Steganalysis of Variable Size Image Based on Efficient Feature Fusion

doi:10.3778/j.issn.1002-8331.2007-0497

Abstract

Abstract:

In order to improve the efficiency and accuracy of steganalysis, and to adapt to multi-dimensional input images, a variable size steganalysis model based on efficient feature fusion is proposed. In the preprocessing layer, the multi-dimensional convolution kernel initialized by the multi-order high pass filters of spatial rich model is added to the network learning to improve the convergence efficiency and detection performance of the model. In the feature extraction layer, based on the idea of feature fusion, two subnetworks composed of Ghost bottleneck layer, residual module and dense connection module are designed. Then, the output abstract steganographic features and nonlinear high-dimensional steganography features are fused to obtain the dependency information of steganographic features, which is conducive to enhance the feature expression ability of the model. The improved spatial pyramid pooling is used to adapt the variable size image samples and enrich the diversity of steganography features. Simulation results show that the model can correctly capture the key steganographic signals, and possesses high convergence efficiency. The detection accuracy of the WOW steganographic algorithm with embedding rate of 0.2 and 0.4 is 82.6% and 96.5%, respectively, and the detection accuracy of S-UNIWARD steganographic algorithm with embedding rate of 0.2 and 0.4 is 81.4% and 95.2%, which are significantly higher than that of SRM and YedroudjNet steganalysis model.

Key words: steganalysis, feature fusion, spatial pyramid, Ghost, residual module, dense connection module

摘要：

为提升隐写分析的效率和准确率，并适应多尺寸输入图像，提出一个基于高效特征融合的可变尺寸图像隐写分析模型。在预处理层中，将经空域富模型的多阶高通滤波器初始化的多尺寸卷积核加入网络学习中，以提升模型的收敛效率和检测性能；在特征提取层中，采用特征融合思想，设计两个由Ghost瓶颈层、残差模块、密集连接模块组成的子网络，并融合输出的抽象隐写语义特征和非线性的高维隐写特征，以获得隐写特征的依赖性信息，增强模型的特征表达能力；采用改良版空间金字塔池化以自适应可变尺寸的图像样本，并丰富隐写特征的多样性。经仿真分析可知，模型能正确捕获关键的隐写信号，具备较高的收敛效率，在嵌入率为0.2、0.4的WOW隐写算法的检测准确率分别为82.6%和96.5%，在嵌入率为0.2、0.4的S-UNIWARD隐写算法的检测准确率分别为81.4%和95.2%，显著高于SRM和YedroudjNet隐写分析模型。

关键词: 隐写分析, 特征融合, 空间金字塔, Ghost, 残差模块, 密集连接模块

XIAO Ruixue, FENG Yingwei, QU Jianping. Steganalysis of Variable Size Image Based on Efficient Feature Fusion[J]. Computer Engineering and Applications, 2021, 57(24): 126-134.

肖瑞雪，冯英伟，屈建萍. 结合高效特征融合的可变尺寸图像隐写分析[J]. 计算机工程与应用, 2021, 57(24): 126-134.

References

[1] CHANDRAMOULI R，LI G，MEMON N D.Adaptive steganography[C]//Proceedings of SPIE 4675，Security and Watermarking of Multimedia Contents IV，San Jose，California，United States，2002.
[2] FRIDRICH J，KODOVSKY J.Rich models for steganalysis of digital images[J].IEEE Transactions on Information Forensics and Security，2012，7（3）：868-882.
[3] IJJINA E P，MOHAN C K.Human action recognition based on recognition of linear patterns in action bank features using convolutional neural networks[C]//2014 13th International Conference on Machine Learning and Applications，Detroit，MI，2014：178-182.
[4] WANG Q，HUANG H.Learning of recurrent convolutional neural networks with applications in pattern recognition[C]//2017 36th Chinese Control Conference（CCC），Dalian，2017：4135-4139.
[5] WAN X，SONG H，LUO L，et al.Pattern recognition of partial discharge image based on one-dimensional convolutional neural network[C]//2018 Condition Monitoring and Diagnosis（CMD），Perth，WA，2018：1-4.
[6] SALOMON M，COUTURIER R，GUYEUX C，et al.Steganalysis via a convolutional neural network using large convolution filters for embedding process with same stego key：a deep learning approach for telemedicine[J].European Research in Telemedicine/La RechercheEuropéenne en Télémédecine，2017，6（2）.
[7] ZHONG K，FENG G，SHEN L，et al.Deep learning for steganalysis based on filter diversity selection[J].Science China Information Sciences，2018，61（12）：196-198.
[8] QIAN Y，DONG J，WANG W.et al.Feature learning for steganalysis using convolutional neural networks[J].Multimedia Tools and Applications，2018，77（2）：19633-19657.
[9] QIAN Y，DONG J，WANG W.et al.Deep learning for steganalysis via convolutional neural networks[C]//Media Watermarking，Security，and Forensics，San Francisco，California，United States，2015.
[10] XU G，WU H，SHI Y.Structural design of convolutional neural networks for steganalysis[J].IEEE Signal Processing Letters，2016，23（5）：708-712.
[11] YE J，NI J，YI Y.Deep learning hierarchical representations for image steganalysis[J].IEEE Transactions on Information Forensics and Security，2017，12（11）：2545-2557.
[12] YEDROUDJ M，COMBY F，CHAUMONT M.Yedrouj-net：an efficient CNN for spatial steganalysis[C]//ICASSP：International Conference on Acoustics，Calgary，Alberta，Canada，2018：2092-2096.
[13] BOROUMAND M，CHEN M，FRIDRICH J.Deep residual network for steganalysis of digital images[J].IEEE Transactions on Information Forensics and Security，2019，14（5）：1181-1193.
[14] WU S，ZHONG S，LIU Y.Deep residual learning for image steganalysis[J].Multimedia Tools and Applications，2018，77：10437-10453.
[15] OZCAN S，MUSTACOGLU A F.Transfer learning effects on image steganalysis with pre-trained deep residual neural network model[C]//2018 IEEE International Conference on Big Data（Big Data），Seattle，WA，USA，2018：2280-2287.
[16] WU S，ZHONG S，LIU Y.Residual convolution network based steganalysis with adaptive content suppression[C]//2017 IEEE International Conference on Multimedia and Expo（ICME），Hong Kong，2017：241-246.
[17] HE K，ZHANG X，REN S，et al.Spatial pyramid pooling in deep convolutional networks for visual recognition[J].IEEE Transactions on Pattern Analysis and Machine Intelligence，2015，37（9）：1904-1916.
[18] HAN K，WANG Y，TIAN Q，et al.GhostNet：more features from cheap operations[C]//Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition，2020：1580-1589.