A Secure Face–Fingerprint Dual-Biometric Staff Attendance System for Confluence University of Science and Technology, Osara

Abstract

The paper presents the design, implementation, and evaluation of a secure dual-biometric staff attendance system for Confluence University of Science and Technology (CUSTECH), Osara. The methodology combined biometric acquisition and preprocessing, pretrained MobileNetV2, ResNet-50, and ViT-B/16 feature extractors,  normalization, cosine-similarity matching, multimodal feature/score fusion, AES-256-GCM template cryptography, SHA-256 integrity safeguard, and SQLite-logging of grounded attendance data. Ten-subjects produced 10-genuine and 90-imposter verification pairs for the backbones. ResNet-50 showed the biggest general verification capability at accuracy of 0.77, EER of 0.2167, FAR of 0.2333, FRR of 0.20, decidability index of 1.565, and mean latency of 335.75 ms. MobileNetV2 produced the least latency of 207.93 ms, while ViT-B/16 attained accuracy of 0.72 and latency of 694.55 ms. ViT-B/16 revealed the greatest comparative performance against the backbone models. AES-256-GCM enlarged template storage from 10,368 to 10,396 bytes, with encryption times of 0.058 and and decryption time of 0.098 ms; SHA-256 hashing time of 0.043 ms, and template recovery of 100%. Initial usability evaluation using 10 volunteers across 60 task observations gave task success of 98.30%, mean task time of 24.30 s, ease rating of 4.20/5, mean SUS score of 80.50/100, and no unrecoverable crashes. This research interconnects verification decisions to secure official records, which runs role-based access and exportable reports, and a reproducible basis for future edge deployment and larger-scale validation studies. The results demonstrate proof-of-concept feasibility while supporting larger-scale validation before deployment.