ILLUMINATION EFFECTS ON FACE RECOGNITION ALGORITHMS: A COMPARATIVE ANALYSIS AND PRACTICAL INSIGHTS

Abstract

Face recognition systems are being deployed everywhere these days, from airport security to smartphone unlocking. But there's a problem that keeps coming up: these systems don't work well when the light changes. This research examines the impact of varying lighting conditions on the precision of face recognition systems. We looked at how earlier methods like Eigenfaces and Fisherfaces relate to current deep learning methods, like FaceNet. What we found is quite clear: when the lighting isn't perfect, standard algorithms have a hard time, while deep learning models do considerably better. We also looked into whether basic picture preprocessing methods, like CLAHE, could help make things more accurate in low light (Awodeyi, Olutayo, & Adetunmbi, 2025). The results were really good. When we used CLAHE preprocessing with FaceNet, accuracy in low light went up from 82.4% to 95.2%. This is important since mistakes in recognition cost money. When a machine doesn't recognize someone, it either puts security at risk or needs help from a person, both of which cost money (Wei & Rodrigo, 2021). Our research indicates that businesses can make facial recognition systems that perform in the real world and don't cost a lot of money by using the proper mix of contemporary algorithms and sensible preprocessing.