Canonical Basis EigenFaces

Canonical Basis (EigenFaces)

[ Github]

📝 Description:

The "Facial Recognition using Fisher Faces vs Eigen Faces with Support Vector Machines" project aimed to develop a robust face recognition algorithm leveraging supervised learning techniques. Two feature extraction methods were explored: Eigenfaces and Fisherfaces. Support Vector Machines (SVMs) were then trained using these features to classify faces accurately.


🖼️ Datasets Used:
  • Olivetti Dataset: 400 images, 40 classes, 4096 features per image.
  • Labelled Faces in the Wild (LFW) Dataset: 1850 images, user-defined classes, 62500 features per image.
  • Labelled Faces in the Wild (unprocessed) Dataset: 62500 features per image.

🛠️ Methodology:
  • Feature Extraction:
    • Eigenfaces: Utilized PCA to extract eigenfaces from the image datasets.
    • Fisherfaces: Implemented Fisher Linear Discriminant Analysis (FLDA) to extract discriminant features.
  • Classification: SVMs with Linear and Radial Basis Function (RBF) kernels were trained using the extracted features.
  • Evaluation:
    • The performance of classifiers was evaluated using metrics like accuracy, precision, recall, and F1-score.
    • Exploratory Data Analysis (EDA) techniques such as t-SNE visualization were employed to analyze dataset distributions.

🔑 Key Findings:
  • Eigenfaces vs Fisherfaces: Fisherfaces showed improved classification performance compared to Eigenfaces, especially for complex datasets.
  • SVM Kernel Comparison: Linear SVMs outperformed non-linear (RBF) SVMs for most datasets.
  • Dataset Complexity: Performance varied based on dataset complexity, with higher accuracy achieved on simpler datasets.
  • Impact of Dimensionality Reduction: Dimensionality reduction techniques like PCA significantly influenced classifier performance.
🏁 Conclusion:

The project demonstrated the efficacy of Fisherfaces combined with SVMs for face recognition tasks. By comparing different feature extraction methods and SVM configurations, valuable insights were gained into the nuances of facial recognition systems. The findings contribute to the ongoing research in computer vision and pattern recognition domains, paving the way for more accurate and efficient face recognition algorithms.