The Chinese facial recognition company Hanwang claims it could recognize people wearing masks:
The company now says its masked facial recognition program has already reached 95 percent accuracy in tests, and even claims that it’s more accurate in true to life, where its cameras take several photos of an individual if the first try to identify them fails.
[…]
Counter-intuitively, training facial recognition algorithms to identify masked faces involves throwing data aside. A group at the University of Bradford released a study last yr showing they might train a facial reputation program to precisely recognize half-faces by deleting elements of the pictures they used to teach the software.
When a face recognition program attempts to identify a person, it requires a photograph of the individual to end up being identified, and reduces it right down to a lot of money, or vector, of amounts that describes the relative opportunities of features upon the face.
[…]
Hanwang’s system functions for masked faces by attempting to guess what all of the faces within its existing data source of photographs would appear to be should they were masked.
