Face Recognition using CNNs and SVMs
Introduction
This study presents the performance of CNNs and SVMs for a supervised face recognition and classification task. To improve the generalisation and performance of the models, hyper-parameters are varied and adjusted in grid searches. In the case of SVMs, image features are extracted using Histogram of Oriented Gradients (HOG), and the results from its grid search are validated with a stratified K-fold cross-validation. The best evaluated models are then compared on their confusion matrices and ROC curves. It is found that CNNs achieved 99.6% accuracy and SVMs achieved 99.8% accuracy. The project is written in MATLAB, and codes can be found in this GitHub Repo.