基于多预训练模型的在线隐写盲分析系统研究与实现

doi:10.11959/j.issn.2096-109x.2017.00164

网络与信息安全学报 ›› 2017, Vol. 3 ›› Issue (5): 32-37.doi: 10.11959/j.issn.2096-109x.2017.00164

基于多预训练模型的在线隐写盲分析系统研究与实现

袁亚飞¹,卢伟¹,冯丙文²,翁健²

¹ 中山大学数据科学与计算机学院，广东广州 510006
² 暨南大学信息科学技术学院，广东广州 510632

修回日期:2017-03-26 出版日期:2017-05-01 发布日期:2017-05-13
作者简介:袁亚飞（1991-），男，河南商丘人，中山大学硕士生，主要研究方向为多媒体信息安全与数字取证。|卢伟（1979-），男，河南南阳人，中山大学副教授、硕士生导师，主要研究方向为多媒体信息安全与数字取证。|冯丙文（1985-），男，山东东营人，博士，暨南大学讲师，主要研究方向为多媒体安全与数字取证。|翁健（1976-），男，广东茂名人，博士，暨南大学教授、博士生导师，主要研究方向为密码学与信息安全。
基金资助:
广东省科技发展专项基金资助项目(2016KZ010103);广东省自然科学基金资助项目(2016A030313350);中央高校基本科研业务费基金资助项目(16lgjc83);中山大学科技成果转化计划基金资助项目

Online universal steganalysis system based on multiple pre-trained model

Ya-fei YUAN¹,Wei LU¹,Bing-wen FENG²,Jian WENG²

¹ School of Data and Computer Science,Sun Yat-sen University,Guangzhou 510006,China
² College of Information Science and Technology,Jinan University,Guangzhou 510632,China

Revised:2017-03-26 Online:2017-05-01 Published:2017-05-13
Supported by:
The Special Funds for Science and Technology Development of Guangdong Province(2016KZ010103);The Natural Science Foundation of Guangdong Province(2016A030313350);The Fundamental Research Funds for the Central Universities(16lgjc83);The Scientific and Technological Achievements Transformation Plan of Sun Yat-sen University

摘要/Abstract

摘要：

在实际应用中，针对未知隐写算法的盲检测难度非常大，结合实际应用设计实现了一个在线盲检测系统。在SRM算法的基础上，简化特征提取算法，提高特征可用性和提取速率；使用多个预训练检测模型，并采用加权投票策略判定检测结果；设计实现一种3层系统架构，分布式后台更加灵活高效；为了进一步满足实际应用要求，引入多线程技术，加快检测速率。实验表明，单张图片的平均检测时间可达0.97 s，并且对多种未知隐写算法均具有良好的检测结果。

关键词: 数字图像隐写, 隐写分析, 多模型, 加权投票, 在线检测

Abstract:

In reality,universal blind steganalysis is still a sensitive issue.A universal online steganalysis system that could be used in practical application was proposed.With reducing the dimensions of SRM,it could improve availability and speed up feature extraction.Some effective pre-trained models and weighted voting strategy were used in this system with a B/S architecture,involving a higher speed.In addition,multithread technology was introduced.Experimental results demonstrate that high detection accuracy can be obtained and about 0.97 seconds for single detection with the system.

Key words: digital image steganography, steganalysis, multi-model, weighted voting, online detection

中图分类号:

TP393

袁亚飞,卢伟,冯丙文,翁健. 基于多预训练模型的在线隐写盲分析系统研究与实现[J]. 网络与信息安全学报, 2017, 3(5): 32-37.

Ya-fei YUAN,Wei LU,Bing-wen FENG,Jian WENG. Online universal steganalysis system based on multiple pre-trained model[J]. Chinese Journal of Network and Information Security, 2017, 3(5): 32-37.

