Image Steganography Analysis Based on Text‑Payload Embedding with Machine Learning Techniques
Traditional cybersecurity systems often fail to detect data exfiltration carried out through steganography, as they primarily focus on other types of threats. Therefore, it is necessary to add strong and flexible steganalysis capability as an additional component to these systems. In this study, a lightweight, modular, and cost-efficient machine learning-based steganalysis framework has been developed. The proposed approach employs a two-stage architecture in which image features are extracted and filtered, which is followed by the detection of hidden data through various algorithms. In the experimental phase, the ALASKA-2 dataset was used, and stego images were generated with embedding rates of 10% and 40% using the F5 (frequency domain) and OpenStego (spatial domain) algorithms. Model performance was evaluated using accuracy, precision, recall, and F1-score. The results revealed that at high embedding rates, the CatBoost algorithm achieved 97% detection in the F5 scenario, while XGBoost achieved 94% in the OpenStego scenario. The findings indicate that the proposed framework can serve as a lightweight, modular, and interpretable machine learning-based detection layer for controlled text-payload steganalysis scenarios in both spatial and JPEG frequency domains.
Authors
- Baha Şen (ORCID: https://orcid.org/0000-0003-3577-2548)
- Yalçın Girgin (ORCID: https://orcid.org/0009-0009-4422-3365)
Institutions
- National Defense University (US)
- Milli Savunma Üniversitesi (TR)
- National Defense University (AZ)
Publication Details
- Journal
- Black Sea Journal of Engineering and Science
- Published
- 2026-11-15
- DOI
- https://doi.org/10.34248/bsengineering.1847042
- Primary Topic
- Advanced Steganography and Watermarking Techniques
- Type
- article
- Field-Weighted Citation Impact
- 0.00