Well, everyone had their photos taken for their IDs, and these photos might’ve easily been stolen or leaked… Wonder if this would be enough for such a system.
Not really. The more pictures that are available for a person, the fewer false positives there are, if you only have one face on photo that can be confidently linked to you, you would only ever be identified by that exact angle and feature set. It would be useless outside of applications that specifically request that (and even then it’s really buggy, any Uber driver can attest to how shit facial recognition can be with limited info).
That is for sure a good question, although I would say it would present limitations.
Let’s put it this way: Id photos are very standard. Front facing, until the neck, white lights, white background. Now lets say everyone’s photo got leaked (or used) and the only source of photos for a certain photo is the Id one only. I didn’t study the matching algorithms, but I will say that variation for a certain subject under different circumstances increase the matching possibilities. If by any chance you try to match someone live and the only source would be id photos, my guess is it would present a big error rate. Possibility could be fine tuned, not sure up to how much.
Now lets consider what we have today. For a single person, I will bet there are on average more than 15 photos of themselves somewhere available (for those chronically attached to stuff such as IG of fb) and a lot of the times tagged as well (and manually as well, there was a time on fb this was a big thing). With this amount of comparison points, I would say the matching for sure works a lot better
Well, everyone had their photos taken for their IDs, and these photos might’ve easily been stolen or leaked… Wonder if this would be enough for such a system.
Not really. The more pictures that are available for a person, the fewer false positives there are, if you only have one face on photo that can be confidently linked to you, you would only ever be identified by that exact angle and feature set. It would be useless outside of applications that specifically request that (and even then it’s really buggy, any Uber driver can attest to how shit facial recognition can be with limited info).
That is for sure a good question, although I would say it would present limitations.
Let’s put it this way: Id photos are very standard. Front facing, until the neck, white lights, white background. Now lets say everyone’s photo got leaked (or used) and the only source of photos for a certain photo is the Id one only. I didn’t study the matching algorithms, but I will say that variation for a certain subject under different circumstances increase the matching possibilities. If by any chance you try to match someone live and the only source would be id photos, my guess is it would present a big error rate. Possibility could be fine tuned, not sure up to how much.
Now lets consider what we have today. For a single person, I will bet there are on average more than 15 photos of themselves somewhere available (for those chronically attached to stuff such as IG of fb) and a lot of the times tagged as well (and manually as well, there was a time on fb this was a big thing). With this amount of comparison points, I would say the matching for sure works a lot better
Ah, then I can sleep calmly I guess. There are some group pictures with me, but none have me tagged specifically so I guess that’s not as bad.