图/表 6

表1

表2

表3

表4

表5

表6

参考文献 19

[1]	CHEDDAD A , CONDELL J , CURRAN K ,et al. Digital image steganography:survey and analysis of current methods[J]. Signal Processing, 2010,90(3): 727-752.
[2]	SHARP T , . An implementation of key-based digital signal stegano-graphy[C]// The International Workshop on Information Hiding. 2001: 13-26.
[3]	LUO W , HUANG F , HUANG J . Edge adaptive image steganography based on LSB matching revisited[J]. IEEE Transactions on Information Forensics and Security, 2010,5(2): 201-214.
[4]	PEVNY T , FILLER T , BAS P . Using high-dimensional image models to perform highly undetectable steganography[C]// The International Workshop on Information Hiding. 2010: 161-177.
[5]	HOLUB V , FRIDRICH J . Designing steganographic distortion using directional filters[C]// Information Forensics and Security(WIFS). 2012: 234-239.
[6]	SULLIVAN K , MADHOW U , CHANDRASEKARAN S ,et al. Steganalysis of spread spectrum data hiding exploiting cover memory[C]// The International Society for Optics and Photonics,Electronic Imaging. 2005: 38-46.
[7]	ZOU D , SHI Y Q , SU W ,et al. Steganalysis based on Markov model of thresholded prediction-error image[C]// IEEE International Conference on Multimedia and Expo. 2006: 1365-1368.
[8]	PEVNY T , BAS P , FRIDRICH J . Steganalysis by subtractive pixel adja-cency matrix[J]. IEEE Transactions on Information Forensics and Security, 2010,5(2): 215-224.
[9]	凌轶华, 蔡晓霞, 陈红 . 应用共生矩阵特征的改进自适应 JPEG 隐写[J]. 计算机应用研究, 2014(11): 3490-3493.
	LING Y H , CAI X X , CHEN H . Improved adaptive steganographic algorithm for JPEG images based on co-occurrence matrix feature[J]. Application Research of Computers, 2014(11): 3490-3493.
[10]	FRIDRICH J , KODOVSKY J , HOLUB V ,et al. Steganalysis of con-tent-adaptive steganography in spatial domain[C]// The International Workshop on Information Hiding. 2011: 102-117.
[11]	FRIDRICH J , KODOVSKY J . Rich models for steganalysis of digital images[J]. IEEE Transactions on Information Forensics and Security, 2012,7(3): 868-882.
[12]	CORTES C , VAPNIK V . Support-vector networks[J]. Machine Learning, 1995,20(3): 273-297.
[13]	LYU S , FARID H . Detecting hidden messages using higher-order statistics and support vector machines[C]// The International Workshop on Information Hiding. 2002: 340-354.
[14]	KODOVSKY J , FRIDRICH J , HOLUB V . Ensemble classifiers for stegana-lysis of digital media[J]. IEEE Transactions on Information Forensics and Security, 2012,7(2): 432-444.
[15]	DUDA R O , HART P E , STORK D G . Pattern classification[M]. New York: WileyPress, 1973.
[16]	陈世媛, 汤光明, 高瞻瞻 . 基于权值分配的隐写分析算法[J]. 计算机应用研究, 2016,(11): 3468-3471.
	CHEN S Y , TANG G M , GAO Z Z . Steganalysis method based on weight allocation[J]. Application Research of Computers, 2016(11): 3468-3471.
[17]	GOLJAN M , FRIDRICH J , HOLOTYAK T . New blind steganalysis and its implications[C]// The International Society for Optics and Photonics,Electronic Imaging. 2006: 1-13.
[18]	FRIDRICH J,KODOVSKY J , HOLUB V , et al . Breaking HUGO–the process discovery[C]// The International Workshop on Information Hiding. 2011: 85-101.
[19]	HOLUB V , FRIDRICH J , DENEMARK T . Universal distortion function for steganography in an arbitrary domain[J]. EURASIP Journal on Information Security, 2014(1): 1.

选取子特征算法	子特征维度
S1MINMAX24	325
S1MINMAX34	325
S1MINMAX54	325
S1SPAM14	338
S2MINMAX21	325
S2MINMAX32	325
S2MINMAX41	325
S2SPAM12	338
S3MINMAX24	325
S3MINMAX34	325
S3MINMAX54	325
S3SPAM14	338
S33MINMAX24	325
S33MINMAX41	325
S33SPAM14	338
S35SPAM11	338
S55MINMAX24	325
S55MINMAX41	325
S55SPAM14	338
SPAM	686
WAM	27
总计	6 966

特征提取算法	最小错误率	最大错误率	平均错误率特征维度
SRM	26.90%	28.90%	27.63%	34 671
SRMQ1	27.90%	30.00%	29.13%	12 753
Sub-SRM	27.80%	30.35%	29.26%	6 966

预训练模型	cover		stego
预训练模型	cover	隐写算法	bpp=0.25	bpp=0.5
		LSBM	5.6%	4.8%
Model_LSBM0.25	7.6%	EA	59.6%	16.4%
		HUGO	88.0%	59.4%
		LSBM	4.8%	4.4%
Model_LSBM0.5	9.0%	EA	57.6%	15.8%
		HUGO	84.2%	59.0%
		LSBM	23.6%	20.2%
Model_EA0.25	17.0%	EA	17.8%	7.8%
		HUGO	61.0%	25.8%
		LSBM	44.4%	41.4%
Model_EA0.5	7.6%	EA	45.4%	6.6%
		HUGO	85.2%	55.6%
		LSBM	4.8%	5.2%
Model_HUGO0.25	30.6%	EA	11.8%	2.8%
		HUGO	28.4%	11.8%
		LSBM	11.0%	10.4%
Model_HUGO0.5	20.8%	EA	16.6%	6.4%
		HUGO	51.8%	17.0%

