隱寫術(Steganography)是將若干影像甚至音視訊等資料做為載體，再把大量機密資訊嵌入於其中以達成秘密通訊的效果，而隱寫分析(Steganalysis)則是偵測可能的載體以確認秘密資訊是否藏於其中。關於隱寫分析，以往多採用人工設計之特徵擷取，但需要耗費較多人力以及依賴相關研究經驗，近期則以深度學習技術為主，但也多使用指定的濾波器對待測資料進行人工預處理，無法達成完全的自動學習。本研究主要為影像隱寫分析，使用雙路徑卷積神經網路(Dual Path Networks, DPN)達成端到端(end-to-end)架構，以ResNet 擷取特徵，再以 DenseNet 提取更深層且細微的特徵，結合兩者的優勢，組成權值共享的雙通道區塊(DPN blocks)，並採用 ResNeXt 的分組卷積降低計算量，使用不同參數的雙通道區塊組合以利隱寫分析。SRNet 為目前隱寫分析模型中效果較為優異者，當中採用了 ResNet 作為特徵擷取，我們將其替換為雙通道區塊進行比較，偵測準確度有所提升，也證實了 DPN 有助於隱寫特徵的擷取。接著我們將整體架構改為以 DPN 為主，與以往的隱寫分析架構不同，並與這些架構比較以彰顯所提出架構的可行性。

關鍵字：隱寫分析、隱寫術、深度學習、卷積神經網路。

Steganography is a technique to embed a large amount of information in such carriers as images, audio, videos and even texts to achieve effective secret communications. On the other hand, steganalysis is the adversarial technique
aimed at determining whether the investigated carriers contain hidden information. In the field of steganalysis, heuristic features were usually adopted. Recently deep
learning techniques are often employed but most existing methods still use certain high-pass filters to apply pre-processing. In this research, we focus on image
steganalysis and adopt the dual path networks (DPN) to achieve an end-to-end architecture. The proposed scheme uses ResNet to extract features, and then employs DenseNet to extract deeper and smaller features. It combines the advantages of both networks to form a DPN blocks with shared weights. The scheme uses the group convolution to reduce the amount of computation. Finally,
dual path blocks with different parameters are tested to build suitable steganalysis architectures. SRNet, which uses ResNet, performs quite well in image steganalysis. We first replace its ResNet blocks with DPN blocks for comparison. The detection accuracy is improved and confirms that the structure using DPN is helpful to steganalysis. We then use DPN blocks to build our architecture and then compare the performance with the existing steganalysis architectures. Finally, we use the ALASKA II dataset to verify the feasibility of the proposed scheme.
Index Terms - Steganalysis, steganography, deep learning, convolutional neural networks.

