Eigenfaces Demo: Educational Facial Recognition App

An interactive Streamlit application that breaks down and visualizes the Eigenfaces facial recognition algorithm (Turk & Pentland, 1991).

This project is designed for students and enthusiasts to understand the mathematics behind early computer vision, specifically how Principal Component Analysis (PCA) is used to reconstruct and recognize human faces.

Features

The app is divided into 7 educational tabs:

0. Training Deconstruction: See exactly how the algorithm learns from training data:
   • Calculate the Average Face
   • Compute Difference Faces
   • Build the Covariance Surrogate Matrix
   • Find Eigenvectors and Eigenvalues
   • Derive the final Eigenfaces
1. Component Explorer: Inspect individual Eigenfaces and view a Scree Plot (Explained Variance) to understand data compression.
2. Step-by-Step Reconstruction: A detailed mathematical walkthrough showing how a face is built layer-by-layer:
   • Centering ($\text{Face} - \text{Mean}$)
   • Projection (Calculating Weights/DNA)
   • Weighted Sum (Mixing the ingredients)
   • Final Reconstruction
3. Live Recognition: Pick a random "Unknown" face and see the two-path algorithm (Face Detection + Identity Check) with threshold explanations.
4. Webcam Recognition: Take a photo of yourself to see how the algorithm processes real-world images.
   • Auto-Face Detection: Uses OpenCV/Haar Cascades to automatically find and crop your face.
   • Real-time Processing: Preprocesses your image to match the training data format (64x64 grayscale).
5. History & Context: Learn about the 1991 revolution in Computer Vision.
6. Modern AI: A comparison between Eigenfaces (Holistic) vs. Modern Deep Learning/CNNs (Feature-based).

Installation

1. Clone the repository

   git clone https://github.com/DavidHavoc/eigenFaces.git
   cd eigenFaces

2. Install dependencies

   pip install -r requirements.txt

3. Run the App

   streamlit run main.py

Technologies Used

• Python 3.x
• Streamlit: For the interactive web interface.
• Scikit-Learn: For the PCA implementation and Olivetti Faces dataset.
• OpenCV: For real-time face detection and image processing.
• Pillow (PIL): For handling image uploads and formatting.
• NumPy: For linear algebra calculations (Dot products, Norms).
• Matplotlib: For plotting face images and graphs.

Educational Value

This project helps answer:

• How can math describe a face?
• What is a "Weight Vector" or "Face DNA"?
• How does dimensionality reduction (PCA) work on images?
• Why did we switch from Eigenfaces to Deep Learning?

References

• Turk, M., & Pentland, A. (1991). "Eigenfaces for Recognition." Journal of Cognitive Neuroscience, 3(1), 71-86.
  IEEE Xplore

Common Issues & Troubleshooting

1. Images Not Loading / Broken Images

Issue: Sometimes the face images or plots appear as broken icons or do not render immediately. Fix: This is often a Streamlit caching or rendering hiccup.

• Refresh the page (Cmd+R or F5).
• If running locally, try clearing the Streamlit cache by clicking the "Hamburger menu" (top right) -> "Clear cache" and then rerun.

2. "ModuleNotFoundError"

Issue: Passages like ModuleNotFoundError: No module named 'cv2' or sklearn. Fix: Ensure you have installed the requirements in your active environment:

pip install -r requirements.txt

3. Webcam Not Starting

Issue: The webcam component stays blank or shows an error. Fix:

• Ensure your browser has permission to access the camera.
• Make sure no other application (like Zoom or Teams) is currently using the camera.
• If on macOS, check System Settings -> Privacy & Security -> Camera to ensure your terminal/browser has access.

4. Port Already in Use

Issue: Error "Port 8501 is already in use" or the app doesn't launch. Fix:

• Option A: Close the other terminal window where Streamlit is currently running (Press Ctrl+C).
• Option B: Run on a different port manually:

  streamlit run main.py --server.port 8502

License

This project is open-source and available under the MIT License.