Optimizing Facial Recognition: An Analytical Comparison of Traditional and Deep Learning Approaches

In the rapidly evolving domains of AI and Internet tech, face recognition, a key machine learning application, is increasingly used in security, identity verification, and public monitoring. As this technology progresses, its applications are expanding. A critical factor in its growing popularity is the advancement of the underlying algorithms, which drives its effectiveness in various applications. This paper investigates facial recognition technologies using the Olivetti faces dataset. It compares the efficacy of Principal Component Analysis (PCA), Eigenface method, Support Vector Machine (SVM) with linear kernel, a standard Convolutional Neural Network (CNN), and a CNN-Transfer Learning approach with MobileNet. The transfer learning is trained with the Adam optimizer (learning rate 0.0004) and categorical cross entropy loss. Over 100 epochs with a batch size of 64, the model's performance, in terms of accuracy and loss, is meticulously tracked and visualized, providing insights into its learning progression. The final evaluation includes a detailed classification report, highlighting precision, recall, and F1-scores, and these metrics are graphically represented for a comprehensive understanding of the model's performance. The CNN model achieved 80% accuracy, while PCA and SVM exhibited high accuracies, reaching 91% and 97.5%, respectively. We have also applied the eigenface method and achieved 91% accuracy in face recognition. Notably, the CNN-Transfer Learning method demonstrates superior performance with a 95% accuracy, highlighting the potential of deep learning in facial recognition applications.