3D Face Imaging in 40 Milliseconds 170
Roland Piquepaille writes "Computer scientists at Sheffield Hallam University, UK, have developed a new face recognition software which can produce an exact 3D image of a face within 40 milliseconds. A pattern of light is projected on your face, creating a 2D image, from which an accurate 3D representation is generated. This technology should speed airport check-ins, but it could also be used in banks or for checking ID cards as it allows full identification in less than one second."
Airport check ins??? (Score:5, Insightful)
Airport checkins take time because they are a security procedure. The "recognition" part of it takes seconds as it is - just swipe the passport or other form of ID. What takes the time is confirming that the traveller's luggage hasn't been modified, finding a decent seat on the plane, and labelling up the luggage they drop. I've never been held up because they couldn't figure out who I was. Ever.
Re:database? (Score:3, Insightful)
- Create biometric data out of face holding attached to person with card.
- Verify face to biometric data on ID card
- Verify ID card codes to central database for authenticity
Biometric verification can be done parallel to waiting for the data about whether the card is authentic or not.Heck, before the legislation setting up the ID cards passed, this speedy new biometric verification wasn't even broken to the public yet. All of the debate about verifying the biometric data in passing this ID card law was under the old school assumptions that it would be a friendly cashier or steward "verifying" that the person looked vaguely like the photo and was approximately the same height, weight, race, and age as listed in government records. Verifying the central records match the card is a problem that has to be taken care of regardless of the "speed" in verifying the biometric part with the human.
Of course, It's sort of disturbing that people knew this "recognition revolution" was in the pipeline but deliberately didn't say anything until after the legislative debate was over and all of the people worried about abuse of a system like this could be swept aside as science fiction freaks.
I'm still skeptical of the actual technique. (Heck, it's hard enough to verify "automatically" that a JPEG says HELLO WORLD when one has to deal with anti-aliasing and different fonts. How are they going to deal with people who gain or loose a lot of weight since the biometrics were taken.) But the real worry appears to be that the people developing it and supporting its development wanted it kept under wraps until all moral debate was silenced.
Re:database? (Score:2, Insightful)
Surely they could update the card each time it was verified. This way the changes would be extremely small and it should still pass.
There would still be the problem if someone didn't use their card in a long time, and gained/lost a lot of weight (for example), but I am assuming that the ID cards would be used for almost everything eventually (payments, keycards, etc), so the chance of not having your card verified for an extended period of time would be much less likely. Also, I suppose they could allow a slight difference between the scan taken and the data stored but it gets noted, and a proper match has to be made within 3 attempts.
There will obviously be problems if someone has an injury which drastically alters their face, etc, but they will probably be able to work around this eventually.
Who knows..
but it could also be used in banks (Score:4, Insightful)
I believe the biggest problem banks have is ordinary robberies. I can't say I've heard of any situation of someone using someone else's identity to wipe out their bank account.
Even in countries that do not have identification cards (and, after all, the english speaking world fell into this category until only fairly recently) I haven't heard anything to suggest banks are having/have had troubles identifying customer's correctly. (The identification collected when opening an account now is for Patriot Act purposes.) The lowly ATM with 4 digit pin is used successfully without identification (phishing is its main weakness.) Thanks to debit/credit cards, identification is now even less relevant to banking.
Why's this crap always being pushed on banks?
Re:This Is A Good Thing (Score:2, Insightful)
Congratulations. You've invented precrime.
Foolproof Security? (Score:5, Insightful)
FTA: giving a digital mapping of a face that would form part of a fool-proof security system.
I have yet to hear of a "fool-proof" security system of any sort. I've heard many security schemes touted over the years as being "unbreakable" or "fool-proof", and yet somehow someone manages to break them or fool them. This is what happens when PR hype takes over instead of substantive information.
This is an interesting technology that may have applications down the line, but it's still new, hasn't been given a wide range of testing, and appears to be something that can be spoofed by facial changes. The PR hack at the university needs to switch to decaf.
Re:database? (Score:3, Insightful)
If you can recognize your brother after he has gained weight, I'm sure the biometrics can recognize him, too. This is why the algorithms are programmed to have a margin of error. Like "it's 80% probable that this picture belongs to the person in question".
Re:database? (Score:4, Insightful)
If you can recognize your brother after he has gained weight, I'm sure the biometrics can recognize him, too.
Sure, as long as the matching algorithm is as sophisticated and accurate as the one in your brain, the one developed over millions of years of evolution of a social animal, and then fine-tuned by years and decades of personal feedback training.