预训练模型	cover		stego
预训练模型	cover	隐写算法	bpp=0.25	bpp=0.5
		LSBM	5.8%	6.4%
Model_HUGO0.15	38.0%	EA	10.4%	3.6%
		HUGO	25.4%	10.0%
		LSBM	4.6%	5.8%
Model_HUGO0.2	34.2%	EA	10.2%	3.0%
		HUGO	26.4%	11.0%
		LSBM	4.8%	5.2%
Model_HUGO0.25	30.6%	EA	11.8%	2.8%
		HUGO	28.4%	11.8%
		LSBM	6.0%	6.6%
Model_HUGO0.3	29.2%	EA	13.0%	3.2%
		HUGO	32.4%	12.2%
		LSBM	5.8%	6.4%
Model_HUGO0.35	35.0%	EA	14.0%	2.4%
		HUGO	36.6%	13.6%
		LSBM	7.6%	8.0%
Model_HUGO0.4	25.4%	EA	13.6%	4.4%
		HUGO	42.4%	13.2%
		LSBM	10.0%	8.4%
Model_HUGO0.45	22.0%	EA	16.8%	3.6%
		HUGO	47.8%	17.2%
		LSBM	11.0%	10.4%
Model_HUGO0.5			16.6%	6.4%
		HUGO	51.8%	17.0%

预训练模型		w_i
预训练模型	w_i1		w_i2
Model _HUGO0.15	0.6		1.5
Model _HUGO0.2	0.6		1.3
Model _HUGO0.45	1.2		0.8
Model _HUGO0.5	1.3		0.8

基于多预训练模型的在线隐写盲分析系统研究与实现

Online universal steganalysis system based on multiple pre-trained model

在线阅读

pdf下载

可视化

摘要/Abstract

引用本文

使用本文

图/表 6

参考文献 19

相关文章 15

Metrics

推荐阅读 0

[1]	任奎, 孟泉润, 闫守琨, 秦湛. 人工智能模型数据泄露的攻击与防御研究综述[J]. 网络与信息安全学报, 2021, 7(1): 1-10.
[2]	张青青, 汤红波, 游伟, 李英乐. 基于免疫算法的网络功能异构冗余部署方法[J]. 网络与信息安全学报, 2021, 7(1): 46-56.
[3]	张涛, 伍前红, 唐宗勋. 基于比特币区块链的隐蔽信息传输研究[J]. 网络与信息安全学报, 2021, 7(1): 84-92.
[4]	王滨, 陈靓, 钱亚冠, 郭艳凯, 邵琦琦, 王佳敏. 面向对抗样本攻击的移动目标防御[J]. 网络与信息安全学报, 2021, 7(1): 113-120.
[5]	苗春雨, 范渊, 李晖, 葛凯强, 张小孟. 基于移动Sink的WSN安全数据收集方法[J]. 网络与信息安全学报, 2021, 7(1): 121-129.
[6]	陈璐, 汤红波, 游伟, 柏溢. 移动边缘计算安全防御研究[J]. 网络与信息安全学报, 2021, 7(1): 130-142.
[7]	孙澄, 胡浩, 杨英杰, 张红旗. 融合宏观与微观的双层威胁分析模型[J]. 网络与信息安全学报, 2021, 7(1): 143-156.
[8]	张颖君,刘尚奇,杨牧,张海霞,黄克振. 基于日志的异常检测技术综述[J]. 网络与信息安全学报, 2020, 6(6): 1-12.
[9]	熊小兵,舒辉,康绯. 基于融合编译的软件多样化保护方法[J]. 网络与信息安全学报, 2020, 6(6): 13-24.
[10]	张鑫,羌卫中,吴月明,邹德清,金海. 基于卷积神经网络恶意安卓应用行为模式挖掘[J]. 网络与信息安全学报, 2020, 6(6): 35-44.
[11]	周天昱,申文博,杨男子,李金库,秦承刚,喻望. Docker组件间标准输入输出复制的DoS攻击分析[J]. 网络与信息安全学报, 2020, 6(6): 45-56.
[12]	尚菁菁,朱宇佳,刘庆云. 电子邮件安全扩展协议应用分析[J]. 网络与信息安全学报, 2020, 6(6): 69-79.
[13]	王标,刘兴洋,许卡,刘汪洋,王克楠,夏宇清. 政府开放数据的国家安全风险评估模型研究[J]. 网络与信息安全学报, 2020, 6(6): 80-87.
[14]	熊钢,葛雨玮,褚衍杰,曹卫权. 基于跨域协同的网络空间威胁预警模式[J]. 网络与信息安全学报, 2020, 6(6): 88-96.
[15]	吴奇,陈鸿昶. 低故障恢复开销的软件定义网络控制器布局算法[J]. 网络与信息安全学报, 2020, 6(6): 97-104.