FACENET MODEL APPLICATION IN THE CONSTRUCTION AND DEVELOPMENT OF FACE RECOGNITION SYSTEM AT HANOI UNIVERSITY OF INDUSTRY

Abstract

The face recognition system is one of applications, based on the foundation of photography editing and machine
learning methodology, which assists computers in confirming and recognising someone from a picture or a video frame.
There have been a lot of algorithms mentioned and one of them can be listed as the comparison of facial characteristics
determined from pictures with a database of faces collected previously (one-to-many matching) [1]. However, the fact that
using those common algorithms solely, even with a small image database, can lead to the waste of resources and time for
the recognition system in calculations while the accuracy rate of a prediction remains low. In recent years, the significant
development of deep learning, especially the development of convolution neural networks, has contributed to the focus
and enhance more than ever of the recognition systems. The Facenet model was introduced in 2015, which has been
applied to almost all recognition systems until now, having a remarkable advantage in the development of Siamese
network architecture, co-operated with the utilization of a flexible loss function for the training in large image databases. In
this article, the authorities will analyse as well as provide methodologies to enhance Facenet model for the application in
constructing and developing a suitable recognition system meeting the requirement of large numbers of students in taking
attendance and managing students at Hanoi University of Industry.