DeepFake Image Detection

A deep learning web application that detects whether a face image is Real or Fake (AI-generated/manipulated). It uses MTCNN for face detection and a fine-tuned ResNet50 model for binary classification.

How It Works

1. User uploads an image via the web interface
2. MTCNN detects and crops the face from the image
3. The cropped face is resized to 128×128 and normalized
4. A fine-tuned ResNet50 model predicts Real or Fake
5. Result is displayed on the result page

If no face is detected in the image, the app returns a "no face" response.

Tech Stack

• Model: ResNet50 (fine-tuned, Keras .h5)
• Face Detection: MTCNN
• Backend: Flask (Python)
• Frontend: HTML, CSS, JavaScript
• Image Processing: OpenCV, PIL, scikit-image

Project Structure

├── app.py                        # Flask app with upload, predict, result routes
├── model/
│   └── deepfake_resnet50.h5      # Trained model (not tracked in git)
├── static/
│   └── uploads/                  # Uploaded images stored here
├── templates/
│   ├── index.html
│   ├── upload.html
│   ├── result.html
│   └── 404.html
└── requirements.txt

Setup & Run

# Install dependencies
pip install -r requirements.txt

# Run the app
python app.py

Note: Place deepfake_resnet50.h5 inside a model/ directory before running.

Model Details

• Architecture: ResNet50 (transfer learning)
• Input Size: 128 × 128 × 3
• Classes: Real | Fake
• Face Extraction: MTCNN (detects face bounding box before classification)