Face Scan With Glasses, Masks and Head Coverings
TIME&SPACE · Event Technology
What actually affects a face match is smaller than people expect. Glasses and head coverings are mostly fine. What breaks it is anything across the middle of the face.
Face Scan With Glasses, Masks and Head Coverings
In short: Face matching reads the geometry between the eyes, nose and mouth. Glasses and head coverings usually leave that intact. Anything covering the centre of the face does not, and sunglasses are the most common culprit at events.
This comes up as a practical question at the gallery and as a fairness question long before that. Both deserve a straight answer.
What the match is actually reading
A face match is not comparing pictures. It builds a mathematical description of the spatial relationships across a face, and compares descriptions.
That means most of the signal lives in the middle: the region spanning the eyes, the bridge of the nose and the mouth. Whatever leaves that region visible tends to work. Whatever covers it does not.
This is why the intuitive list of problems is wrong. People assume hair, hats and glasses are the issue. They mostly are not.
The things that are usually fine
Prescription glasses. Clear lenses leave the eye region visible. Matching works normally. Heavy frames can reduce confidence slightly but rarely enough to matter.
Head coverings. A hijab, a turban, a cap or a headscarf frames the face without crossing it. The geometry the system reads is untouched. This one is worth stating plainly, because people wearing them often assume they will be excluded and they are not.
Beards, makeup, a new haircut. Surface appearance changes far more than the underlying geometry. Somebody who grew a beard between the selfie and the event is normally still matched.
Hats. Unless the brim throws the whole face into shadow, which is a lighting problem rather than a covering one.
The things that genuinely break it
Sunglasses. The most common real problem at events, especially outdoor and daytime ones. Dark lenses remove the eye region entirely, which is the single most informative part of the face.
Face masks. Covering the nose and mouth removes a large share of the geometry. Matching becomes unreliable.
Anything held in front of the face. A drink, a phone, a hand mid-laugh. Very common in candid event photography.
Extreme angle. Not a covering, but the same effect. A face turned far enough away presents a partial view.
What this means in practice
For guests, the useful instruction is short: take the selfie without sunglasses, looking at the camera, in reasonable light. That is the whole thing.
For organisers, the honest position is that a guest photographed all evening in sunglasses may be harder to find, and no amount of processing changes that. Say so on the gallery rather than leaving them to conclude the system is broken.
For the sceptical question underneath all of this, whether a system works less well for some groups than others: it is a real and serious question, it is not answered by anecdotes, and it is not answered by a vendor saying it is fine. It is answered by testing across groups and publishing what you find. Aggregate accuracy figures do not substitute for that, and anyone quoting only an aggregate number has not answered the question.
Do not loosen the threshold to compensate
When matches fail, the tempting fix is to accept weaker matches. It returns more results, including other people's photographs.
For an event gallery that trade is a bad one. Returning fewer, correct results is the right failure mode, and the reasoning is set out in how facial recognition accuracy works.
The short version
Glasses and head coverings are mostly fine because they do not cross the centre of the face. Sunglasses, masks and anything held in front of the face are what actually break a match. Tell guests to lose the sunglasses for the selfie, and be honest on the gallery about what will not be found.
Founder, TIME&SPACE