[1]Yunpeng Chen, et al. “Dual Path Networks”, arXiv preprint arXiv:1707.01083, 2017.
[2]Kaiming He, et al. “Deep residual learning for image recognition”, Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition. 2016.
[3]Gao Huang, et al. “Densely Connected Convolutional Networks”, Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition. 2017.
[4]Mehdi Boroumand, et al. “Deep Residual Network for Steganalysis of Digital Images.”, Proceedings of the IEEE Transactions on Information Forensics and Security. 2019.
[5]S. M. Masud Karim, et al., “A new approach for LSB based image steganography using secret key”, 14th International Conference on Computer and Information Technology, 2011.
[6]D.G. Lowe, “Object recognition from local scale-invariant features,” Proceedings of the Seventh IEEE International Conference on Computer Vision: 1150–1157. 1999.
[7]Bin Li, et al., “A new cost function for spatial image steganography”, Proceedings of the IEEE International Conference on Image Processing (ICIP), 2014.
[8]Vojtěch Holub and Jessica Fridrich, “Designing steganographic distortion using directional filters”, Proceedings of the IEEE International Workshop on Information Forensics and Security (WIFS), 2012.
[9]Vojtech Holub, et al., “Universal distortion function for steganography in an arbitrary domain”, EURASIP Journal on Information Security, vol.2014, no. 1, Dec. 2014.
[10]Linjie Guo, et al., “An efficient JPEG steganographic scheme using uniform embedding”, Proceedings of the IEEE International Workshop on Information Forensics and Security (WIFS), 2012.
[11]Chang Wang and Jiangqun Ni, “An efficient JPEG steganographic scheme based on the block entropy of DCT coefficients”, Proceedings of the IEEE International Conference on Acoustics, Speech and Signal Processing, 2012.
[12]Corinna Cortes and Vladimir Vapnik, “Support-vector networks”, Machine Learning. 20 (3): 273–297., 1995.
[13]Jan Kodovsky, at el., “Ensemble Classifiers for Steganalysis of Digital Media.”, Proceedings of the IEEE Transactions on Information Forensics and Security, vol.7, 2012.
[14]Jessica Fridrich, at el., “Breaking HUGO—The Process Discovery.”, Information Hiding-international Conference, pp. 85-101, 2011.
[15]Jan Kodovský and Jessica Fridrich, “Steganalysis in High Dimensions: Fusing Classifiers Built on Random Subspaces.”, Proceedings of SPIE—The International Society for Optical Engineering, pp. 181-197, 2011
[16] Jessica Fridrich and Jan Kodovsky, “Rich Models for Steganalysis of Digital Images.”, Proceedings of the IEEE Transactions on Information Forensics and Security, 2012
[17]Vojtech Holub and Jessica Fridrich, “Random Projections of Residuals for Digital Image Steganalysis.”, Proceedings of the IEEE Transactions on Information Forensics and Security, 2013
[18]Tomáš Pevny et al., “Steganalysis by subtractive pixel adjacency matrix”, Proceedings of the IEEE Transactions on Information Forensics and Security, 2010.
[19]Vojtech Holub and Jessica Fridrich, “Random projections of residuals for digital image steganalysis.”, Proceedings of the IEEE Transactions on Information Forensics and Security, 2013.
[20]Tomas Pevny and Jessica Fridrich, “Merging Markov and DCT features for multi-class JPEG steganalysis.”, Security, Steganography, and Watermarking of Multimedia Contents IX., 2007.
[21]Jan Kodovský and Jessica Fridrich, “Steganalysis of JPEG images using rich models”, Media Watermarking, Security, and Forensics, 2012.
[22]Vojtěch Holub and Jessica Fridrich, “Low-Complexity Features for JPEG Steganalysis Using Undecimated DCT.”, Proceedings of the IEEE Transactions on Information Forensics and Security, 2015.
[23]Vojtěch Holub and Jessica Fridrich, “Phase-aware projection model for steganalysis of JPEG images.”, Media Watermarking, Security, and Forensics, 2015.
[24]Xiaofeng Song, et al., “Steganalysis of Adaptive JPEG Steganography Using 2D Gabor Filters”, Proceedings of the 3rd ACM Workshop on Information Hiding and Multimedia Security, 2015.
[25]Tomas Denemark, et al., “Selection-channel-aware rich model for Steganalysis of digital images”, Proceedings of the IEEE Transactions on Information Forensics and Security, 2014.
[26]Weixuan Tang, et al., “Adaptive steganalysis against WOW embedding algorithm.”, Proceedings of the 2nd ACM workshop on Information hiding and multimedia security, pp. 91-96, 2014.
[27]Yinlong Qian, et al., “Deep learning for steganalysis via convolutional neural networks.”, Media Watermarking, Security, and Forensics 2015, 2015.
[28]Guanshuo Xu, et al., “Structural Design of Convolutional Neural Networks for Steganalysis.”, Proceedings of the IEEE Signal Processing Letters, 2016.
[29]Jian Ye, et al., “Deep Learning Hierarchical Representations for Image Steganalysis.”, Proceedings of the IEEE Transactions on Information Forensics and Security, 2017.
[30]Guanshuo Xu, “Deep Convolutional Neural Network to Detect J-UNIWARD.”, Proceedings of the 5th ACM Workshop on Information Hiding and Multimedia Security, 2017
[31]Mehdi Yedroudj, et al., “Yedrouj-Net: An efficient CNN for spatial steganalysis.”, International Conference on Acoustics, Speech and Signal Processing, 2018.
[32]Ru Zhang, et al., “Efficient feature learning and multi-size image steganalysis based on CNN.”, arXiv preprint arXiv:1807.11428, 2018.
[33]Christian Szegedy, et al., “Going Deeper With Convolutions.”, Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition (CVPR), 2015.
[34]Christian Szegedy, et al., “Inception-v4, Inception-ResNet and the Impact of Residual Connections on Learning.”, AAAI Publications, Thirty-First AAAI Conference on Artificial Intelligence, 2017.
[35]Rohollah Soltani and Hui Jiang, “Higher order recurrent neural networks”. arXiv preprint arXiv:1605.00064, 2016.
[36]Saining Xie, et al., “Aggregated Residual Transformations for Deep Neural Networks.”, arXiv:1611.05431v2, 2017.
[37]Mo Chen, et al., “JPEG-phaseaware convolutional neural network for steganalysis of JPEG images.”, in Proc. 5th ACM Workshop Inf. Hiding Multimedia Secur., 2017.
[38]Tomas Denemark, et al., “Selection-channel-aware rich model for Steganalysis of digital images.”, 2014 IEEE International Workshop on Information Forensics and Security (WIFS), 2014.
[39]Tomáš Denemark Denemark, et al., “Steganalysis Features for Content-Adaptive JPEG Steganography.”, IEEE Transactions on Information Forensics and Security, 2016.
[40]Diganta Misra, “Mish: A Self Regularized Non-Monotonic Neural Activation Function.”, arXiv:1908.08681, 2019.
[41] Patrick Bas, et al., “‘Break our steganographic system’: The ins and outs of organizing BOSS”, Proc. 13th Int. Conf. Inf. Hiding in Lecture Notes in Computer Science, vol. 6958, pp. 59–70., 2011
[42]Prajit Ramachandran, et al., “Searching for Activation Functions”, arXiv:1710.05941, 2017.
[43]Diederik P. Kingma and Jimmy Ba, “Adam: A method for stochastic optimization”, arXiv:1412.6980, 2014.
[44]Kaiming He, et al., “Delving Deep into Rectifiers: Surpassing Human-Level Performance on ImageNet Classification”, Proceedings of the IEEE International Conference on Computer Vision (ICCV), 2015.
[45]Dan Hendryck and Kevin Gimpel, “GAUSSIAN ERROR LINEAR UNITS (GELUS)”, arXiv:1606.08415v4, 2020.
[46]P. Bas and T. Furon., “BOWS-2.”, http://bows2.ec-lille.fr, Jul. 2007.
[47]https://alaska.utt.fr/#timeline
[48]Rémi Cogranne, et al., “Steganography by Minimizing Statistical Detectability: The cases of JPEG and Color Images.”, ACM Information Hiding and MultiMedia Security, Jun. 2020.
[49]Linjie Guo, et al., “Using statistical image model for JPEG steganography: Uniform embedding revisited.”, IEEE Transactions on Information Forensics and Security, vol. 10, no. 12, pp. 2669–2680, 2015.