Related Experiment Video
Updated: May 21, 2026
03:31
End-To-End Deep Neural Network for Salient Object Detection in Complex Environments
Published on: December 15, 2023
Deepfake face detection using hybrid bag-of-visual-words and multi-CNN feature fusion
Maher Alrahhal1, Fatimah Alqahtani2, Rohaya Latip3
1Research Institute of Sciences and Engineering, University of Sharjah, Sharjah, UAE. maherrahal92@gmail.com.
Scientific Reports
|May 19, 2026
Summary
This study introduces a hybrid deepfake face detection framework combining local forensic features and deep learning. The novel approach enhances accuracy and generalization for robust deepfake identification.
Area of Science:
- Computer Vision
- Digital Forensics
- Artificial Intelligence
Background:
- Deepfake technology poses risks to digital security and public trust.
- Current deep learning detection methods lack generalization and interpretability.
- Need for robust and reliable deepfake face detection solutions.
Purpose of the Study:
- To propose a hybrid deepfake face detection framework.
- To improve generalization and interpretability of deepfake detection.
- To enhance robustness against cross-dataset and forensic challenges.
Main Methods:
- Integrating handcrafted local forensic descriptors (HOG, SURF, FAST, BRISK) with multi-CNN deep semantic representations (ResNet-50, MobileNet, ShuffleNet).
- Utilizing a Bag-of-Visual-Words (BoVW) model for manipulation-sensitive regions.
- Feature-level fusion of local and deep features, classified using Support Vector Machine (SVM).
Main Results:
- Achieved up to 97.55% accuracy on six benchmark datasets.
- Demonstrated consistent outperformance over state-of-the-art methods.
- Maintained robustness under cross-dataset evaluation and challenging forensic conditions.
Conclusions:
- The hybrid approach effectively integrates explicit forensic features with deep representations.
- The proposed framework offers a robust, interpretable, and generalizable solution for deepfake face detection.
- Highlights the significance of combining diverse feature types for advanced deepfake detection.