Face recognition systems have been spreading owing to advances in AI. It relies on machine learning ( a subfield of AI ) in which computers teach themselves to do tasks that their programmers are unable to explain to them explicitly. They are rewarded when they correctly identify a face, and penalised when they do not. Thus it can be taught to recognise images that contains faces from those that do not. Once it has an idea of what a face consists of, it begins to distinguish one face from the other. The specifics vary depending on the algorithm. Usually, it involves mathematical representation of a number of crucial anatomical points, such as the location of a nose relative to the other facial features, or the distance between the eyes. In lab tests, such systems could be very accurate.

There are many weaknesses in computer vision which is nothing like human vision.

CV Dazzle

Make up can, fool face recognisers. Bright colours, high contrast, graded shading and asymmetric stylings confound an algorithm’s assumptions of what a face looks like.

Hyperface aims to hide faces from dozens of fakes. The idea is to disguise the real thing.

Baseball cap fitted with tiny light-emitting diodes that project infra-red dots onto the wearer’s face could be used.

FaceNet

It is a face recognition system developed by Google. The researchers find that the right amount of infra-red illumination prevent a computer from recognising that it was looking at a face at all.

More sophisticated attacks are possible by searching for faces that are mathematically similar.

Adversial Machine Learning

Training one algorithm to fool another is known as adversial ML. Images are created to mislead CV. Innocuous looking abstract patterns printed on paper and stuck onto the frame of a pair of glasses could convince a CV system that a male research worker is a female actress.

Thus all these systems have constraints.