北京邮电大学学报

  • EI核心期刊

北京邮电大学学报 ›› 2018, Vol. 41 ›› Issue (6): 7-13.doi: 10.13190/j.jbupt.2018-032

• 论文 • 上一篇    下一篇

基于图像描述的文本信息隐藏

薛一鸣1, 周雪婧1, 周小诗1, 牛少彰2, 文娟1   

  1. 1. 中国农业大学 信息与电气工程学院, 北京 100083;
    2. 北京邮电大学 计算机学院, 北京 100876
  • 收稿日期:2018-02-02 出版日期:2018-12-28 发布日期:2018-12-24
  • 作者简介:薛一鸣(1968-),男,副教授;文娟(1982-),女,讲师,E-mail:wenjuan@cau.edu.cn.
  • 基金资助:
    国家自然科学基金项目(61802410,61872368)

Text Steganography Based on Image Caption

XUE Yi-ming1, ZHOU Xue-jing1, ZHOU Xiao-shi1, NIU Shao-zhang2, WEN Juan1   

  1. 1. College of Information and Electrical Engineering, China Agricultural University, Beijing 100083, China;
    2. School of Computer Science, Beijing University of Posts and Telecommunication, Beijing 100876, China
  • Received:2018-02-02 Online:2018-12-28 Published:2018-12-24

摘要: 针对文本信息隐藏嵌入容量低和语义连贯性差的问题,提出了一种基于神经网络图像描述的文本信息隐藏模型.将卷积神经网络与长短期记忆网络相结合,把图像特征和生成语句进行关联.从收发双方能否共享图像及模型参数的不同应用前提出发,设计了多种概率采样方式,从而生成载密的图像描述文本.实验结果表明,该算法具有较高的隐藏容量,载密描述句能较好地表达图像内容.该模型归属于"无载体"自然语言生成式信息隐藏,具有较好的隐蔽性和安全性.

关键词: 文本信息隐藏, 图像描述, 卷积神经网络, 长短期记忆网络

Abstract: Aiming at the problem of low embedding capacity and poor semantic coherence of text steganography, a text steganographic scheme based on neural image caption is proposed. An encode-decode structure with a combination of long short term memory and convolution neural network is used to model the joint probability distributions between image features and the descriptive sentences. Two methods with different sampling process are designed from the perspectives of sharing and non-sharing models. Experimental results show that the proposed model can achieve high embedding capacity and desirable text quality. This scheme belongs to "carrier-free" steganography and has good security.

Key words: text steganography, image caption, convolutional neural network, long short term memory

中图分类号: