Studying the Resilience of Steganography Algorithms to Detection by Neural Networks
DOI:
https://doi.org/10.47451/tec2025-10-01Keywords:
steganography, steganalysis, color space, deep learning modelsAbstract
This study analyzes the resilience of steganography algorithms used in the spatial domain against detection using steganalysis deep learning models. We evaluate the impact of geometric transformations on the robustness against detection. The study investigates the effect of image transformations on the embedded message integrity and, more importantly, on the detectability of the algorithms by different models. Proven CNN models, as EfficientNet and SRNet with a Convolutional Block Attention Module, were used to compare detectability after different transformations. As expected, the embedded messages were significantly corrupted, while the models were still able to identify their presence. Introducing noise before embedding reduced detectability and increased robustness of steganographic algorithms. Alongside common LSB and modern S-UNIWARD, evaluating a spatial color space embedding algorithm, previously unknown to the model, significantly decreased accuracy. Due to the complex impact of the Color Space algorithm on uniform areas and smooth color transitions, we observe a higher number of false positives after fine-tuning models. Nevertheless, accuracy and generalization were increased to expected levels consistent with another modern research. The results underscore the need to focus on developing models that can withstand real-world image alterations, as well as improve detection capabilities to keep up with unseen stenographic methods. In practice, this will help practitioners select detection models best suited for operational environments and support future advancements in detection model development and design.
Downloads
References
AbdelRaouf, A. (2021). A new data hiding approach for image steganography based on visual color sensitivity. Multimedia Tools and Applications, 80. https://doi.org/10.1007/s11042-020-10224-w
Alrusaini, O. A. (2025). Deep learning for steganalysis: Evaluating model robustness against image transformations. Frontiers in Artificial Intelligence, 8, 1532895. https://doi.org/10.3389/frai.2025.1532895
Apau, R., Asante, M., Twum, F., Ben Hayfron-Acquah, J., & Peasah, K. O. (2024). Image steganography techniques for resisting statistical steganalysis attacks: A systematic literature review. PLOS ONE, 19(9), e0308807. https://doi.org/10.1371/journal.pone.0308807
Bradski, G. (2000). The OpenCV library. Dr. Dobb’s Journal of Software Tools.
Chaumont, M. (2020). Deep learning in steganography and steganalysis. In M. Hassaballah (Ed.), Digital media steganography: Principles, algorithms, advances (pp. 321–349). Elsevier. https://doi.org/10.1016/B978-0-12-819438-6.00022-0
Clark, A. (2015). Pillow (PIL fork) documentation. Read the Docs. https://pillow.readthedocs.io/
Deng, J., Dong, W., Socher, R., Li, L.-J., Li, K., & Fei-Fei, L. (2009). ImageNet: A large-scale hierarchical image database. 2009 IEEE Conference on Computer Vision and Pattern Recognition, 248–255. https://doi.org/10.1109/CVPR.2009.5206848
Hasilin, D. L., & Zhuravel, I. M. (2025). Analysis of the robustness of steganographic algorithms against geometric attacks and compression. Modern Information Protection, (2). https://doi.org/10.31673/2409-7292.2025.020965
Holub, V., Fridrich, J., & Denemark, T. (2014). Universal distortion function for steganography in an arbitrary domain. EURASIP Journal on Information Security, 2014(1), 1–13. https://doi.org/10.1186/1687-417X-2014-1
Hu, X., Ni, J., & Shi, Y. Q. (2018). Efficient JPEG steganography using domain transformation of embedding entropy. IEEE Signal Processing Letters, 25(6), 859–863. https://doi.org/10.1109/LSP.2018.2818674
Lin, T.-Y., Maire, M., Belongie, S., Hays, J., Perona, P., Ramanan, D., Dollár, P., & Zitnick, C. L. (2014). Microsoft COCO: Common objects in context. European Conference on Computer Vision (ECCV), 740–755. https://doi.org/10.1007/978-3-319-10602-1_48
Margalikas, E., & Ramanauskaitė, S. (2019). Image steganography based on color palette transformation in color space. EURASIP Journal on Image and Video Processing, 2019(1), 84. https://doi.org/10.1186/s13640-019-0484-x
Ni, J., Hu, X., & Shi, Y. Q. (2014). Universal distortion function for steganography in an arbitrary domain. EURASIP Journal on Information Security, 2014(1), 1. https://doi.org/10.1186/1687-417X-2014-1
Paszke, A., Gross, S., Massa, F. et al. (2019). PyTorch: An imperative style, high-performance deep learning library. Advances in Neural Information Processing Systems, 32, 8024–8035.
Tan, M., & Le, Q. V. (2019). EfficientNet: Rethinking model scaling for convolutional neural networks. Proceedings of the 36th International Conference on Machine Learning (ICML 2019), 6105–6114. PMLR. https://arxiv.org/abs/1905.11946
Timofte, R., Agustsson, E., Van Gool, L., Yang, M., & Zhang, L. (2017). NTIRE 2017 Challenge on Single Image Super-Resolution: Dataset and Study. arXiv preprint. arXiv:1708.02567. https://arxiv.org/abs/1708.02567
Woo, S., Park, J., Lee, J.-Y., & Kweon, I. S. (2018). CBAM: Convolutional block attention module. arXiv. https://doi.org/10.48550/arXiv.1807.06521
Published
Issue
Section
License
Copyright (c) 2025 European Scientific e-Journal
This work is licensed under a Creative Commons Attribution 4.0 International License.
The European Scientific e-Journal (ESEJ) is an open access journal. Articles are available free of charge as PDF files on the website of the European Institute for Innovation Development. PDF files can be previewed with Acrobat Reader from www.adobe.com.
All articles of the “Tuculart Student Scientific” are published under a Creative Commons Attribution 4.0 Generic (CC BY 4.0) International license.
According to the Creative Commons Attribution 4.0 Generic (CC BY 4.0) International license, the users are free to Share — copy and redistribute the material in any medium or format for any purpose, even commercially (the licensor cannot revoke these freedoms as long as you follow the license terms).
Under the following terms:
- Attribution — You must give appropriate credit, provide a link to the license, and indicate if changes were made. You may do so in any reasonable manner, but not in any way that suggests the licensor endorses you or your use.
- No additional restrictions — You may not apply legal terms or technological measures that legally restrict others from doing anything the license permits.
