Published in: Journal of Computer Science and Engineering in Innovations and Research (ISSN No: 3049-1762 online)
Publication Date: May 15, 2025
Face masks significantly impair traditional face recognition systems by obscuring crucial facial features like the nose, mouth, and chin. The variety of mask types, including surgical and cloth masks, further complicates recognition due to differences in materials, shapes, and colors. This research aims to address the challenges posed by facial occlusion, particularly mask-wearing, and improve recognition system performance. The study leverages the Facenet deep learning model, which uses a convolutional neural network (CNN) to generate facial embeddings for identification or verification purposes. To overcome the limitations of existing models in recognizing masked faces, a Refined Facenet model is proposed. The experiment compares three models—Facenet, Refined Facenet, and VGG16—paired with five classifiers: Support Vector Machine (SVM), Decision Tree, k-Nearest Neighbors (KNN), Random Forest, and Gaussian Naïve Bayes. Datasets used include Unmasked, Masked, Merged, and Augmented versions of these datasets, with evaluations performed using K-fold cross-validation and Bootstrap sampling. Results demonstrate that the Refined Facenet model with KNN consistently achieves higher accuracy and optimized CPU execution time across different datasets, providing a more robust solution for face recognition under occlusion conditions.
Image Processing, Convolutional Neural Networks (CNNs), k-Nearest Neighbors (KNN), Machine Learning, Facenet, Face Recognition