Information Steganography Algorithm for Automatic Generation of Song Ci

doi:10.3969/j.issn.1671-1122.2022.07.006

Abstract

Abstract:

It is hard to balance the contradiction between hiding capacity and imperceptibility by traditional text steganography methods. Taking advantage of the semantic richness and syntactic flexibility of Song Ci, an algorithm to generate steganographic Song Ci based on the Seq2Seq model with bidirectional encoder representation from transformers (BERT) and attention mechanism was proposed in this paper. BERT was used as the semantic vector transformation in the generation model. Its rich word vector space ensured the semantic coherence between the generated sentences and improved the quality of the generated Song Ci. In addition, the algorithm designed the word selection method using the rule template and mutual information to restrict the generation of steganographic sentences, which enhanced the security of the hiding algorithm. Compared with the existing and advanced text information hiding algorithms, the experimental results indicate that the embedding rate of the algorithm proposed in this paper is improved by more than 7% in comparison with Ci-stega, and it has good performance in security and robustness.

Key words: text information steganography, Song Ci, metrical template, mutual information

CLC Number:

TP309

YANG Wanxia, CHEN Shuai, GUAN Lei, YANG Zhongliang. Information Steganography Algorithm for Automatic Generation of Song Ci[J]. Netinfo Security, 2022, 22(7): 46-54.

Figures/Tables 12

References 21

[1]	ZHANG Zhenyu, LI Qianmu, QI Yong. Text Watermarking Design Based on Invisible Characters[J]. Journal of Nanjing University of Science and Technology, 2017, 41(4): 405-411.
	张震宇, 李千目, 戚湧. 基于不可见字符的文本水印设计[J]. 南京理工大学学报(自然科学版), 2017, 41(4): 405-411.
[2]	WU Guohua, GONG Lichun, YUAN Lifeng, et al. Review of Information Hiding on Chinese Text[J]. Journal on Communications, 2019, 40(9): 145-156.
	吴国华, 龚礼春, 袁理锋, 等. 中文文本信息隐藏研究进展[J]. 通信学报, 2019, 40(9): 145-156.
[3]	BEHROOZ K, BEHNAM K, BAHMAN K, et al. A New Method for pdf Steganography in Justified Texts[J]. Journal of Information Security and Applications, 2019(45): 61-70.
[4]	YU Zhenshan, HUANG Liusheng, CHEN Zhili, et al. High Embedding Ratio Text Steganography by Ci-Poetry of the Song Dynasty[J]. Journal of Chinese Information Processing, 2009, 23(4): 55-63.
	余振山, 黄刘生, 陈志立, 等. 用宋词实现高嵌入率文本信息隐藏[J]. 中文信息学报, 2009, 23(4): 55-63.
[5]	ALOTAIBI R A, ELREFAEI L A. Improved Capacity Arabic Text Watermarking Methods Based on Open Word Space[J]. Journal of King Saud University-Computer and Information Sciences, 2018, 30(2): 236-248. doi: 10.1016/j.jksuci.2016.12.007 URL
[6]	ZHANG Wei, WANG Xiaomei, ZHANG Chenxu. Coverless Text Steganography Method Based on Tone Feature Mapping[J]. Journal of Information Engineering University, 2021, 22(1): 38-43.
	张维, 王晓梅, 张晨旭. 一种基于声调特征映射的无载体信息隐藏方法[J]. 信息工程大学学报, 2021, 22(1): 38-43.
[7]	CHEN Xianyi, CHEN Sheng, WU Yulei. Coverless Information Hiding Method Based on the Chinese Character Encoding[J]. Journal of Internet Technology, 2017, 18(2): 313-320.
[8]	ZHANG Jianjun, WANG Lucai, LIN Haijun. Coverless Text Information Hiding Method Based on the Word Rank Map[J]. Journal of Internet Technology, 2017, 18(2): 427-434.
[9]	YANG Zhongliang, GUO Xiaoqing, CHEN Ziming, et al. RNN-Stega: Linguistic Steganography Based on Recurrent Neural Networks[J]. IEEE Transactions on Information Forensics and Security, 2019, 14(5): 1280-1295. doi: 10.1109/TIFS.2018.2871746 URL
[10]	KANG Huixian, WU Hanzhou, ZHANG Xinpeng. Generative Text Steganography Based on LSTM Network and Attention Mechanism with Keywords[J]. Electronic Imaging, 2020(4): 291-298.
[11]	LIU Yanchen, WANG Jian, WANG Zhibin, et al. A Technique of High Embedding Rate Text Steganography Based on Whole Poetry of Song Dynasty[C]// Springer. Proceedings of the Second International Conference on Cloud Computing and Security. Berlin: Springer, 2016: 178-189.
[12]	LUO Yubo, HUANG Yongfeng, LI Fufang, et al. Text Steganography Based on Ci-Poetry Generation Using Markov Chain Model[J]. KSII Transactions on Internet and Information Systems, 2016, 10(9): 4568-4584.
[13]	YI Xiaoyuan, LI Ruoyu, SUN Maosong. Generating Chinese Classical Poems with RNN Encode-Decoder[C]// IEEE. Proceedings of the 16th Chinese Computational Linguistics. New York: IEEE, 2017: 211-223.
[14]	YI Xiaoyuan, SUN Maosong, LI Ruoyu, et al. Chinese Poetry Generation with a Working Memory Mode[C]// ACM. Proceding of the 27th International Joint Conference on Artificial Intelligence. New York: ACM, 2018: 4553-4559.
[15]	LUO Yubo, HUANG Yongfeng. Text Steganography with High Embedding Rate: Using Recurrent Neural Networks to Generate Chinese Classic Poetry[C]// ACM. Proceedings of the Seventh ACM Workshop on Information Hiding & Multimedia Security. New York: ACM, 2017: 99-104.
[16]	QIN Chuan, WANG Meng, SI Guangwen, et al. Constructive Information Hiding with Chinese Quatrain Generation[J]. Chinese Journal of Computers, 2021, 44(4): 773-785.
	秦川, 王萌, 司广文, 等. 基于绝句生成的构造式信息隐藏算法[J]. 计算机学报, 2021, 44(4): 773-785.
[17]	ZHENG Haitao, WANG Wei, CHEN Wang, et al. Automatic Generation of News Comments Based on Gated Attention Neural Networks[J]. IEEE Access, 2018, 6: 702-710. doi: 10.1109/ACCESS.2017.2774839 URL
[18]	MA Liang, WANG Zili, YANG Fan, et al. Robust State of Charge Estimation Based on a Sequence-to-Sequence Mapping Model with Process Information[J]. Journal of Power Sources, 2020(474): 30-35.
[19]	LI Xuehong, GUO Hui, ANG Chaoqun. Text Digital Watermarking Algorithm Based on Improve Line Shift Coding[J]. Computer & Digital Engineering, 2014, 42(9): 1687-1690, 1712.
	李雪红, 郭晖, 昂朝群. 改进行移编码的文本数字水印算法研究[J]. 计算机与数字工程, 2014, 42(9): 1687-1690,1712.
[20]	KIM M Y, ZAIANE O R. Natural Lanuage Watermarking Based on Syntactic Displacement and Morphological Divsion[C]// IEEE. Computer Software and Applications Conference Workshops (COMPSACW). New York:IEEE, 2010: 164-169.
[21]	CHAPMAN M, DAVIDA G. Hiding the Hidden: A Software System for Concealing Ciphertext in Innocuous Text[C]// Springer. Proceedings of the International Conference on Information and Communications Security. Berlin: Springer, 1997: 335-345.

