Page 277 - 《软件学报》2021年第9期

软件学报 ISSN 1000-9825, CODEN RUXUEW E-mail: jos@iscas.ac.cn
Journal of Software,2021,32(9):2901−2915 [doi: 10.13328/j.cnki.jos.005980] http://www.jos.org.cn
©中国科学院软件研究所版权所有. Tel: +86-10-62562563

∗
基于卷积神经网络的低嵌入率空域隐写分析

1,2
1,3
1,2
沈 军 , 廖 鑫 , 秦 拯 , 刘绪崇 3
1
(湖南大学 信息科学与工程学院,湖南 长沙 410012)
2 (大数据研究与应用湖南省重点实验室(湖南大学),湖南 长沙 410082)
3 (网络犯罪侦查湖南省重点实验室(湖南警察学院),湖南 长沙 410138)
通讯作者: 廖鑫, E-mail: xinliao@hnu.edu.cn

摘 要: 近年来,基于深度学习的空域隐写分析研究在高嵌入率下已经取得了较好的成果,但是对低嵌入率的检
测效果还不太理想.因此设计了一种卷积神经网络结构,使用 SRM 滤波器进行预处理来获取隐写噪声残差,采用 3
个卷积层并对卷积核大小进行合理设计,通过适当选择批量归一化操作和激活函数来提升网络的性能.实验结果表
明:与现有方法相比,所提出的网络结构对 WOW,S-UNIWARD 和 HILL 这 3 种常见的空域内容自适应隐写算法取得
了更好的检测效果,且在低嵌入率 0.2bpp,0.1bpp 和 0.05bpp 下的检测效果有非常明显的提升.还提出了逐步迁移
(step by step)的迁移学习方法,进一步提升低嵌入率条件下的隐写分析效果.
关键词: 隐写分析;卷积神经网络;低嵌入率;迁移学习
中图法分类号: TP309

中文引用格式: 沈军,廖鑫,秦拯,刘绪崇.基于卷积神经网络的低嵌入率空域隐写分析.软件学报,2021,32(9):2901−2915.
http://www.jos.org.cn/1000-9825/5980.htm
英文引用格式: Shen J, Liao X, Qin Z, Liu XC. Spatial steganalysis of low embedding rate based on convolutional neural network.
Ruan Jian Xue Bao/Journal of Software, 2021,32(9):2901−2915 (in Chinese). http://www.jos.org.cn/1000-9825/5980.htm

Spatial Steganalysis of Low Embedding Rate Based on Convolutional Neural Network
1,2
1,2
1,3
SHEN Jun , LIAO Xin , QIN Zheng , LIU Xu-Chong 3
1 (College of Computer Science and Electronic Engineering, Hunan University, Changsha 410012.China)
2 (Hunan Key Laboratory of Big Data Research and Application (Hunan University), Changsha 410082, China)
3 (Hunan Key Laboratory of Cybercrime Reconnaissance (Hunan Police Academy), Changsha 410138, China)
Abstract: In recent years, the research of spatial steganalysis based on deep learning has achieved sound results under high embedding
rate, but the detection performance under low embedding rate is still not ideal. Therefore, a convolutional neural network structure is
proposed, which uses the SRM filter for preprocessing to obtain implicit noise residuals, adopts three convolution layers and designs the
size of convolution kernel reasonably, and selects appropriate batch normalization operations and activation functions to improve the
network performance. The experimental results show that compared with the existing methods, the proposed network can achieve better
detection performance for WOW, S-UNIWARD, and HILL, three common adaptive steganographic algorithms in spatial domain, and

∗ 基金项目: 国家自然科学基金(61972142, 61402162, 61772191); 湖南省自然科学基金(2017JJ3040); 模式识别国家重点实验室
开放课题(201900017); 湖南省科技计划(2015TP1004, 2016JC2012); 网络犯罪侦查湖南省普通高校重点实验室开放课题(2017
WLFZZC001)
Foundation item: National Natural Science Foundation of China (61972142, 61402162, 61772191); Hunan Provincial Natural
Science Foundation of China (2017JJ3040); Open Project Program of the National Laboratory of Pattern Recognition (201900017);
Science and Technology Key Projects of Hunan Province (2015TP1004, 2016JC2012); Open Research Fund of Key Laboratory of
Network Crime Investigation of Hunan Provincial Colleges (2017WLFZZC001)
收稿时间: 2019-06-18; 修改时间: 2019-09-23; 采用时间: 2019-11-06; jos 在线出版时间: 2020-04-21