对空域信息的依赖比原始网络更少。对改进前后的2个网络进行了大量鲁棒性实验,对比发现,本文算法在CIFAR-10数据集上对抵抗椒盐噪声和旋转、移除像素行(列)等攻击优势显著;在ImageNet数据集上对抵抗椒盐高斯噪声、旋转、中值滤波、高斯滤波、JPEG压缩、裁剪等攻击优势显著,特别是对随机移除行(列)和椒盐噪声有较强的鲁棒性。
关键词: 图像水印多项式分布学习神经网络架构搜索(NAS)    
Abstract: In order to solve the problem of large computation and model redundancy when applying deep learning to image watermarking,and to improve the robustness of image watermarking algorithm against noise,rotation,and cropping attacks,this paper adopts the robust image neural architecture search (NAS). Multinomial distribution learning for effective the neural architecture search watermarking network algorithm based on algorithm to find the optimal network structure in the preset search space,and performs efficient embed-ding and robust extraction of image watermarks. Firstly,the linearly connected fully convolutional layers in the sub-network are changed to independent neural unit structure,then we parameterize the connection of nodes in the structural unit,and set the search space for each neuron operation in the structural unit. Secondly,after completing a batch of data set training each time,the probabilities of being selected for the next operation are dynamically updated according to the number of samples and the average loss function value.Finally,the network structure after the search is retrained,which reduces the model parameters by more than 92% comparing with original model as well as the model training time. Since the obtained network structure is more compact,the proposed algorithm can achieve higher performance and better experimental results.Compared to the original network,the newly-developed network relies less on spatial information when hiding images.A larger number of experiments are performed to illustrate the advantages of the proposed algorithm including resisting salt and pepper noise,rotation and flipping,and removing pixel rows (columns) and other attacks on the CIFAR-10 data set; and has significant advantages against attacks of median filtering,Gaussian filtering,JPEG compression,and cropping on ImageNet,especially for the random removal of rows (columns) and salt and pepper noise.
Key words: multinomial distribution learning    image watermarking    neural architecture search (NAS)
收稿日期: 2020-09-11 出版日期: 2021-05-20
CLC:  TP 391  
基金资助: 天津市高等学校基本科研业务费项目(2018KJ222).
通讯作者: ORCID:https://orcid.org/0000-0001-8675-4564,E-mail:neduhjp307@163.com.     E-mail: neduhjp307@163.com
作者简介: 王小超(1987—),ORCID:https://orcid.org/0000-0001-7852-2106,男,博士,副教授, 主要从事计算机图形学、图像处理等研;
服务  
把本文推荐给朋友
加入引用管理器
E-mail Alert
RSS
作者相关文章  
王小超
张雷
余元强
胡坤
胡建平

引用本文:

王小超, 张雷, 余元强, 胡坤, 胡建平. 基于快速神经网络架构搜索的鲁棒图像水印网络算法[J]. 浙江大学学报(理学版), 2021, 48(3): 261-269.

WANG Xiaochao, ZHANG Lei, YU Yuanqiang, HU Kun, HU Jianpiang. Robust image watermarking network algorithm based on effective neural architecture search. Journal of Zhejiang University (Science Edition), 2021, 48(3): 261-269.

链接本文:

https://www.zjujournals.com/sci/CN/10.3785/j.issn.1008-9497.2021.03.001        https://www.zjujournals.com/sci/CN/Y2021/V48/I3/261

