A sampling of commercial facial recognition algorithms shows that face biometrics systems are currently too dependent on specific attributes to distinguish between people of the same race and gender. The problem is investigated in a paper using five unnamed commercial facial recognition algorithms and a commercial iris recognition algorithm as a control.
According to the paper, facial recognition algorithms that search large databases of homogeneous images tend to treat people differently based on race and gender. The research builds on the scientist’s previous work, which showed that images of people from the same demographic group tend to be scored more similar to each other than when compared to people from a different demographic group, a phenomenon they call broad homogeneity.
The analysis of the principal components used by biometrics algorithms reveals that most variations between vectors, or facial appearance, are not related to race or gender, though approximately 10% are. It means that face biometrics algorithms should be able to avoid delivering unfair accuracy differentials more consistently.
Furthermore, the separation between mated and non-mated score distributions was only slightly reduced when reconstructed exclusively using principal components that do not cluster individuals by race or gender, indicating that CFRAs (commercial facial recognition algorithms) can maintain acceptable performance even when ignoring face features associated with race and gender.
Related Content: Facial Recognition System Can See Through Masks