Web-based Face Detection and Recognition using YOLO and Dlib

The electronic-based government system (SPBE) is an effort initiated by the Indonesian government to create a clean and trustworthy governance and public service system. However, the government’s public services involve sensitive data owned by Indonesian citizens. An identity verification system is needed to ensure the security and integrity of the system and prevent system abuse. A face recognition system is arguably the most accessible biometric verification method available. This paper used YOLOv5-Face and YOLOv7-Face pre-trained models for face detection. The models were then integrated with the Dlib library to identify faces based on their identities. Testing was performed by determining a confusion matrix for each weight used. The best-performing face recognition system was then implemented as a web application. The yolov7-lite-s weight achieved an accuracy of 94.5% with an average processing time of 237.45 ms per image on a private dataset.