EFFECT OF LIGHTING AND BACKGROUND ON THE ACCURACY OF FACE RECOGNITION SYSTEMS: A CNN-BASED EVALUATION IN NIGERIAN INSTITUTIONAL ENVIRONMENTS

Abstract

Face recognition technology is being used more often in educational and institutional environments for managing attendance, controlling access, and identity verification. However, the accuracy of the system based on convolutional neural network (CNN) can be greatly affected by environmental factors, especially the amount of lighting intensity and how complex the background is. This study looks at how differences in lighting and background impact the performance of face recogni-tion systems, with a focus on the typical conditions found in Nigerian institutions. A CNN-based system employing MTCNN for face detection and a ResNet-50 backbone for feature extraction was evaluated on a custom dataset of facial images from volunteer participants under five lighting conditions bright, normal, dim, backlit, and side-lit and three background types: plain, moderately cluttered, and highly cluttered. The system's performance was measured using recognition accuracy, False Acceptance Rate (FAR), and False Rejection Rate (FRR). The results showed that recognition accuracy varied from 93.2% in the best conditions to 65.8% in the worst combined scenario.  Statistical analysis confirmed that both factors exert significant independent and interactive effects on recognition performance. These findings suggest the need for training data that reflects real-world environments, adaptive image preprocessing techniques, and deployment strategies that take context into account to ensure reliable and fair use of face recognition in African institutional settings.