The fact that you can recognize your brother after he has gained weight is evidence that it's possible to perform such recognition. It, however, in no way means that any matching algrothm we can create will match the feat.
In this particular case (weight gain), most facial recognition matchers probably won't be fooled, as long as the gain isn't too great. Not because the matching algorithms are anywhere near as good as the human brain, but because the matchers focus primarily on bone structure, which doesn't change with weight -- unless the change is so great that it hides previously visible structure under a smooth, round layer of fat.
In general, though, facial recognition technology is lousy. It's improving, but it's not remotely as good as what people can do.
Imaging != Identification (Score:4, Insightful)
Excuse my whilst I almost jump up and down with glee. I mean it's not as if a typical high res photograph can be taken in 1/300th of a second (given decent light) and a bunch of them can't be taken simultaneously with a bank of cameras - leaving almost all of the remainder of 1/25th of a second to quickly calculate a 3D model using the same digital photogrametry that's been around for years on a powerful enough system.
To put it in context, there have been camera systems that can film an actor "in 3D" - and then use that co-ordinate data to manipulate a 3D character - for TV use for the last half decade or so. By definition, at 25 frames per second, it too builds a 3D model within 1/25 of a second (40ms). The only difference is higher accuracy.
So, OK, they've come up with a new technique for projecting a dot pattern that makes it even easier to record a set of points than the old annoying stick on black dots method. Even so, quickly capturing a 3D image isn't radically new - a bank of cameras can capture it in far less time than 40ms and you can do the processing in a staggering 2-3 whole seconds as the person steps away before the next person steps up.
The slow bit has always been comparing a complex 3D model against a huge database and identifying matches when people move their facial muscles between each image.
Of note is the simple fact: This talks about how "fast" 3D imaging is now available (although it has been by years but we'll ignore that) which is a totally different concept to actually comparing that information against several million, if not hundreds of millions, of other entries.
OK, so fair enough, the article talks about comparing someone to a specific record to see if they're who they claim to be. Again, nothing that couldn't already be done with a bank of decent CMOS based cameras. They imply that this is "more" accurate (which I still dispute is any more accurate than a bank of 10MP CMOS cameras and traditional photogrametry) but make absolutely no reference to cracking the real problem of people's 3D facial structure changing as their muscles move, as they gain weight, etc.
So - they have a quick method of creating a 3D snapshot, which could be done already, and haven't solved any of the real problems that make a simple 3D snapshot useless (comparing against large numbers of possibilities, parameterizing 3D points that move as faces do). So absolutely nothing then? Sweet.
I wonder if I could come up with a new, different, but absolutely no better technology for something people also still can't do very well. If so, I wonder how much a breathy press release and vapid article would net me in grants for my research?
Re:database? (Score:3, Insightful)
I think you need ot shove that statistic right back up your ass where you got it from. Like I said in the post, that was a number for that State, but even still, the US population is less than 300 million. So I guess like everyone between the ages of 18-60 has a warrant for their arrest.
Re:database? (Score:4, Insightful)
As someone who has written, studied, tested, validated, and worked with 3D target recognition and facial recognition algorithms, I agree to a point. I only agree because it's always good to be skeptical of these things until the (unbiased) evidence has been presented.
That being said, the jump from 2D to 3D recognition is nothing but revolutionary in terms of recognition and verification accuracy. Beards, cosmetic surgery, and even swelling are not major impediments. They do reduce the match, but less than you'd think. There are areas of the face that don't change much even in such cases. 3D face recognition is more about measuring the shape of the skull. The skin generally follows the shape of the skull except where there are heavy fat deposits, or of course if you grow a beard. Unless you're changes were to the extent of a normal face to something like the Elephant Man or Eric Stoltz in "Mask" (no, not the Jim Carry movie), it's still going to do quite well.
And even then, the application here is verification. If the changes were enough to make it fail, that would just mean you'd have to go through either a secondary manual security check and/or update the biometrics. It's not like you could, say, pretend to be Tom Cruise and say your face changed. If you don't look like Tom Cruise with a beard or fatter face, and have the same shaped skull, and have the other security info to back you up, you're going to get your biometrics updated.
As a general comment (not specific to the parent post), I know /. likes to pick apart these things, but /. readers are not of so superior intelligence that they can come up with flaws in 2 minutes that experts in the area of research have not addressed, studied, or solved in years of research.