Analysis of the Effect of Lighting Variations on the Performance of a Face Recognition System Based on FaceNet and MTCNN

Abstract

Face recognition is a biometric technology widely used in automated attendance and access control systems. However, variations in lighting conditions pose a significant challenge that can substantially affect system accuracy in real-world environments. This study aims to analyze the effect of lighting variation on the performance of an MTCNN and FaceNet-based face recognition system and to identify the lighting conditions that yield the best and worst recognition results. This study employed an experimental systems engineering approach using a dataset of 50 facial images collected from 10 students of the Information Technology Study Program at ISTEK ‘Aisyiyah Kendari under five lighting conditions: very bright (>1,000 lux), bright, normal, dim, and very dim (<100 lux). MTCNN was used for face detection and alignment, FaceNet for extracting 128-dimensional embeddings, and Euclidean Distance with a threshold of 0.8 for identification. System performance was evaluated using accuracy, precision, recall, F1-score, and confusion matrix. The results indicate that the system achieved its best performance under very bright conditions, with an accuracy of 70.00%, precision of 58.33%, recall of 70.00%, and an F1-score of 61.67%. Under very dim conditions, all metrics declined by 10%, with a higher average Euclidean Distance (0.8922) compared to very bright conditions (0.8558). Lighting variation was found to have a significant and cascading impact on system performance, affecting image quality, embedding stability, and final classification outcomes. Future development is recommended to incorporate image enhancement techniques such as CLAHE or histogram equalization to improve system robustness under extreme lighting conditions.