Face recognition for banking web-app using neural network

Abstract

In recent years, facial recognition technology has made more and more progress in real world applications and systems. Primarily, security and privacy are among the top concerns regarding the applicability of face recognition. With the ability to identify each individual from unique facial structures and features, face recognition has the potential to become a reliable security method. However, the performance and accuracy of face recognition algorithms are challenging problems that prevent people from using them. This thesis aims to create a simulated banking web application, and apply face recognition to enhance security when authenticating and transferring. Specifically, Dlib[19] is used as the face detector, Facenet512 [1] is the face recognition model in the application. Face expression analysis with HyperExtended LightFace [28] is also employed to prevent or reduce the risk of using face photos instead of real faces. Moreover, to choose the most suitable face recognition model for the application, experiments are performed to evaluate performance between state-of-the-art face recognition models. At the end, Facenet512 is the chosen one thanks to its excellent precision, high accuracies and fast execution time