Which Characteristics May or May not Bias Facial Recognition Software?

Biometric technology, such as facial recognition, has been known to falsely identify people of color more often than other demographics. This issue has led to the loss of civil liberties for those who are the victim of false-negative rates (Buolamwini, & Gebru, 2018). This is a study to identify characteristics that may bias biometric readings, with particular interest in facial recognition software. This study uses facial recognition software on a personal device from a Samsung Android S10 smartphone. Three individuals with varying age, gender, skin colors and tones participated in this study. Glasses, sunglasses, a hat, a black mask, and a white mask served as variables to obscure portions of their faces. Interesting patterns emerged while analyzing the results of this study. Most surprising was the dark-skinned African American male unlocked the Android device while wearing dark sunglasses along with a hat, which resulted in a false-positive match. The Android device sensor was unable to differentiate his eyes from the dark sunglasses or his forehead from the hat, which may indicate a limitation of the facial recognition software to distinguish darker skin tones. Ascertaining biases against people of color within the facial recognition feature of a personal-use device requires further study