Comments - Facial Recognition On 60 Minutes

Did anyone see this segment about facial recognition on 60 Minutes?

I am skeptical of the speed they portray this happening in. Some of the opt in stuff, yeah I get that. Small database with good pictures is easy. The one I struggle with is the random pictures processed through Facebook and then, once the name is found, getting a Soc #. They make it look like he did it all in a few minutes.

Thanks for sharing. A few thoughts:

  • The face recognition drone is a pretty neat demo. Not very practical as it needs to be close to the person and the person needs to be relatively stationary but still neat.
  • 'Face prints' for home use makes sense because that's just matching, not recognition.
  • Gender and age is a lot easier than face. Like the difference between hitting walking and running a mile in sub 4 minutes.
  • The Europe point is important as privacy rights will be a much bigger issue than the US.
  • Btw, the main person talking, Joseph Atick, is one of the pioneers of biometrics (Visionics, Identix). I am not sure whatever became of his early efforts (most of all of this has flopped) and now he seems to have become a consultant.
  • The countermeasures (e.g., face paint) are just stupid.
  • App called 'facedeals' - that's worth checking out (here's a video and our discussion on it).
  • The experiment of taking photos on campus and matching to social networks is interesting and believable.
  • Facebook has more photos than government - funny and maybe true.
  • The FBI example at the end (11 minute mark) is interesting though this is focused on mugshots, not surveillance photos (which obviously have bigger issues).
  • The FBI equivocating on using social media photos is important (on the one hand, no law against, on the other no authorization to do so).

Good segment though no real or new claims on the video surveillance side.

Here's my question: Why are there so many issues with facial recognition when it comes to biometrics, but Facebook seems to have it down? You can upload a picture to Facebook, and it can recognize your friends and suggest that you tag them. Seems like that technology could be worked into cameras somehow right?

Granted all the photos are usually head on and people are usually looking at the camera, but it seems to work pretty well. It's not in real time, but it's in the few seconds it takes to upload the photo. What's the difference with this stuff and what facebook is doing (besides all the other variables with security cameras like light, etc). What am I missing?

Well, the photos being head on and people posing for the camera are two huge pluses! Let's not discount that.

Also, consider that the average smartphone camera is 5MP, 8MP, more? The average image from a smartphone is far higher quality than the one from a surveillance cameras. While smartphone images are worse in the dark, facebook photos are overwhelmingly well lit, posed, etc.

The other thing, and I am just speculating here, is that Facebook might use the social graph to better 'guess' at who is who. Cameras have no such metadata to use as a second data point. For example, if 3 people comes up as strong matches but only one person is a close connection of the uploader, Facebook can use that to make a better match. Also, Facebook likely knows what cities people live in and may know what location the photo was taken in. These can be used as well to increase accuracy. Again, I do not know what exactly Facebook does but these would be straightforward things to do and almost certainly boost accuracy.

Analyzing video on the fly at 7 - 30 FPS is much more rapid than 1 Frame over 1s? 2s? or more?