Steganalysis model for detecting and recovering stego images
No Thumbnail Available
Date
2026
Authors
Journal Title
Journal ISSN
Volume Title
Publisher
Abstract
The National Information Technology Authority of Uganda (NITA-U) lacks effective steganalysis capabilities, leaving government systems vulnerable to covert data exfiltration and hidden communication channels exploited by malicious actors. While encryption secures message content, it does not conceal the existence of communication—a limitation that steganography overcomes by hiding information within innocuous digital media.
This study designed, implemented, and experimentally evaluated a steganalysis model to detect, decipher, and recover hidden information from digital image files within the NITA-U context. Using an experimental design, the Least Significant Bit (LSB) technique was implemented in Python with OpenCV and Pillow. A dataset of 45 images was assembled from USC-SIPI, BOSSBase v1.01, and NITA-U operations. Evaluation metrics included PSNR, SSIM, MSE, chi-square analysis, classification metrics, and three feature detection methods (Shi-Tomasi, ORB, Harris).
Results showed successful hiding and retrieval of text and image payloads without quality loss. The tool achieved PSNR values of 52.34–54.18 dB (exceeding the 40 dB threshold) and SSIM values of 0.9978–0.9984, confirming imperceptibility. Chi-square
statistics (2.14, 1.87) fell below the critical threshold of 3.84, confirming statistical undetectability.
No feature detection method distinguished stego images from cover images at significant levels (p = 0.68, 0.72; 94.7% match rate). The proposed model achieved 93.3% detection accuracy (F1 = 0.903) and significantly outperformed Steghide, OpenPuff, and
F5 (ANOVA: F = 16.98, p < 0.001).
Description
Undergraduate research report
Keywords
Citation
Nabirye, B. (2026) Steganalysis model for detecting and recovering stego images. Research report. Busitema University