Computer Engineering and Applications ›› 2018, Vol. 54 ›› Issue (15): 74-77.DOI: 10.3778/j.issn.1002-8331.1711-0210

Previous Articles     Next Articles

Image steganalysis based on densely connected convolutional networks

GAO Peixian1,2, WEI Lixian1,2, LIU Jia1,2, LIU Mingming1,2   

  1. 1.Key Laboratory for Network and Information Security of Chinese Armed Police Force, Engineering University of Chinese Armed Police Force, Xi’an 710086, China
    2.Department of Electronic Technology, Engineering University of Chinese Armed Police Force, Xi’an 710086, China
  • Online:2018-08-01 Published:2018-07-26

基于密集连接网络的图像隐写分析

高培贤1,2,魏立线1,2,刘  佳1,2,刘明明1,2   

  1. 1.武警工程大学 网络与信息安全武警部队重点实验室,西安 710086
    2.武警工程大学 电子技术系,西安 710086

Abstract: In view of the shortcomings of the traditional steganalysis technology for the feature set demanding higher and higher requirements, a Steganalysis-Densely Connected Convolutional Network(S-DCCN) model is constructed for image steganalysis to avoid the feature extraction and improves the efficiency of steganalysis. Firstly, a High-Pass Filter(HPF) filter is added in front of the network layer to speed up model training. And the filtered image enters two convolution layers for feature extraction. After the convolution layer using five groups of dense connection module to solve the network deepen gradient disappear dense connection between modules to control the network the width of the transition layer. Experimental results show that compared with the traditional image steganalysis algorithm and the convolutional neural network technology, the proposed model can effectively improve the accuracy and generalization performance of steganalysis.

Key words: steganalysis, neural networks, dense connectivity, gradient vanishing

摘要: 针对目前传统的隐写分析技术对特征集要求越来越高的问题,构建了一个密集连接网络模型(Steganalysis-Densely Connected Convolutional Networks,S-DCCN)进行图像隐写分析,避免了人工提取特征,提高了隐写分析效率。首先,在网络层之前添加高通滤波层(HPF)进行滤波,加快模型训练速度。经过滤波后的图像进入两层卷积层进行特征提取,在卷积层之后使用了5组密集连接模块来解决网络加深带来的梯度消失问题,密集连接模块之间通过过度层来控制整个网络的宽度。实验结果表明,相比传统的图像隐写分析算法和卷积神经网络技术,该模型有效提高了隐写分析的准确率和泛化性能。

关键词: 隐写分析, 神经网络, 密集连接, 梯度消失