基于新残差网络的图像隐写分析方法

卷积神经网络在隐写分析领域取得了一系列进展,但现有网络结构大多都是专用隐写分析,只针对某一类隐写算法有效.为了提高模型的泛化能力,提出了一种基于新残差网络的图像隐写分析算法.构建了残差分组融合网络结构(W-R2 N),采用分组融合的方式来提高提取多尺度特征的能力,增大每层网络的感受野范围,并且增加每组卷积的对角相关性.相对于Xu-Net和SRNET在S-UNIWARD嵌入率为0.4 bpp情况下隐写分析准确率分别提高了17.13％和0.81％.实验结果表明,相对于现有卷积神经网络,该模型泛化能力更好,并且能够有效提高隐写分析的准确率.

Image steganalysis method based on new residual network

Convolutional neural network has achieved a series of progress in steganalysis, but most of the existing network structures are dedicated steganalysis, which is only effective for a certain kind of steganalysis. In order to improve the model''s generality, this paper proposed a new residual network based on image steganalysis algorithm. It constructed a residual packet fusion network structure(W-R2N), which improved the ability of extracting multi-scale features by means of grouping convolution and fusion, increased the receptive field scope of each layer of network, and increased the convolution of diagonal correlations of each group. Compared with XU-NET and SRNET with S-UNIWARD embedding rate of 0.4 bpp, the accuracy of steganalysis was improved by 17.13% and 0.81% respectively. The experimental results show that this model has better generalization ability and can effectively improve the accuracy of steganalysis compared with the existing convolutional neural network.