This project is an AI-powered system designed to detect deepfake videos by analyzing visual features and facial regions within video frames.
It was developed as part of an academic graduation project to address the growing risks of deepfake media and provide a practical filtering solution.
Deepfake content can threaten privacy, trust, and information security.
Our system aims to automatically detect manipulated videos by analyzing faces, extracting features, and classifying content in real time.
A key part of this project is implementing a Deepfake Filter that can be integrated into a user-facing mobile application to flag suspicious videos immediately.
- Face Detection & Extraction: Automatically detects and crops faces from video frames using OpenCV.
- Deep Feature Extraction: Uses a deep learning model (Xception) to analyze subtle visual patterns.
- Real/Fake Classification: Predicts the authenticity of faces and frames.
- Video-Level Aggregation: Combines frame-level results to classify entire videos.
- Flask REST API: Serves predictions for easy integration with other applications.
- Mobile Filter Application: Provides a practical Deepfake Filter that works through a mobile app (built with Flutter) to detect deepfake videos directly on user devices.
- Python, TensorFlow, Keras — Model training and prediction
- OpenCV — Video processing & face extraction
- Flask — REST API for serving the detection model
- Flutter — Cross-platform mobile app implementing the Deepfake Filter
- FaceForensics++ — Base dataset for training and evaluation
1️⃣ Upload Video: User uploads a video through the mobile app.
2️⃣ Frame Extraction: The system splits the video into frames and extracts faces.
3️⃣ Prediction: The AI model classifies each face as real or fake.
4️⃣ Filtering: The Deepfake Filter aggregates predictions and shows the user a result in the app.
5️⃣ Response: The user receives a clear score and recommendation (e.g., safe or suspicious content).
- Bahaa aldin Alzhouri
- Mahmut Basmaci
Graduation Project — Firat Universitesy — 2025
- FaceForensics++ Dataset — GitHub
- Xception Model — Paper
- Flutter — flutter.dev
This project is part of an academic study and is not intended for commercial use.
Contact us for questions or possible collaboration.