Multi-model Face Recognition Pipeline and Anti-spoofing for a Generic Attendance System

Abstract

Face recognition has many use cases, and the attendance system is one of the most promising areas of them. But choosing the right face detection and recognition model for the right purpose is always a billion-dollar decision before rolling out a real-time computer vision-based project. Although it is a very common understanding that face recognition is a closed problem, in reality it still has a lot of areas for improvements in the context of implementations. Development of a general-purpose face recognition-based attendance system is completely dependent on the availability of generalized face detection and recognition algorithms which can keep balance between speed, accuracy and anti-spoofing at the higher side of the benchmark. In some of the use cases, speed affects the usability and in some other use cases accuracy affects the usability. So in this paper, we have tried to find face detection and recognition models which can address all types of use cases and in the absence of such a model we tried to create alternative architecture to achieve the best out of the existing model by right positioning them in a pipeline.