Official repository for the HDBK (Hybrid Deep, Block, and Keypoint) framework and a centralized hub for image forgery detection benchmarks[cite: 1]. This project provides the implementation and datasets for robust detection and pixel-level localization of copy-move manipulations[cite: 1].
The HDBK model addresses the limitations of traditional forensic methods by integrating a triple-architecture deep learning ensemble with Genetic Algorithm (GA) optimization and keypoint-based descriptors[cite: 1]. It is designed to be resilient against geometric transformations such as scaling and rotation, as well as post-processing attacks like blurring and noise[cite: 1].
- Triple-Architecture Ensemble: Leverages VGG16, MobileNet, and EfficientNetB0 for high-accuracy image-level detection[cite: 1].
- Genetic Algorithm Optimization: Uses GA to optimize block-matching processes, significantly reducing false positives in complex regions[cite: 1].
- Pixel-Level Localization: Employs SIFT, SURF, and FAST descriptors for precise forensic mask generation[cite: 1].
- High Performance: Achieved 97.34% accuracy and an 87.17% IoU on the CoMoFoD benchmark[cite: 1].
This repository includes data and benchmarks for the following industry-standard forensic datasets used in our research[cite: 1]:
| Dataset | Description | Key Focus |
|---|---|---|
| CoMoFoD | 10,000 images ( |
Post-processing attacks (Blur, Noise, JPEG)[cite: 1] |
| MICC-2000 | 2,000 images (various sizes)[cite: 1] | Geometric transformations (Scale/Rotation)[cite: 1] |
| COVERAGE | 100 original/forged pairs | Similar object copy-move detection |
| CASIA | V1 and V2 benchmarks | Splicing and copy-move forgery |
| GRIP |
|
High-precision pixel-level localization[cite: 1] |
If you use this dataset or code in your research, please cite the following papers:
- A hybrid model for image forgery detection using deep learning with block and keypoint methods
Scientific Reports, 2026.[cite: 1] - Copy-move forgery detection and localization using deep learning
International Journal of Pattern Recognition and Artificial Intelligence. - An optimal hybrid method to detect copy-move forgery
Journal of AI and Data Mining. - A survey on deep learning-based image forgery detection
Pattern Recognition, 2023. - Deep learning framework to extract anatomy for mosquito image classification
Scientific Reports.
- Python 3.8+
- TensorFlow / Keras[cite: 1]
- OpenCV (for SIFT/SURF extraction)[cite: 1]
- NumPy & Matplotlib
# Clone the repository
git clone https://github.com/your-username/image-forgery-detection-dataset.git
# Install dependencies
pip install -r requirements.txt
# Run the HDBK detection script
python main_hdbk_detection.py --input ./dataset/test_image.pngResults obtained on the CoMoFoD dataset[cite: 1]:
- Accuracy: 97.34%[cite: 1]
- Precision: 92.04%[cite: 1]
- Recall: 91.72%[cite: 1]
- F1-Score: 91.85%[cite: 1]
- IoU: 87.17%[cite: 1]
This project is licensed under the Creative Commons Attribution-NonCommercial 4.0 International License[cite: 1]. For commercial use, please contact the authors.
Contact: For questions regarding the JAICT journal or this research, please open an issue or contact the corresponding author at Meybod University.