Face Recognition Approach Using ArcFace Embeddings and Traditional Machine Learning

Le résumé fourni par la source

Facial recognition systems have evolved over the years, becoming more complex and powerful. Despite the increasing complexity of facial recognition models, we demonstrate that a streamlined approach can still achieve competitive performance. This paper presents a facial recognition pipeline that combines ArcFace-based facial embeddings with traditional machine learning models to identify individuals with their associated metadata. The images undergo several preprocessing stages, including face extraction using Deep Neural Network, frontal face alignment via Haar Cascade classifiers based on eye positions. We used ArcFace to generate face embeddings, which were then used as input features for classifiers and got 100 % accuracy on Logistic Regression, K-Nearest Neighbors, Support Vector Machine and Ensemble model using both ORL and Yale datasets. Furthermore, we got 93.1% accuracy on Logistic Regression using our mixed dataset of 12,372 images and 210 labels. This work contributes both a deployable recognition pipeline in resource-constrained environments and valuable insights into model selection for structured facial recognition tasks.

Ce résumé expose les affirmations des auteurs. BNTIC ne l’interprète pas comme une validation indépendante des résultats.