Before any filter or prop appears, the software has to find where a face is and map points such as the eyes, nose and mouth. Telling who that face belongs to is a separate step, and UK data protection law treats it differently.
Pick a step to see what it involves. The bar shows how closely that step comes to identifying a person, which is what UK data protection guidance focuses on.
Based on ICO guidance on biometric data and biometric recognition. General information, not legal advice.
TWO NUMBERS WORTH KNOWING
From a few dots to a full mesh.
One research standard for marking a face, one model built for phone effects.
68 points
Points in a standard face mark-up
The 300 Faces in-the-Wild challenge at ICCV 2013 benchmarked landmark detection using a 68-point mark-up, tested on 300 indoor and 300 outdoor face images.
MARK-UP · 300-W, IMPERIAL COLLEGE
468 points
Points in a face mesh for effects
A 2019 model built for augmented reality effects predicts a 3D face mesh of 468 vertices from a single camera, at 100 to over 1,000 frames per second on phone graphics chips.
AR MESH · CVPR WORKSHOP 2019
Speed matters because the props must follow your face live. In 2001, Viola and Jones detected faces in 384 by 288 pixel images at 15 frames per second using only a grey-scale picture.
FOUR THINGS TO ASK AT A BOOTH
Enjoy the props, know the process.
Most party booths only need to find and map a face. These questions help you tell a fun filter from a system that recognises people.
Does it only place effects?
Props and filters need detection and landmarks, not a name. Ask whether the booth does anything beyond that.
Does it match faces to people?
Recognition compares your face with stored references. The ICO treats data used this way as special category biometric data.
Is anything guessed about you?
Some tools estimate age or other traits. The ICO says inferring ethnic origin can make data special category too.
Where do the photos go?
If you can be identified from a photo, it is personal information. Ask who keeps the pictures and for how long.
AI-generated image, not a photograph of a real person or place.
DETECTION VS RECOGNITION
Where a face is, or whose it is.
AI-generated image, not a photograph of a real person or place.
The ICO separates biometric verification, a one-to-one check such as a passport eGate asking whether you are who you claim to be, from identification, a one-to-many search asking who a person is.
Both need data that can uniquely identify someone. A plain photo is not enough on its own: the ICO says it is when specific techniques extract facial features that allow unique identification that the result becomes biometric data.
Questions
Questions people ask
How does a filter know where my eyes are?
First a detector finds the face, then a landmark model places points on features such as the eyes, brows, nose, lips and jaw. A common research mark-up uses 68 points; some AR models use a 468-point mesh.
Is face detection the same as face recognition?
No. Detection finds where a face is in an image. Recognition compares a face with stored references to verify or identify a person, which the ICO describes as biometric recognition.
Is my booth photo biometric data?
Not automatically. The ICO says a photo showing physical features is not enough on its own; it becomes biometric data when specific technical processing makes unique identification possible.
Why can filters run live on a phone?
Models are built for speed. A 2019 face mesh model reported 100 to over 1,000 frames per second on phone graphics chips, depending on the device and model version.
What is special category biometric data?
The ICO says it is biometric data used to uniquely identify someone. Organisations may only process it if they can identify a valid condition for doing so.
