Steganalysis Model for Detecting and Recovering Stego Images.
| dc.contributor.author | Okiror, Charles. | |
| dc.date.accessioned | 2026-09-30T14:43:06Z | |
| dc.date.available | 2026-09-30T14:43:06Z | |
| dc.date.issued | 2026-09 | |
| dc.description | Postgraduate Thesis/Dissertation | |
| dc.description.abstract | Steganographic concealment hides the existence of information rather than its content, allowing hidden messages to evade conventional content-inspection controls. The National Information Technology Authority of Uganda (NITA-U) receives image evidence during digital forensic investigations but lacks a validated procedure for determining whether images contain concealed payloads, resulting in suspected cases being treated as inconclusive. This study therefore designed, implemented, and experimentally evaluated an integrated steganalysis workflow for detecting and recovering hidden information in digital images. An experimental design was employed using sequential single-bit Least Significant Bit (LSB) embedding and recovery to generate s with known ground truth. A threshold-based detector combining chi-square analysis and feature-preservation measures was calibrated using a disjoint subset. Evaluation involved 45 PNG images from USC-SIPI, BOSSBase v1.01, and NITA-U operational systems. Performance was assessed using PSNR, SSIM, MSE, chi-square statistics, classification metrics, and feature analysis, with comparisons against Steghide, OpenPuff, and F5. The results showed successful embedding and recovery of text and image payloads with minimal perceptible image degradation. PSNR ranged from 52.34–54.18dB, SSIM from 0.9978– 0.9984, and payload recovery accuracy exceeded 99.9%. Chi-square statistics ranged from 1.87– 2.14, below the critical value of 3.84 at α = 0.05. Shi-Tomasi, ORB, and Harris feature analysis did not significantly distinguish s from s (p = 0.68 and p = 0.72; 94.7% ORB match rate), indicating that gradient-based inspection provided no useful signal at the evaluated embedding rates. The proposed detector achieved 93.3% accuracy and an F1 score of 0.903, while differences in chi-square statistics among the evaluated systems were statistically significant. The study provides preliminary evidence of the potential applicability of the proposed workflow within NITA-U and similar digital forensic environments. However, the findings are limited by laboratory-only testing, the 45-image corpus, PNG-only evaluation, and thresholds calibrated using the study’s own generator. The contribution is primarily integrative and contextual, combining detection, recovery, feature analysis, and controlled benchmarking within one workflow, including benchmark results from operational images from a Ugandan government agency. | |
| dc.description.sponsorship | Mr. Lusiba Badru, Dr. Odongtoo Geofrey | |
| dc.identifier.citation | Okiror, C. (2026). Steganalysis Model for Detecting and Recovering Stego Images. Busitema University. | |
| dc.identifier.uri | https://bdears.busitema.ac.ug/handle/123456789/9811 | |
| dc.language.iso | en | |
| dc.publisher | Busitema University | |
| dc.title | Steganalysis Model for Detecting and Recovering Stego Images. | |
| dc.type | Thesis |