Hybrid Real-Time Image Tampering Detection & Localization — Reproducible Research Release v1.0.0
Latest
This release corresponds to the first reproducible research version of our hybrid framework for real-time digital image tampering detection and localization.
The release is archived for research reproducibility and DOI generation via Zenodo.
Included in this Release
- Complete hybrid detection framework (ViT + EfficientNet + LightGBM + YOLOv8-Seg)
- Training and inference pipeline
- Confidence-weighted ensemble fusion
- Conditional localization logic
- Experiment notebook used for final results
- Dataset preprocessing and evaluation scripts
- Reproducibility configuration and environment notes
Experimental Environment
- GPU: NVIDIA Tesla T4 (Google Colab Free Tier)
- Python 3.10
- CUDA 12.x
- PyTorch / TensorFlow GPU builds
Datasets Used
- CASIA v2 — Copy-Move and Splicing
- Inpainting Localization Dataset
- Deepfake and Real Images Dataset
- JPEG / Seam-Carving Forgery Dataset
- EXIF Metadata Manipulation Dataset
(Dataset links are provided in README)
Purpose of this Release
This release is created to:
- Ensure full reproducibility of the research
- Freeze the exact version used in the paper
- Generate DOI via Zenodo for academic citation
- Provide transparent access to code and experiments
Notes
- This release represents the stable research version used for publication
- Future improvements and experiments may appear in later releases
- Pretrained weights may be added in a future release depending on dataset licensing
Citation
If you use this repository, please cite:
Kamble, V.B., Uke, N.J.
Hybrid Framework for Real-Time Detection and Localization of Digital Image Tampering
(Under submission / 2026)
Contact
Vitthal B. Kamble
Research Scholar — VIIT Pune
Email: vitthalk13@gmail.com