Char1	Char2	Char3	Char4	Char5	Char6	Char7
风 00	人 00	隔 00	烟 00	水 00	忧 00	江 00
白 01	舟 01	照 01	薄 01	月 01	峰 01	早 01
断 10	柔 10	日 10	窗 10	春 10	雪 10	归 10
众 11	云 11	响 11	林 11	雨 11	鸣 11	暖 11

编码器输入	编码器输出
秋月	春花秋月何时了
往事&春花秋月何时了	往事知多少
往事&春花秋月何时了&往事知多少	小楼昨夜又东风
…	…
东流&春花秋月何时了&往事知多少&小楼昨夜又东风&故国不堪回首月明中&雕栏玉砌应犹在&只是朱颜改&问君能有几多愁&恰是一江春水向东流	恰是一江春水向东流

参数	值
Song ci vocab size	6000
Type vocab size	2
Number hidden layers	8
Number attention heads	8
Max position embeddings	100
Hidden size	128
Hidden dropout prob	0.1
Hidden activate function	gelu
Intermediate size	3072

信息隐藏算法	各隐藏算法的复杂度	信息提取算法的复杂度
基于宋词的算法^[4]	O(1)	O(n)
基于全宋词的算法^[11]	O(1)	O(n)
基于马尔可夫模型的宋词生成算法（Ci-stega）^[12]	O(${{n}^{2}}$)	O(n)
本文算法	O(${{n}^{2}}$)	O(${{n}^{2}}$)

信息隐藏算法	修改格式	修改内容	语义分析	统计分析
基于文本格式的算法^[19]	Bad	Bad	Good	Good
基于语义生成的算法^[20]	Good	Bad	Good	Bad
基于全宋词的算法^[11]	Good	Bad	Bad	Bad
基于Markov的宋词生成算法^[12]	Good	Bad	Good	Bad
本文算法	Good	Bad	Good	Good