Dennis Delgado

The Dark Database series looks at the bias inherent in facial recognition systems. Using Open Computer Vision and the programming language Python, a facial detection algorithm was run on a series of films by directors of color. Each of the faces detected was stacked and centered according to a portrait template. The median pixel value for each detected face is calculated to create one single portrait of the film that serves as a kind of record of all the faces the facial detection algorithm was able to detect. Current studies show that facial recognition systems are less able to detect a face in an image when the skin tone of that face is not Caucasian. The undetected faces are then not enrolled in the systems’ database. As a result many of the systems are trained using datasets which contain fewer people of color, making those same systems inaccurate when recognizing or identifying individuals of color. The Dark Database is a kind of record of visibility and representation as seen through the eyes of artificial intelligence.