Explainable-AI in Multiview Facial Authentication Using Automatic Image Segmentation

Abstract

Accurate authentication of a human user without considering specific proximity, context, and environmental conditions, is a complex yet crucial task in applications such as surveillance and proof of identity. While significant research has been conducted in the direction of human authentication, completing authentication with an explainable process is still far from being perfect. This is mainly due to depending on machine learning to create behind-the-scene calculations that the user can't fully understand. Another issue is the requirement of distance to the device and neutral face expressions. Multiview facial authentication relates to the ability to capture a face from the front or from the side and use these views for the authentication. This allows for an increase in the flexibility of the process.

In this research, the proposed method includes a flexible authentication solution based on an accompanying explainable process. The explainable authentication approach uses a facial feature set that can be segmented automatically once a face has been successfully detected. The proposed feature set is made up of a combination of facial minutiae and contours. Additionally, the proposed solution will include an authentication solution for faces from the side, referred to as a face-profile.