🧠 SVM Face Recognition

A machine learning project that detects and recognizes faces of political figures using Support Vector Machine (SVM) and Principal Component Analysis (PCA) with OpenCV.

---

👤 Recognizes

• Barack Obama
• Donald Trump
• George W. Bush

---

📊 Model Performance

Metric	Score
Accuracy	85.29%
Mean CV Accuracy	83%
Standard Deviation	0.07

Classification Report

Person	Precision	Recall	F1-Score	Support
Barack Obama	1.00	0.71	0.83	7
Donald Trump	0.92	0.86	0.89	14
George W Bush	0.75	0.92	0.83	13

Confusion Matrix

              Predicted
              Obama  Trump  Bush
Actual Obama  [  5     0     2 ]
       Trump  [  0    12     2 ]
       Bush   [  0     1    12 ]

---

🛠️ Technologies Used

Library	Purpose
OpenCV	Face detection & image handling
NumPy	Array operations
Scikit-learn	SVM, PCA, model evaluation
Matplotlib	Plotting & visualization
Seaborn	Confusion matrix heatmap
Joblib	Saving & loading the model

---

📁 Project Structure

SVM Project/
│
├── train_data_2/
│   ├── Barack Obama/         # Training images
│   ├── Donald Trump/         # Training images
│   └── George W Bush/        # Training images
│
├── test_data/                # Test images
│
├── Face_Recognition_model.ipynb       # Main notebook
├── haarcascade_frontalface_default.xml  # Face detector
├── SVM-Face Recognition.sav          # Saved trained model
├── .gitignore
└── README.md

---

⚙️ Installation

pip install opencv-python numpy scikit-learn matplotlib seaborn joblib

---

🔄 How It Works

1. Load images from each person's folder
         ↓
2. Detect faces using Haar Cascade Classifier
         ↓
3. Manually approve detected faces (press 'y' to approve)
         ↓
4. Resize approved faces to 50×50 pixels
         ↓
5. Flatten faces from 50×50 → 2500 features
         ↓
6. Reduce features using PCA (2500 → 150)
         ↓
7. Train SVM classifier on reduced features
         ↓
8. Evaluate model — accuracy, report & confusion matrix
         ↓
9. Save trained model as .sav file

---

🚀 How to Run

Step 1 — Clone the repository

git clone https://github.com/Isi2004-star/SVM-Face-Recognition.git

Step 2 — Install dependencies

pip install opencv-python numpy scikit-learn matplotlib seaborn joblib

Step 3 — Open the notebook

jupyter notebook Face_Recognition_model.ipynb

Step 4 — Run all cells in order

• During face collection, a window will pop up showing each detected face
• Press y to approve and add to dataset
• Press any other key to skip

---

📈 Dataset Summary

Category	Samples
Barack Obama	92
Donald Trump	127
George W Bush	114
Total	333

Split	Count
Training set	299
Testing set	34

---

🧠 Model Architecture

Raw Face Image (50×50 grayscale)
        ↓
Flatten → 2500 features
        ↓
Normalize → pixel values 0 to 1
        ↓
PCA → reduce to 150 components (whiten=True)
        ↓
SVM → kernel='rbf'
        ↓
Prediction → Obama / Trump / Bush

---

📌 Key Concepts

PCA (Principal Component Analysis) Reduces 2500 pixel features down to 150 most important components, removing noise and speeding up training.

SVM (Support Vector Machine) Finds the optimal boundary between each person's facial features using the RBF kernel, which handles non-linear patterns effectively.

Cross Validation Model was validated using 5-fold cross validation achieving a mean accuracy of 83% with a standard deviation of 0.07, indicating a stable and consistent model.

---

👩‍💻 Author

Isiri Gangamini Jayanetti

• GitHub: @Isi2004-star