1 顾宇鑫,马小虎. 采用稀疏变换和拉普拉斯金字塔的数字水印算法[J]. 计算机辅助设计与图形学学报,2018,30(5):901-910. GU Y X,MA X H. Digital watermarking algorithm using sparse transform and Laplace pyramid[J].Journal of Computer Aided Design & Computer Graphics,2018,30(5):901-910.
2 丁玮,闫伟齐,齐东旭. 基于Arnold变换的数字图像置乱技术[J]. 计算机辅助设计与图形学学报,2001,13(4):338-341. DING W,YAN W Q,QI D X.Digital image scrambling technology based on Arnold transformation[J].Journal of Computer-Aided Design & Computer Graphics,2001,13(4):338-341.
3 VERMA V S,BHARDWAJ A,JHA R K. A new scheme for watermark extraction using combined noise-induced resonance and support vector machine with PCA based feature reduction[J]. Multimedia Tools and Applications,2019,78(16):23203-23224. DOI:10.1007/s11042-019-7599-z
4 SUN L,XU J,LIU S,et al. A robust image watermarking scheme using Arnold transform and BP neural network[J].Neural Computing and Applications,2018,30(8):2425-2440. DOI:10.1007/s00521-016-2788-4
5 CHEN J,KANG X,LIU Y,et al. Median filtering forensics based on convolutional neural networks[J].IEEE Signal Processing Letters,2015,22(11):1849-1853. DOI:10.1109/lsp.2015.2438008
6 MUN S M,NAM S H,JANG H,et al. Finding robust domain from attacks:A learning framework for blind watermarking[J]. Neurocomputing,2019,337:191-202. DOI:10.1016/j.neucom.2019.01.067
7 BALUJA S. Hiding images within images[J]. IEEE Transactions on Pattern Analysis and Machine Intelligence,2020,42(7):1685-1697. DOI:10.1109/tpami.2019.2901877
8 ELSKEN T,METZEN J H,HUTTER F. Neural architecture search:A survey[J]. Journal of Machine Learning Research,2019,20:1-21.
9 ZOPH B,VASUDEVAN V,SHLENS J,et al.Learning transferable architectures for scalable image recognition[C]//Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition. Salt Lake City:IEEE,2018:8697-8710. DOI:10.1109/cvpr.2018.00907
10 PHAM H,GUAN M,ZOPH B,et al. Efficient neural architecture search via parameters sharing[C]//International Conference on Machine Learning.Stockholm:IMLS,2018:4095-4104.
11 CAI H,YANG J,ZHANG W,et al.Path-level network transformation for efficient architecture search[C]//International Conference on Machine Learning. Stockholm:IMLS,2018:678-687.
12 XIE L,YUILLE A. Genetic cnn[C]//Proceedings of the IEEE International Conference on Computer Vision.Venice:IEEE,2017:1379-1388. DOI:10.1109/iccv.2017.154
13 REAL E,AGGARWAL A,HUANG Y,et al.Regularized evolution for image classifier architecture search[C]//Proceedings of the AAAI Conference on Artificial Intelligence.Hawaiian:AAAI,2019:4780-4789. DOI:10.1609/aaai.v33i01.33014780
14 LIU H X,KAREN S,YIMING Y.Darts:Differentiable architecture search[C]//International Conference on Learning Representations.New Orleans:ICLR,2019:1-13.
15 ZHENG X W,JI R R,TANG L,et al. Multinomial distribution learning for effective neural architecture search[C]//The IEEE International Conference on Computer Vision,Los Angeles:IEEE,2019:1304-1313. DOI:10.1109/ICCV.2019.00139
16 HUANG G,LI Y X,PLEISS G,et al. Snapshot ensembles:Train 1,get M for free[C]// International Conference on Learning Representations. Toulon:ICLR,2017:1-14. DOI:10.1109/robio.2017.8324553
17 LI X,CAI J. Robust transmission of JPEG2000 encoded images over packet loss channels[C]//2007 IEEE International Conference on Multimedia and Expo. Beijing:IEEE,2007:947-950.
18 THOMOS N,BOULGOURIS N V,STRINTZIS M G. Optimized transmission of JPEG2000 streams over wireless channels[J]. IEEE Transactions on Image Processing,2005,15(1):54-67.
19 WANG Z,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
20 ZHANG K,ZUO W,ZHANG L. FFDNet:Toward a fast and flexible solution for CNN-based image denoising[J].IEEE Transactions on Image Processing,2018,27(9):4608-4622. DOI:10.1109/tip.2018.2839891
21 GUO S,YAN Z,ZHANG K,et al.Toward convolutional blind denoising of real photographs[C]//Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition.Long Beach:IEEE,2019:1712-172. DOI: 10.1109/cvpr.2019.00181
[1] 谭晓东,赵奇,文明珠,王小超. 基于BEMD、DCT和SVD的混合图像水印算法[J]. 浙江大学学报(理学版), 2023, 50(4): 442-454.
[2] 方于华,叶枫. MFDC-Net:一种融合多尺度特征和注意力机制的乳腺癌病理图像分类算法[J]. 浙江大学学报(理学版), 2023, 50(4): 455-464.
[3] 孔翔,陈军. 一类带4个形状参数的同次三角曲面构造算法[J]. 浙江大学学报(理学版), 2023, 50(2): 153-159.
[4] 张远鹏,陈鸿韬,王伟娜. 基于非凸非光滑变分模型的灰度图像泊松噪声移除算法[J]. 浙江大学学报(理学版), 2023, 50(2): 160-166.
[5] 李军成,刘成志,罗志军,龙志文. 空间参数曲线的双目标能量极小化方法及其应用[J]. 浙江大学学报(理学版), 2023, 50(1): 63-68.
[6] 全浩荣,刘成志,李军成,杨炼,胡丽娟. 张量积型Said-Ball曲面的预处理渐近迭代逼近法[J]. 浙江大学学报(理学版), 2022, 49(6): 682-690.
[7] 虞瑞麒,刘玉华,沈禧龙,翟如钰,张翔,周志光. 表征学习驱动的多重网络图采样[J]. 浙江大学学报(理学版), 2022, 49(3): 271-279.
[8] 律睿慜,张陶洁,席旭,王濛濛,孟磊,张克俊. 笔法与结构对楷书文字美学品质影响的量化研究[J]. 浙江大学学报(理学版), 2022, 49(3): 261-270.
[9] 祝锦泰,叶继华,郭凤,江蕗,江爱文. FSAGN: 一种自主选择关键帧的表情识别方法[J]. 浙江大学学报(理学版), 2022, 49(2): 141-150.
[10] 钟颖,王松,吴浩,程泽鹏,李学俊. 基于SEMMA的网络安全事件可视探索[J]. 浙江大学学报(理学版), 2022, 49(2): 131-140.
[11] 朱强,王超毅,张吉庆,尹宝才,魏小鹏,杨鑫. 基于事件相机的无人机目标跟踪算法[J]. 浙江大学学报(理学版), 2022, 49(1): 10-18.
[12] 杨猛,丁曙,马云涛,谢佳翊,段瑞枫. 基于纹理特征的小麦锈病动态模拟方法[J]. 浙江大学学报(理学版), 2022, 49(1): 1-9.
[13] 傅汝佳, 冼楚华, 李桂清, 万隽杰, 曹铖, 杨存义, 高月芳. 面向表型精确鉴定的豆株快速三维重建[J]. 浙江大学学报(理学版), 2021, 48(5): 531-539.
[14] 余鹏, 刘兰, 蔡韵, 何煜, 张松海. 基于单目摄像头的自主健身监测系统[J]. 浙江大学学报(理学版), 2021, 48(5): 521-530.
[15] 林俊聪, 陈萌, 施渝斌, 雷钧, 郭诗辉, 高星, 廖明宏, 金小刚. 面向高级定制的个性化虚拟服装展示[J]. 浙江大学学报(理学版), 2021, 48(4): 418-426.