nadatelwazane / face-recognition

Face recognition model that implements dimensionality reduction to reduce memory usage and overfitting

Face Recognition

Table of Contents

• Description
• Implementation
• Results
• Notebook
• Collaborators

Description

This project is a face recognition system that uses AT&T Face database to train a model that can recognize faces. The model is trained using KNN algorithm. It is also a practical introduction to the use of PCA and LDA for dimensionality reduction.

Implementation

We implemented PCA and LDA from scratch using numpy, as well as KNN algorithm.

Additionally, we implemented a binary classification model for identifying faces vs non-faces.

As a bonus, we also implemented KPCA, KLDA as well as DLDA.

Results

The model was able to recognize faces with an accuracy of 0.958 with PCA and 0.95 with LDA. For the LDA and PCA variations we got the following results:

Method	Accuracy
PCA	0.958
LDA	0.95
KPCA	0.975
KLDA	1.0
DLDA	0.94

As for the binary classification model, we got an accuracy of 0.9925 with LDA and 0.995 with PCA.

Notebook

The notebook can be found here.

Collaborators

• Manar Amgad
• Ahmed Dusuki