4 ms·

Tech-related question: How far are we from having sufficient facial recognition technology that we could clip a photo like this, drop it into Facebook image se

by scarmig 14y ago

Tech-related question: How far are we from having sufficient facial recognition technology that we could clip a photo like this, drop it into Facebook image search (or whatever), and the person would come up? Is that feasible with today's technology?

throwaway_21 14y ago
This is possible now. You need a database of faces. You could store and search them without compression in like clouds. Or you can compress them. 7 billion isn't that many pictures. One problem is the databases of tagged faces typically have passport style photos and not sides. It would be easier with 3d type data of faces, like when we start using depth cameras at the DMV too. Or, we will develop sufficient methodology to figure it out well enough from the existing photos... depth from a single image. It sounds sort of disturbing to me. Google reverse image search for faces... that team has done it or could do it easily. But privacy.
- ricardobeat 14y ago
  face.com offered that before they were sweeped by facebook, and, if my memory is not failing me, it could match different head orientations and give you back the angles.
- eru 14y ago
  What makes you think we need to check 7 billion pictures? You have lots of other information, so you don't need to check every human on the planet.
  - vidarh 14y ago
    Post facto for a specific incident, you are right, but you still need to assemble that database from somewhere. I got the impression he is talking about the general case, though.
    - eru 14y ago
      OK. I'd go with more than one picture per user then, though.
dmishe 14y ago
Well facebook can recognize your friends right now, so I would think if you give them a couple of hours they could find a match on a random photo. Provided that photo has actual face
- Resident_Geek 14y ago
  Sure, but recognizing a picture of one of your friends is a matter of searching O(100) people. Recognizing a picture of any Facebook user is a matter of searching O(100 million) people. It's a totally different problem, and there's no reason to think Facebook, which hasn't solved the smaller problem perfectly, could solve the larger problem at all.
- lostlogin 14y ago
  Our daughter is 2. I recently dropped some photos of her into iPhoto and told it who she was with about 20 of them. I have probably 60gigs of photos in its library, mostly of people. More recently I loaded my wife's pregnancy US scan into iPhoto, including some stills from the 4D scan. iPhoto popped up a message asking if this was my daughter.
crucifiction 14y ago
Yes, profile pics (e.g. side of face) are hard but it is definitely feasible.
mazsa 14y ago
Yes: https://news.ycombinator.com/item?id=3640291 https://news.ycombinator.com/item?id=3640291
heifetz 14y ago
it's probably just a matter of time, or maybe it's already happened, when the police can go to Facebook or Google and ask them to use their computing resources and data to look up someone based on a picture. In this case, I hope they do, and there is some positive result. I'll love to see this link back to the culprits facebook profile, with his email address, phone number, and all his messages.

by scarmig 14y ago

throwaway_21 14y ago

This is possible now. You need a database of faces. You could store and search them without compression in like clouds. Or you can compress them. 7 billion isn't that many pictures. One problem is the databases of tagged faces typically have passport style photos and not sides. It would be easier with 3d type data of faces, like when we start using depth cameras at the DMV too. Or, we will develop sufficient methodology to figure it out well enough from the existing photos... depth from a single image. It sounds sort of disturbing to me. Google reverse image search for faces... that team has done it or could do it easily. But privacy.

ricardobeat 14y ago
face.com offered that before they were sweeped by facebook, and, if my memory is not failing me, it could match different head orientations and give you back the angles.
eru 14y ago
What makes you think we need to check 7 billion pictures? You have lots of other information, so you don't need to check every human on the planet.
- vidarh 14y ago
  Post facto for a specific incident, you are right, but you still need to assemble that database from somewhere. I got the impression he is talking about the general case, though.
  - eru 14y ago
    OK. I'd go with more than one picture per user then, though.

dmishe 14y ago

Well facebook can recognize your friends right now, so I would think if you give them a couple of hours they could find a match on a random photo. Provided that photo has actual face

Resident_Geek 14y ago
Sure, but recognizing a picture of one of your friends is a matter of searching O(100) people. Recognizing a picture of any Facebook user is a matter of searching O(100 million) people. It's a totally different problem, and there's no reason to think Facebook, which hasn't solved the smaller problem perfectly, could solve the larger problem at all.
lostlogin 14y ago
Our daughter is 2. I recently dropped some photos of her into iPhoto and told it who she was with about 20 of them. I have probably 60gigs of photos in its library, mostly of people. More recently I loaded my wife's pregnancy US scan into iPhoto, including some stills from the 4D scan. iPhoto popped up a message asking if this was my daughter.

crucifiction 14y ago

Yes, profile pics (e.g. side of face) are hard but it is definitely feasible.

mazsa 14y ago

Yes: https://news.ycombinator.com/item?id=3640291 https://news.ycombinator.com/item?id=3640291

heifetz 14y ago

it's probably just a matter of time, or maybe it's already happened, when the police can go to Facebook or Google and ask them to use their computing resources and data to look up someone based on a picture. In this case, I hope they do, and there is some positive result. I'll love to see this link back to the culprits facebook profile, with his email address, phone number, and all his messages.