Mostly poorly.
The idea is to encode a set of facial features and then search a database of those those features. This technique was developed for fingerprints, and works really well for that.
However, fingerprints are essential “line” or “no line” in each pixel of the image. That’s a very low information density. Faces are much more complex, and the metrics like distance between the eyes and nose, have proven much less unique and powerful than fingerprint features.
Latest Answers