摘要
针对目前图像隐写分析准确率较低的问题,构建一个基于多层感知卷积层的卷积神经网络隐写分析模型。使用多层感知卷积层代替传统的线性卷积层,提高模型的非线性能力,提取载体/隐写图像更抽象的特征。采用全局平均池化层代替全连接层,以减少网络的参数并提高模型的训练效率。实验结果表明,相比传统的图像隐写分析算法和现有的卷积神经网络隐写分析模型,该模型能够有效提高隐写分析的检测准确率,对S-UNIWARD嵌入算法的隐写分析检测准确率达到90. 87%。
Aiming at the situation that the accuracy of image steganalysis is not high,a steganalysis model of Convolutional Neural Network(CNN)based on Multi-layer percepual convolution layer(Mlpconv)is constructed.The model uses multi-layer perceptual convolution layer instead of the traditional linear convolution layer to improve the nonlinear capability of the model and extract more abstract features of the carrier/steganographic image;the global average pooling layer is used to instead of the fully connection layer,which effectively reduces the network parameters and improves the training efficiency of the model.Experimental results show that compared with the traditional image steganalysis algorithm and the existing steganalysis model of convolutional neural network,the model can effectively improve the detection effect of steganalysis,and the accuracy of steganalysis detection of S-UNIWARD embedding algorithm reaches 90.87%.
作者
高培贤
魏立线
刘佳
刘明明
GAO Peixian;WEI Lixian;LIU Jia;LIU Mingming(Key Laboratory of Network and Information Security under the Chinese Armed Police Force Engineering University of the Chinese Armed Police Force,Xi’an 710086,China;College of Cryptography,Engineering University of the Chinese Armed Police Force,Xi’an 710086,China)
出处
《计算机工程》
CAS
CSCD
北大核心
2018年第10期309-313,共5页
Computer Engineering
基金
国家自然科学基金(61403417)
关键词
隐写分析
卷积神经网络
多层感知卷积层
池化层
全连接层
steganalysis
Convolutional Neural Network(CNN)
Multi-layer perceptual convolution layer(Mlpconv)
pool layer
fully connected layer
