REAL AND SIMULATED MASKED FACE RECOGNITION WITH A PRE-TRAINED MODEL

Audrey, Albert and Chai, Soo See and Goh, Kok Luong and Chin, Kim On (2023) REAL AND SIMULATED MASKED FACE RECOGNITION WITH A PRE-TRAINED MODEL. Journal of Theoretical and Applied Information Technology, 101 (19). pp. 5855-5864. ISSN 1817-3195

PDF REAL AND.pdf Download (811kB)

Abstract

Facial recognition has currently become indispensable owing to the efficacy of precise identification verification. Because of the distinctiveness of human biometrics, face recognition enables humans to communicate with technology while maintaining their privacy. Advancements in pre-trained models such as FaceNet have enabled improvement in identification accuracy in face recognition technology. Response to the Covid-19 pandemic has led to the replacement of conventional face recognition with masked face recognition. This change has encouraged the use of collaboration to resolve the related issues, which has resulted in the development of algorithms for face occlusion, collection of data on masked and unmasked faces and improvement of pre-trained models. Current research has utilised custom datasets or a specially produced dataset for masked face recognition. To increase the amount of data available for modelling, some studies have implemented mask simulation in facial photos. In this study, FaceNet is evaluated on two datasets: the real-masked face recognition dataset and the simulated masked face recognition dataset. Particularly, we highlight the performance of FaceNet on simulated masked faces. Using simulated masks achieved 67% accuracy, while the use of real masks achieved 84.3%. Results from the two datasets are compared with each other and with other studies using different pre-trained models with similar datasets. This study reveals that simulated masked faces perform less effectively than real masked faces, as corroborated by various other studies.

Actions (For repository members only: login required)

View Item