Tugas Akhir ( Perancangan Sistem Absensi Mahasiswa Berbasis Face Recognition Menggunakan metode CNN )

Abstract

Abstract—Conventional student attendance systems can be inefficient and susceptible to attendance fraud. This study presents a web-based student attendance system using one-to-one face verification, where a captured face is verified only against the facial embedding associated with the authenticated user account. The system integrates a Next.js frontend, a Flask backend, and Supabase for authentication and database management. A pretrained ArcFace model accessed through the DeepFace library generates 512 dimensional facial embeddings, which are compared using cosine similarity with an empirical verification threshold of 0.75. The system also applies image-quality prechecking, a simple liveness check, and performance logging. Experimental evaluation covered capture distance, illumination, processing latency, facial attribute variation, and a screen-based presentation attack. The highest verification success rate was 100% at a capture distance of 50 cm. Successful verification was observed from 2.16 to 955 lux in the tested illumination scenarios. Average processing time was 1557.25 ms for check-in, 1913.75 ms for check-out, and 4756.82 ms for enrollment. The tested screen-based presentation attack was rejected by the implemented quality and liveness checks.