Nope, not Viola–Jones. I'm describing the initial fast test that is a preprocessor to pretty much all facial recognition, before the FR model processes anything

Nope, not Viola–Jones. I'm describing the initial fast test that is a preprocessor to pretty much all facial recognition, before the FR model processes anything. FR models process potential face rectangles, and this method goes after the preprocessing. This works on all the top FR models. I wrote one of the global top 5, with over a decade running as certified by the annual NIST FR Vendor's Test.