Face Recognition System using PCA, LDA, and Machine Learning

Face Recognition System using PCA, LDA, and Machine Learning

This project implements a face recognition system using classical machine learning techniques and dimensionality reduction methods. The system was built using the ORL Face Dataset, which contains 400 grayscale images from 40 individuals, with each image converted into a feature vector for machine learning processing

project2 ML.ipynb

.

The project pipeline includes several important steps:

Data Preparation

Images were loaded using OpenCV in grayscale format.

Each image was flattened into a 10,304-dimensional feature vector.

Feature matrices and label vectors were constructed for model training and evaluation

project2 ML.ipynb

.

Dimensionality Reduction

To handle the high dimensionality of image data, two techniques were applied:

• Principal Component Analysis (PCA)

Used to reduce the dimensionality of image vectors while preserving most of the variance in the dataset.

• Linear Discriminant Analysis (LDA)

Applied to maximize class separability by analyzing within-class and between-class variance.

Classification Algorithms

The reduced feature representations were evaluated using multiple classifiers:

K-Nearest Neighbors (KNN) for multiclass face recognition.

Support Vector Machine (SVM) with an RBF kernel for binary classification (Face vs. Non-Face).

Different K values (1, 3, 5, 7) were tested to analyze model sensitivity and optimize classification performance

project2 ML.ipynb

.

Experimental Analysis

Compared different variance thresholds in PCA (0.80 – 0.95).

Evaluated different training/testing splits (7:3 vs. 5:5).

Visualized classification accuracy and model behavior across configurations.

Results

LDA achieved better class separability than PCA for multiclass classification.

SVM performed strongly in binary face detection tasks.

Increasing training samples improved model accuracy.

This project provided practical experience in image processing, dimensionality reduction, and machine learning classification techniques for computer vision tasks.