Comparative Study of FaceNet and ArcFace Performance in Customer Facial Recognition for the Lao Credit Information System

Abstract

This study compared the performance of two facial recognition technologies, FaceNet and ArcFace, for potential application in the Lao Credit Information Bureau system. The evaluation focused on two key factors: processing speed and recognition accuracy. The methodology involved testing with a sample of 200 individuals drawn from 10 financial institutions (20 participants per institution). Each participant contributed 20 facial images captured from multiple angles (front, 45-degree, and side views) under controlled lighting and distance conditions. Standardized hardware and testing procedures were employed to ensure consistency. Findings revealed significant performance differences between the two technologies. FaceNet demonstrated faster processing speeds, averaging 27.24 seconds per task—approximately 15.01% faster than ArcFace’s average of 31.27 seconds. In contrast, ArcFace achieved higher recognition accuracy, with an average rate of 94.60%, surpassing FaceNet’s 90.75% by 4.26%. The choice between the two technologies depends on implementation priorities. FaceNet is better suited for applications requiring faster processing, whereas ArcFace is preferable in contexts where higher recognition accuracy is critical. These insights can inform the adoption of effective biometric solutions within credit information systems.