HIGH PERFORMANCE FACIAL RECOGNITION MATCHER

Abstract

The utilization of biometric products is an expanding landscape; from general consumers employing it for authenticating into their devices to governments deploying it at the forefront of crime and border control. One sizeable organization required an expansion in their offering within the industryThis study aims to develop a facial matching solution that offers high performance and meets the requirements of the organization’s biometric Subject Matter Experts in order to meet the current gap in the offering. A facial recognition approach known as FaceNet was utilized along with the GO language and MongoDB to produce an application capable of performing enrolments and matches against a persistent set of candidates. This solution was validated against the labeled Faces in the Wild dataset, a challenging set of facial biometric data in function, performance, and accuracy testing. For a subset of 6000 images from the dataset, a 100 % accuracy was recorded from multiple test runs demonstrating no false matches. The application's performance against this subset was averaged over multiple executions using two concurrent connections, which concluded an average enroll response time of 70ms and 236ms for match requests giving transactions per second values of 29 and 8 respectively.