摘要
数字图像隐写将秘密信息嵌入到图像的冗余部分,并通过公开信道(如E-mail、社交平台等)传输以实现隐蔽通信.近年来,主流的数字图像隐写技术往往通过设计多样化的对称/非对称失真函数,将信息自适应嵌入到图像纹理复杂区域,以提升方法的抗统计检测性能.相比对称失真函数,最新的基于图像恢复或滤波原理的非对称失真函数能更好地指示信息嵌入时的修改方向.然而,现有基于图像恢复的非对称失真隐写往往耗时较长,基于滤波的非对称失真隐写抗检测能力还有待提升.为此,本文首先分析指出现有基于滤波的非对称失真隐写会弱化图像中物体的边缘与轮廓,降低边信息图像的质量;然后,提出超像素滤波的概念及算法,将描绘物体边缘与轮廓的超像素加入到滤波过程中,以提高构造边信息图像的质量;接着,给出一种超像素去噪算法,避免过小的超像素成为图像的“干扰噪声”,以更好地描绘物体的边缘与轮廓;最后,利用生成的边信息图像,结合现有非对称失真函数设计策略,得到能够更准确指示信息嵌入修改方向的非对称失真函数,进而给出基于超像素滤波的非对称失真隐写方法.基于常用的Bossbase-1.01、UCID、BOWS2、IStego100K等多个图像库,在0.1~0.5 bpnzAC的嵌入率、65~95的JPEG压缩质量因子下进行了大量实验,结果表明:与现有基于滤波构造边信息图像的典型隐写方法相比,本文方法所得边信息图像的峰值信噪比平均提高1.05 dB、结构相似度平均提高0.04,验证了超像素滤波的优势;与J-UNIWARD、UED、UERD等对称失真隐写典型方法相比,所提方法的抗统计检测性能分别平均提升7.6%、4.2%、6.0%;与现有基于图像恢复的非对称失真隐写典型方法相比,所提方法生成边信息图像的耗时降低约87%;与现有最新的基于均值滤波的非对称失真隐写方法相比,抗统计检测性能提升了0.3%~12%.
Digital image steganography embeds secret messages into the redundant part of an image and transmits the image through open channels(such as E-mail,social platforms,etc.)to realize secure covert communication.In recent years,the dominant research direction in digital image steganography has been adaptive steganography.It can adaptively embed messages into the complex texture regions of an image by designing diversified distortion functions to continuously improve the anti-statistical detection performance of a method.Distortion functions are divided into the symmetric distortion function and the asymmetric distortion function according to whether the addition or subtraction distortion of a position is the same.Compared with the symmetric distortion function,the asymmetric distortion function can better guide the modification direction of the message embedding with the help of side-information,and thus has more advantages in anti-statistical detection performance.In the existing asymmetric distortion steganographic methods,there are two main design directions:the method based on image restoration and the method based on filtering.However,image restoration based methods usually take a long time,while the anti-detection ability of filtering based methods still needs to be improved.To address these issues,this manuscript firstly analyzes and points out the problems of the existing filtering based asymmetric distortion steganography:such methods weaken the contours and edges of objects in an image to a certain extent,and thus reduce the quality of a side-information image.Then,to remedy this defect,the concept and algorithm of superpixel filtering are proposed,and the superpixel depicting the edges and contours of objects are added into a filtering process.In this way,the negative influence of filtering on contour and edge can be reduced,and the quality of a side information image can be improved.Next,to avoid tiny superpixels becoming“interference noises”of an image,the superpixel denoising algorithm is given.It
作者
朱利妍
罗向阳
张祎
王金伟
卢伟
刘粉林
ZHU Li-Yan;LUO Xiang-Yang;ZHANG Yi;WANG Jin-Wei;LU Wei;LIU Fen-Lin(School of Cyberspace Security,PLA Strategic Support Force Information Engineering University,Zhengzhou 450001;Key Laboratory of Cyberspace Situation Awareness,Zhengzhou 450001;Henan Police College,Zhengzhou 450001;School of Computer Science,Nanjing University of Information Science and Technology,Nanjing 210044;School of Computer Science and Engineering,Sun Yat-sen University,Guangzhou 510006)
出处
《计算机学报》
EI
CAS
CSCD
北大核心
2023年第7期1473-1493,共21页
Chinese Journal of Computers
基金
国家自然科学基金(62172435,62202495,62261160653,62102101)
中原科技创新领军人才项目(214200510019)
阿里巴巴创新研究计划资助。
关键词
隐写
非对称失真
超像素滤波
边信息
抗统计检测
steganography
asymmetric distortion
superpixel filtering
side information
anti-statistical detection
