The idea in one sentence
Instead of guests looking for themselves in thousands of photos, the photos are searched for the guest's face.
Step by step
- Finding faces. When photos are uploaded, a face detector marks every face in every photo, including small faces at the back of group shots.
- Making face signatures. Each face is straightened and turned into a face signature: 512 numbers that describe its structure. Similar faces have similar numbers. A signature can't be turned back into a picture.
- The selfie. The guest's selfie gets a signature the same way.
- Comparing. The selfie's signature is compared with every face in the event. With modern hardware, 250,000 faces are compared in well under a second.
- Deciding. Clear matches are shown. Borderline faces are shown only if they also match the guest's face in other photos they were confidently found in.
How accurate is it?
For clear, front or three-quarter faces, accuracy is very high. In our own tests on crowd-sized, partly blurred and darkened faces, fewer than 1 in 100 matches was wrong. What reduces accuracy:
- Dark sunglasses or masks in the selfie or the photos.
- Faces turned fully away, or hidden behind other people.
- Very dark or strongly backlit photos.
- Twins and siblings who look very alike.
A good system is tuned to leave a photo out rather than show it to the wrong person, and lets guests remove mistakes with one tap.
What about privacy?
Questions to ask any provider:
- Who sees the photos? Each guest should see only their own photos, never the whole event.
- What is stored? Selfies, face signatures and guest details should be deleted on a set date.
- Is there consent? Guests should agree before their selfie is used, and be able to ask for deletion.
Read how Vortex Media Event Photos works and our privacy policy.