Performance Analysis of a Robust Face Recognition System Using the Fisherfaces Method Under Varying Environmental Conditions

Abstract

Face recognition remains a fundamental biometric technique for identification, yet it continues to face significant challenges regarding illumination and expression variations (Belhumeur et al., 1997). This study presents a face recognition system based on the Fisherfaces method, which leverages Linear Discriminant Analysis (LDA) to maximize class separability (Martinez & Kak, 2001). The proposed methodology integrates image preprocessing steps, including grayscale conversion and normalization, with a dual-stage feature extraction process involving Principal Component Analysis (PCA) and LDA. Classification is performed using a Nearest Neighbor (NN) approach to determine identity. Experimental evaluations on a standard dataset demonstrate an overall recognition accuracy of 90%, with performance levels ranging from 85% to 95%. These results indicate that the Fisherfaces framework remains robust under diverse conditions (Zhao et al., 2003), offering a reliable solution for real-world applications in security and surveillance.