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?
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.
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.
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.
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
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.
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.
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.
It's looking highly likely this guy in the white hat could be the culprit. He was seen with a backpack earlier and in the newer high resolution image he does not appear to have his backpack any more. There could be a perfectly legitimate explanation for this: in the heat of the moment, his instinct was to run and leave his bag behind. Much like your instinct in a bad building or house fire is to run and save yourself. But then that would mean his bag would have been left behind and someone would have found it or reported it as being lost/authorities discovered it.
It's not looking good for whoever this guy is and if he's innocent, it would be in his best interests now we have a high resolution photo to turn himself in before a member of the public or worse members of the public find this guy and potentially harm an innocent person out of irrational anger. Another incriminating fact is Jeff Bauman the guy who lost both legs in the blast woke up heavily drugged in the hospital and said he had looked into the eyes of a bomber (which is no doubt what helped the feds narrow down the search). Lets keep level-headed about this and assist the authorities the best we can and let them do the investigating. If we all start jumping to conclusions and taking the law into our own hands we could do more harm than good. They'll find the culprit, it's only a matter of time.
... Well that and after thousands of hours of study by highly skilled experts combing thousands of sources he was one of two people deemed suspects. For all we know they have video of him dropping the bag, they only released a small portion of what they had to help gather tips from the public.
In the press conference earlier today, the chief investigator said they have footage of this suspect (suspect #2) dropping a bag near the bomb site and leaving it.
Or that he handed it to someone else, such as often happens when my ex and I take our son out together and hand the bag holding his stuff from one to the other.
Saw this on reddit (http://www.reddit.com/r/findbostonbombers/comments/1cn4wh/ne...) and posted a cropped/comparison version of it to my twitter...and immediately was told by people that it must be a fake. I guess people today are overly skeptical, but photoshop fakery of this quality is not trivial, especially just a couple hours after the official photos were released. The JPEG artifacting on the suspect is perfectly consistent with every other object on the same plane. More importantly, this is a new angle of the suspect's face not yet published, which means the faker would've had to find someone who looked like the suspect with the same clothes, or created a convincing composite even before placing it in the scene.
And perhaps most importantly...why? If it was a hoax it is surely one of the most boring hoaxes of all time, showing the suspect vaguely with no new interesting details except that he has no bag in the photo.
Edit: I just saw this Tweet from a digital manager at the New York Times. He's since admitted his mistake but this should be preserved for posterity as an example of how NOT to do digital forensics...it's as if he's never zoomed into a JPG before in his entire life https://mobile.twitter.com/jeremyzilar/status/32504129646716...
Why did this guy wear a distinct-looking hat backwards? It's really easy to pick him out of a crowd. Any witness will say, "Guy with backward white hat". Seems obviously stupid. Either they're innocent or they had a foolproof escape plan.
Why do you think criminals are masterminds? The perpetrator of the truck bombing of the World Trade Center -- a plot far more complicated and would have been more destructive than 9/11 had they parked the truck differently -- was captured when he attempted to get the deposit back for the truck used in the bombing: http://gothamist.com/2013/02/26/today_is_the_20th_anniversar...
Just because someone's willing to be a indiscrimate murderer doesn't mean that they're geniuses or well put-together
I wasn't considering "smart" interpolation methods, I was assuming the best available interpolation was already implemented in Chrome. I think it would load faster with client-side interpolation actually, 17 MB takes a long time on most connections.
But I don't see why client-side can't do "powerful resizing". Not since <canvas> appeared, anyway.
>I think it would load faster with client-side interpolation actually, 17 MB takes a long time on most connections.
You'd still have to move the 5-6 MB original, and then it would hog a lot of people's browsers to work on the image in the canvas (depending on client CPU/memory).
>But I don't see why client-side can't do "powerful resizing". Not since <canvas> appeared, anyway.
Who said it cannot do? I said the options aren't available. The guy did a simple resize. Should he also spend hours coding smart interpolation on javascript instead of just putting the image up?
For most connections, a 17MB download is eaten for breakfast, anyway.
Like so (edit: note that you'll have to edit the image first to make it grayscale, and even then this uses a lot of memory...):
$ python fftresize.py suspect-number-2.JPG 2.0
I just uploaded the image to S3 to share. The script isn't great (grayscale only) because it was written for a blog post to demonstrate the interpolation method, but I might as well work on it this weekend.
50 comments
[ 3.0 ms ] story [ 111 ms ] threadHow 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?
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.
You have lots of other information, so you don't need to check every human on the planet.
Bottom left corner.
Edit: Higher Res
HN needn't fear, though. Reddit's on it.
Ref: http://graphics8.nytimes.com/images/blogs/thelede/posts/susp...
http://robertgawron.blogspot.com/2011/01/super-resolution-al... More details on it here: https://github.com/RobertGawron/supper-resolution/wiki
It's not looking good for whoever this guy is and if he's innocent, it would be in his best interests now we have a high resolution photo to turn himself in before a member of the public or worse members of the public find this guy and potentially harm an innocent person out of irrational anger. Another incriminating fact is Jeff Bauman the guy who lost both legs in the blast woke up heavily drugged in the hospital and said he had looked into the eyes of a bomber (which is no doubt what helped the feds narrow down the search). Lets keep level-headed about this and assist the authorities the best we can and let them do the investigating. If we all start jumping to conclusions and taking the law into our own hands we could do more harm than good. They'll find the culprit, it's only a matter of time.
Or somebody would have stolen it. There was even looting in the area just after the bombings.
(Personally I very rarely take the bother to declare something being lost with the authorities).
And perhaps most importantly...why? If it was a hoax it is surely one of the most boring hoaxes of all time, showing the suspect vaguely with no new interesting details except that he has no bag in the photo.
Edit: I just saw this Tweet from a digital manager at the New York Times. He's since admitted his mistake but this should be preserved for posterity as an example of how NOT to do digital forensics...it's as if he's never zoomed into a JPG before in his entire life https://mobile.twitter.com/jeremyzilar/status/32504129646716...
Just because someone's willing to be a indiscrimate murderer doesn't mean that they're geniuses or well put-together
http://www.eliteraspberries.com/images/suspect-number-2-749.... (17MB)
Besides, server side he has access to powerful smart resizing programs and algorithms. Client size, not so much.
But I don't see why client-side can't do "powerful resizing". Not since <canvas> appeared, anyway.
You'd still have to move the 5-6 MB original, and then it would hog a lot of people's browsers to work on the image in the canvas (depending on client CPU/memory).
>But I don't see why client-side can't do "powerful resizing". Not since <canvas> appeared, anyway.
Who said it cannot do? I said the options aren't available. The guy did a simple resize. Should he also spend hours coding smart interpolation on javascript instead of just putting the image up?
For most connections, a 17MB download is eaten for breakfast, anyway.
http://code.google.com/p/fftresize/source/browse/fftresize.p...
Like so (edit: note that you'll have to edit the image first to make it grayscale, and even then this uses a lot of memory...):
I just uploaded the image to S3 to share. The script isn't great (grayscale only) because it was written for a blog post to demonstrate the interpolation method, but I might as well work on it this weekend.