ABOUT
StegVision Project
StegVision is a cybersecurity Final Year Project for multisteganalysis using CNN and Transformer models. The system analyses images, audio, and video through dedicated forensic engines and returns explainable probability scores with evidence channels.
Current deployment engines: video_ensemble_v4b_calibrated, audio_ensemble_v3_realworld, and Stegformer ONNX image ensemble.
IMAGEStegformer transformer + spatial LSB + JPEG-frequency + texture
VIDEO32–128 adaptive frames, H.264/DCT, pixel CNN, screen-capture guard
AUDIOMultisegment CNN + SPA/RS + codec profiling + calibrated fusion
APIFlask REST — POST /predict, GET /health
FRONTENDStatic HTML / CSS / JavaScript
PROJECT TEAM
Research Group
StegVision was developed as a group Final Year Project under the supervision listed below.
Supervisor
Dr. Farhan Hassan
Academic supervisor for the StegVision cybersecurity FYP — guiding system design, steganalysis methodology, evaluation, and deployment.
ANALYSIS ENGINES
Deployed Forensic Stack
VIDEO v4bAdaptive frame sampling, multichannel forensics, benign screen-recording guard, verdict-locked probabilities.
AUDIO v38–48 segments, SPA/RS LSB discrimination, codec-noise profiling, CNN calibration for real uploads (OGG/WhatsApp/music).
IMAGEStegformer ONNX per frame or file plus independent spatial and frequency evidence modules.
DEPLOYMENT
Run and Deploy
Local testing and cloud deployment steps.
1
Install dependencies
pip install -r requirements.txt
2
Prepare image model
python tools/prepare_stegformer_onnx.py --output checkpoints/stegformer.onnx
3
Start API (website + backend)
.\scripts\run-local.ps1
Confirm /health shows audio_engine: v3_realworld and video_engine: v4b_calibrated.