Similarity Measures for Person Identification with Face Mask

• Published in: IEEE Symposium on Industrial Electronics and Applications pp. 1 - 7
• Main Authors: Huzaimi, Muhammad Hakimi, Mazalan, Lucyantie
• Format: Conference Proceeding
• Language: English
• Published: IEEE 15.07.2023
• Subjects: COVID-19; Current measurement; Deep Learning; Face recognition; Face Similarity; Feature extraction; Image Classification; Process control; Surveillance; Traffic control
• ISSN: 2472-7660
• DOI: 10.1109/ISIEA58478.2023.10212158

Facial recognition is a method of identifying or verifying a person's identification by examining their face. People can be identified in pictures, videos, or in real time using facial recognition technology. After facing more than two years of Covid-19 restrictions, face masks are optional but strongly advised for usage. As new cases are still recorded on a daily basis, wearing face mask in a social setting is still the safest way to protect ourselves and others from COVID-19. However, the usage of face mask has brought difficulties for a person or face identity recognition because half of the face features has been covered. The proposed research starts with face mask detection for determining if the subject is wearing a mask or not. The system uses MobileNetV2 for face mask detection resulted to 100% correct when identifying face mask. Secondly, to evaluate the effectiveness of various similarity coefficient formula for person identification with face mask. A dataset of 35 people wearing face masks was examined. The study involving usage of multiple similarity measures formula; Euclidean Distance, Cosine Similarity, Manhattan Distance, and Jaccard Similarity to evaluate the similarity between two images. It is identified that Jaccard Similarity is the most sensible similarity coefficient method for person identification with face mask. This study offers some insight that worth to further analyses the potential of Jaccard coefficient in measuring the person identity with a face mask.