How does mass face recognition work based on a large database?

Question

Algorithms

Let's say the FB volume base needs to be recognized in real time.
How does the fastest recognition work, maybe you can create a set of simplified face metrics, a sort of hash?

Answer 1

Most likely, classification in the form of a set of experts works here. Let's say we have a face. We set him on the first classifier, he says that the face belongs to type No. 1 (and there can be 10 such types, for example). Next, we set the face on the second classifier, which also classifies one of the 10 types of faces (already on a different basis), etc. the main thing is that all classifiers should be trained on orthogonal face features. we get some face metric. If there are enough such classifiers (for example, 10-15), then we can thus classify 10 ^ 10 - persons - and this is a lot :) ... then we just have to select all the faces with the same hash, and compare with the current photo (there will be quite a few such faces) with a more resource-intensive algorithm.
PS These are all my guesses, in reality it can be much more complicated.

Answer 2

More difficult.
Read: www.face-rec.org/interesting-papers