Adaptive Hybrid Reduction for Facial Recognition

Abstract

The rapid growth of large-scale image datasets has highlighted the importance of dimensionality reduction in improving computational efficiency while preserving critical visual information. This paper investigates both single-method and hybrid approaches to dimensionality reduction for facial recognition, with a particular focus on the Face94 dataset. Four single techniques, PCA, LDA, NMF, and UMAP, are compared against hybrid strategies that combine two or three methods. Experimental results demonstrate that hybrid approaches consistently outperform individual methods, achieving higher accuracy, precision, recall, and F1-score, while maintaining reasonable computational costs. Among them, tri-hybrid combinations such as PCA-LDA-UMAP achieve near-perfect recognition performance, confirming the effectiveness of carefully designed hybridization strategies for robust facial recognition.