An Efficient JPEG Steganalysis Model Based on Deep Learning
Convolutional neural networks (CNN) have gained an overwhelming advantage in many domains of pattern recognition. CNN’s excellent data learning ability and automatic feature extraction ability are urgently needed for image steganalysis research. However, the application of CNN in image steganalysis is still in its infancy, especially in the field of JPEG steganalysis. This paper presents an efficient CNN-based JPEG steganographic analysis model which is called JPEGCNN. According to the pixel neighborhood model, JPEGCNN calculates the pixel residual as a network input with a 3 × 3 kernel function. In this way, JPEGCNN not only solves the problem that direct analysis of DCT coefficients is greatly affected by image content, but also solves the problem that larger kernel functions such as 5 × 5 do not effectively capture neighborhood correlation changes. Compared with the JPEG steganographic analysis model HCNN proposed by the predecessors, JPEGCNN is a lightweight structure. The JPEGCNN training parameters are about 60,000, and the number of parameters is much lower than the number of parameters of the HCNN. At the same time of structural simplification, the simulation results show that JPEGCNN still maintains accuracy close to HCNN.
KeywordsSteganalysis Convolutional neural network Transform domain
This work is supported by the National Key R&D Program of China (No. 2017YFB0802703) and Open Foundation of Guizhou Provincial Key Laboratory of Public Big Data (No. 2018BDKFJJ014).
- 1.Pevný, T., Filler, T., Bas, P.: Using High-Dimensional Image Models to Perform Highly Undetectable Steganography. Springer, Berlin (2010)Google Scholar
- 2.Holub, V., Fridrich, J.: Designing steganographic distortion using directional filters. In: IEEE International Workshop on Information Forensics and Security (WIFS), vol. 2, pp. 234–239 (2012). https://doi.org/10.1109/WIFS.2012.6412655
- 3.Holub, V., Fridrich, J.: Digital image steganography using universal distortion. In: Proceedings of the First ACM Workshop on Information Hiding and Multimedia Security, Montpellier, 17–19 June 2013, pp. 59–68 (2013). https://doi.org/10.1145/2482513.2482514
- 6.Fridrich, J., et al.: Breaking HUGO—the process discovery. In: Information Hiding-international Conference, vol. 6958, pp. 85–101 (2011). https://doi.org/10.1007/978-3-642-24178-9_7
- 7.Jan, K., Fridrich, J.: Steganalysis in high dimensions: fusing classifiers built on random subspaces. In: Proceedings of SPIE—The International Society for Optical Engineering, vol. 7880, pp. 181–197 (2011)Google Scholar
- 10.Qian, Y., Dong, J., Wang, W., Tan, T.: Deep learning for steganalysis via convolutional neural networks. In: Proceedings of the SPIE, Media Watermarking, Security, and Forensics 2015, pp. 94 090 J–1–94 090 J–10 (2015)Google Scholar
- 11.Pibre, L., Jérôme, P., Ienco, D., Chaumont, M.: Deep learning is a good steganalysis tool when embedding key is reused for different images, even if there is a cover source-mismatch. In: Proceedings of the Media Watermarking, Security, and Forensics, Part of IS&T International Symposium on Electronic Imaging (EI 2016), February 2016Google Scholar
- 16.Holub, V., Fridrich, J.: Phase-aware projection model for steganalysis of JPEG images. In: Proceedings of the SPIE, Electronic Imaging, Media Watermarking, Security, and Forensics XVII, vol. 9409 (2015)Google Scholar
- 17.Bas, P., Filler, T., Pevný, T.: Break our steganographic system: the ins and outs of organizing boss. In: Information Hiding, pp. 59–70. Springer (2011)Google Scholar
- 19.Zeiler, M.D.: ADADELTA: An adaptive learning rate method. https://arXiv:1212.5701 (2012)
- 20.Glorot, X., Bengio, Y.: Understanding the difficulty of training deep feedforward neural networks. In: Proceedings of the Aistats, vol. 9, pp. 249–256 (2016)Google Scholar