GAN Based Data Watermarking For Text Generative

Abstract

Abstract: Coverless steganography realizes covert writing by establishing mapping relationship between digital watermarking information and image characteristic information, to realize data integrity protection and content traceability tracking of image data transmission in the Internet environment. However, the existing image steganography requires many natural images to be prepared in advance to form the image data set, and the natural image selection bias will lead to incomplete or incorrect information transmission. To solve the above problems, this paper proposes an image watermarking steganography method without carrier based on generating countermeasure network. The method uses the generator of the generated adversarial network to generate the forged image like the original image from random noise, image label and digital watermark information, and the discriminator of the generated adversarial network is responsible for determining the true and false of the input image, and extracting the label and digital watermark information at the same time. After several rounds of confrontation training, the generator finally outputs image data like the original image and containing digital watermark, while the naked eye cannot distinguish the difference between the original image and the generated image. The experimental analysis shows that SCRMQ1 is used for feature extraction, and the error detection rate of integrated classifier is 48.5%. Embedding capacity up to 1 BPP; The accuracy of digital watermark extraction is up to 99.5%.

Key words: data watermarking, text sequence, generative adversarial network, reinforcement learning

摘要： 无载体隐写技术通过将数字水印信息与图像自身特征信息建立映射关系实现隐蔽写入，从而实现图像数据在互联网环境下传播的数据完整性保护和内容溯源追踪．但现有的图像无载体隐写方法存在需要事先准备大量自然图像构成图像数据集的问题，自然图像选择偏差会导致信息传递的不完整或错误．针对上述问题，本文提出一种基于生成对抗网络的图像数据水印无载体隐写方法．该方法利用生成对抗网络的生成器从随机噪声、图像标签和数字水印信息生成类似于原始图像的伪造图像，生成对抗网络的判别器则负责判别输入图像的真、假，并同时提取标签和数字水印信息．通过多轮的对抗训练后，生成器最后输出类似原始图像且含有数字水印的图像数据、同时肉眼无法区分原始图像和生成图像的区别．经过试验分析表明，使用SCRMQ1做特征提取，用集成分类器检测错误率为48.5%；嵌入容量高达1bpp以上；数字水印提取的准确率最高可达99.5%．

关键词: 数据水印, 无载体隐写, 生成对抗网络, 强化学习

邹振婉李明轩杨慧婷. 一种基于生成对抗网络的图像数据水印无载体隐写技术[J]. 信息安全研究, 2022, 8(1): 2-.

References

[1] Ye Jian, Ni Jiangqun, Yi Yang．Deep learning hierarchical representations for image steganalysis [J]. IEEE Trans on Information Forensics and Security, 2017, 12(11): 2545-2557

[2] Ruan Feng, Zhang Xing, Zhu Dawei, et al. Deep learning for real-time image steganalysis: A survey[J]. Journal of Real-Time Image Processing, 2020, 17(1): 149-160

[3] Wang Ze, Chen Mingzhi, Yang Yu, et al. Joint multi-domain feature learning for image steganalysis based on CNN[J]. EURASIP Journal on Image and Video Processing, 2020, 2020(1): 1-12

[4] Yang Chenghsing, Weng Chiyao, Wang Shuihjeng, et al. Adaptive data hiding in edge areas of images with spatial LSB domain systems[J]. IEEE Trans on Information Forensics and Security, 2008, 3(3): 488-497

[5] Pevný T, Filler T, Bas P. Using high-dimensional image models to perform highly undetectable steganography[C] //Proc of the International Workshop on Information Hiding．Piscataway, NJ: IEEE, 2010: 161-177

[6] Holub V, Fridrich J. Designing steganographic distortion using directional filters[C] // Proc of the 2012 IEEE Int Workshop on Information Forensics and Security (WIFS)．Piscataway, NJ: IEEE, 2012: 234-239

[7] Holub V, Fridrich J. Digital image steganography using universal distortion[C] //Proc of the 1st ACM Workshop on Information Hiding and Multimedia Security. New York: ACM, 2013: 59-68

[8] Li Bin, Tan Shunquan, Wang Ming, et al. Investigation on cost assignment in spatial image steganography[J]. IEEE Trans on Information Forensics and Security, 2014, 9(8): 1264-1277

[10] Chen Wenyuan. Color image steganography scheme using set partitioning in hierarchical trees coding, digital Fourier transform and adaptive phase modulation[J]. Applied Mathematics and Computation, 2007, 185(1): 432-448

[11] McKeon R T. Strange Fourier steganography in movies[C] //Proc of the 2007 IEEE Int Conf on Electro/Information Technology. Piscataway, NJ: IEEE, 2007: 178-182

[12] Cox I J, Kilian J, Leighton F T, et al. Secure spread spectrum watermarking for multimedia[J]. IEEE Trans on Image Processing, 1997, 6(12): 1673-1687

[13] Valandar M Y, Ayubi P, Barani M J. A new transform domain steganography based on modified logistic chaotic map for color images[J]. Journal of Information Security and Applications, 2017, 34: 142-151

[14] Bilal M, Imtiaz S, Abdul W, et al. Zero-steganography using DCT and spatial domain[C] //Proc of the 2013 ACS Int Conf on Computer Systems and Applications (AICCSA). Piscataway, NJ: IEEE, 2013: 1-7

[15] Zhou Zhili, Cao Yi, Sun Xingming. Coverless information hiding based on bag-of-words model of image[J]. Journal of Applied Sciences, 2016, 34(5): 527-536

[16] Pan Nan, Qin Jiaohua, Tan Yun, et al. A video coverless information hiding algorithm based on semantic segmentation[J]. EURASIP Journal on Image and Video Processing, 2020, 2020(1): 1-18

[17] Zheng Shuli, Wang Liang, Ling Baohong, et al. Coverless information hiding based on robust image hashing[C] //Proc of International Conference on Intelligent Computing.Berlin:springer,2017:536-547

[18] Duan Xintao, Song Haoxian. Coverless information hiding based on generative model[J]. arXiv preprint, arXiv:1802.03528, 2018

[19] Hu Donghui, Wang Liang, Jiang Wenjie, et al. A novel image steganography method via deep convolutional generative adversarial networks[J]. IEEE Access, 2018, 6: 38303-38314

[20] Zhou Zhili, Cao Yi, Sun Xingming. Coverless information hiding based on bag-of-words model of image[J]. Journal of Applied Sciences, 2016, 34(5): 527-536