摘要
为了进一步提升基于深度学习的图像隐写分析算法的检测性能,结合图像隐写算法的嵌入特点提出了一种融合高通滤波和视觉注意力模块的深度残差图像隐写分析新算法。新算法首先对输入图像进行高通滤波,增强信噪比得到噪声残差图像,再将噪声残差图像放入融合了视觉注意力模块的深度残差网络中进行检测分类。高通滤波和注意力模块强制网络关注隐写图像中的高频信息部分,提高了秘密信息的特征表达,从而增强网络的检测能力。实验结果表明,新算法的检测性能超越了目前最优的传统图像隐写分析算法和基于深度学习的主流算法SRNet。
This research aims to improve the detection of the existing image steganalysis algorithms based on deep learning,based on the embedding characteristics of image steganography algorithms,a new image steganalysis algorithm embedded by a deep residual network,combining high-pass filtering and visual attention module is proposed.The new algorithm first performs high-pass filtering on the input image,enhances the signal-to-noise ratio to obtain a noise residual image,and then puts the residual image into a deep residual network embedded with some visual attention modules for image steganalysis detection and classification.The high-pass filtering and attention modules force the network to pay attention to the high-frequency information areas in the steganographic image,which improves the feature representation of secret information,thereby enhancing the detection ability of the network.Experimental results show that the detection performance of the new algorithm surpasses the current best traditional image steganalysis algorithm and the mainstream deep learning-based SRNet algorithm.
作者
贺丽莎
李清光
HE Li-sha;LI Qing-guang(School of Computer, Electronics and Information, Guangxi University, Nanning 530004, China;Guangxi Key Laboratory of Multimedia Communication and Network Technology, Nanning 530004, China)
出处
《广西大学学报(自然科学版)》
CAS
北大核心
2021年第5期1396-1403,共8页
Journal of Guangxi University(Natural Science Edition)
基金
国家自然科学基金资助项目(61871138)
广西自然科学基金资助项目(2017GXNSFAA198371)
广西大学科研基金资助项目(XGZ170107)。
关键词
隐写分析
深度学习
深度残差网络
注意力模块
高通滤波
image steganalysis
deep learning
deep residual network
attention module
high-pass filtering