Facial Recognition System screening evaluation methodology for complexion biases

Abstract

Over the years, Facial Recognition Systems (FRS) have come under scrutiny from watchdog groups who voice their complaints concerning the potential existence of a FRS bias towards certain cultures of people while such systems are generally deployed in security screening situations. To better understand this potential FRS bias, researchers examine a theory developed from the behavioral sciences known as the "other-race effect." FRS researchers have also used the "other-race effect" theory in an attempt to explain the occurrence of biases associated with algorithms tested during the Face RecognitionVendor Test of 2006. In this paper, we develop a scientific testing methodology based on the factors of illumination, distance, and angle to evaluate whether or not a chosen FRS exhibits a significant bias when presented with two dissimilar three dimensional (3D) facial models for comparison. To test our methodology, we compare a light complexion (3D) facial model with a medium complexion 3D facial model. Our methodology will incorporate a full factorial experiment and Design of Experiments (DOE) Pro statistical software for data processing. Multiple regression and ANOVA are also used to analize the results. This testing methodology has been incorporated in our academic programs and implemented in different Biometric Security and DOE course projects. © 2011 American Society for Engineering Education.