Single-Sample Face Recognition for Attendance Record

Lee, Perng (2021) Single-Sample Face Recognition for Attendance Record. Final Year Project (Bachelor), Tunku Abdul Rahman University College.

Text LEE PERNG_Full text.pdf Restricted to Registered users only Download (9MB)

Abstract

Face recognition system is widely used in the world; it had made the human live become more convenient. The traditional face recognition requires multiple of samples for each person to get high accuracy result, because the pose, expression and illumination variations will affect the similarity between two samples. This is inconvenient for the user because they need to upload more than 10 images to the database to ensure the recognition result is accurate. Besides that, the database will become massive and system efficiency become lower. To overcome this problem, the single sample face recognition (SSFR) had proposed. The core of the single sample face recognition is using 1 sample only to complete the recognition process. The challenge of single sample face recognition is insufficient of sample to recognize pose, expression and illumination variations image. To overcome these problems, this project will present a method which is combine the Multi-Task Convolutional Neural Network (MTCNN) and Facenet model to design a single sample face recognition for attendance record. The accuracy of Facenet model is about 98.68% on LFW dataset and 96.38% on CASIA-FaceV5 dataset. A user interface was created by PyQt to allow user take attendance by face recognition system and manage database.