改进卷积神经网络在图像隐写检测中的应用  

The Application of Improved CNN in Image Steganography

作　　者：何凤英 HE Fengying(Department of Mathematics&Statistics,Fuzhou University,Fuzhou,China,350002)

机构地区：[1]福州大学数学与统计学院,福州350002

出　　处：《福建电脑》2022年第9期1-6,共6页Journal of Fujian Computer

基　　金：福建省中青年教师教育科研项目(No.JAT210026)资助。

摘　　要：针对目前基于卷积神经网络的图像隐写分析模型在低嵌入率下检测效果不太理想的问题,本文提出一种新的卷积神经网络。网络的预处理层在参与网络学习的同时,保持残差的提取形式;卷积层通过改变卷积核的大小和步长来代替池化层,同时配合使用大、小尺寸的卷积核提取隐写特征,并采用优化后的激活函数和批量归一化来提高网络的收敛性能;最后融合三种用不同滤波器训练的网络模型得到分类结果。实验结果表明,所提网络模型对WOW、S-UNIWARD和HUGO三种隐写算法的检测率在多数情况下优于现有方法,尤其在低嵌入率下有较高的隐写分析准确率。Aiming to the problem that the detection rate of image steganalysis model based on convolutional neural network is not ideal under low embedding rate,this study proposes a new convolutional neural network.The preprocessing layer of the network participates in network learning while maintaining the extraction form of residual;The convolution layer cancels the pooling layer by changing the size and step size of the convolution kernel and extracts steganographic features by using large and small convolution kernels,then the optimized activation function and batch normalization are used to improve the convergence performance of the network;Finally,three network models trained with different filters are fused to obtain the classification results.Experiment results demonstrate that compared with the existing methods,the proposed network model can achieve better detection performance for WOW,S-UNIWARD and HUGO in most cases,especially in the case of low embedding rate.

关 键 词：隐写检测 卷积神经网络 低嵌入率 批量归一化 激活函数 

分 类 号：TP391[自动化与计算机技术—计算机应用技术]