Researchers Create 'Master Faces' To Bypass Facial Recognition (vice.com) 38
An anonymous reader quotes a report from Motherboard: Researchers have demonstrated a method to create "master faces," computer generated faces that act like master keys for facial recognition systems, and can impersonate several identities with what the researchers claim is a high probability of success. In their paper (PDF), researchers at the Blavatnik School of Computer Science and the School of Electrical Engineering in Tel Aviv detail how they successfully created nine "master key" faces that are able to impersonate almost half the faces in a dataset of three leading face recognition systems. The researchers say their results show these master faces can successfully impersonate over 40 percent of the population in these systems without any additional information or data of the person they are identifying.
The researchers tested their methods against three deep face recognition systems -- Dlib, FaceNet, and SphereFace. Lead author Ron Shmelkin told Motherboard that they used these systems because they are capable of recognizing "high-level semantic features" of the faces that are more sophisticated than just skin color or lighting effects. The researchers used a StyleGAN to generate the faces and then used an evolutionary algorithm and neural network to optimize and predict their success. The evolutionary strategy then creates iterations, or generations, of candidates of varying success rates. The researchers then used the algorithm to train a neural network, to classify the best candidates as the most promising ones. This is what teaches it to predict candidates' success and, in turn, direct the algorithm to generate better candidates with a higher probability of passing. The researchers even predict that their master faces could be animated using deepfake technology to bypass liveness detection, which is used to determine whether a biometric sample is real or fake.
The researchers tested their methods against three deep face recognition systems -- Dlib, FaceNet, and SphereFace. Lead author Ron Shmelkin told Motherboard that they used these systems because they are capable of recognizing "high-level semantic features" of the faces that are more sophisticated than just skin color or lighting effects. The researchers used a StyleGAN to generate the faces and then used an evolutionary algorithm and neural network to optimize and predict their success. The evolutionary strategy then creates iterations, or generations, of candidates of varying success rates. The researchers then used the algorithm to train a neural network, to classify the best candidates as the most promising ones. This is what teaches it to predict candidates' success and, in turn, direct the algorithm to generate better candidates with a higher probability of passing. The researchers even predict that their master faces could be animated using deepfake technology to bypass liveness detection, which is used to determine whether a biometric sample is real or fake.
Apples should bring back thumb readers on all phon (Score:4, Insightful)
Facial recognition ... never was a huge fan. Always trying to unlock when you pick up your phone to stand up and walk away from your car or desk. Doesn't work with masks. Pain in my ass compared to the old fingerprint readers.
Re:Apples should bring back thumb readers on all p (Score:4, Funny)
If the facial recognition process/fingerprint reader is causing you a pain in the ass, you're using it wrong.
Re: (Score:2)
Steve Jobs, is that you?
Re: (Score:2)
But Face ID works when I just got out of the shower or just washed dishes, when I'm cold, after I applied hand cream, and when I'm over fifty - all those cases when the fingerprint reader doesn't work. It does disown me when I mask, though.
Re: (Score:3)
Re: (Score:2)
Fingerprints are insecure - they fail in simple cases (e.g., washing your hands) routinely. It's also extremely easy to spoof by anyone with access to your fingerprint.
FaceID using a 3D map is more secure, and less spoofable. You can't use a simple picture like normal faci
Re: (Score:2)
Fair enough...reason to come up with an alternative, although I don't think that is a good enough reason to ditch it entirely.
The only time I ever have had any problem using touch id is when my hands are wet, which is a lot less often than I'd have a problem using face id, between wearing a mask on public transit and having a degenerative motor disability that prevents me from holding a phone steady enough for face id to register, I'm sticking with my iphone 6+.
Re: (Score:2)
> the sadistic, underhanded scumbags at Google
You're too kind to them.
Master face pros and cons (Score:2, Funny)
On the one hand you'll be able to unlock anyone's phone but on the other hand you'll get shot as a terrorist at the airport. Swings and rundabouts I guess.
Re: (Score:2)
Unlock 40 % of the phones that have the feature. 60 % it will not, plus some use more than just faces (increased data the article talks about) so that number increases. In short what problem is this solving that other methods don't?
Naah... Don't worry. (Score:2)
Master face is an old white dude.
The paper also notes that white males over the age of 60 in the University of Massachusetts' Labeled Faces in the Wild (LFW) dataset tended to be less varied compared to younger groups, so much of that group could be covered by a single older master face.
Additionally, only two of the nine master faces created were female, which the paper notes matches the "much lower frequency" of female faces in the LFW dataset (22 percent.)
You'll be fine. Mostly. [wikipedia.org]
Just remember to be an ACTUAL threat to government, democracy and the police. [nwaonline.com]
Then you get house arrest and spend more time at home - instead of a fractured skull and spending a month at a hospital while being attacked in the media by the Fascist-in-chief. [apnews.com]
Progress waits for noone (Score:1)
From "master face", to "master race". Keep at it, bois
Skeleton Face? (Score:4, Funny)
If generated after the fact and not for a master system, wouldn't that be a skeleton face?
Obvious solution: (Score:2)
Having an old face is now illegal! ;)
With a mask on? (Score:2)
Master face wears no "Geezer Goggles". Or beards. (Score:2)
From TFA:
The "master key" faces tended to be older, and didn't have glasses or facial hair.
Re: (Score:2)
I wonder what the skew in ages is in facial databases; I wouldn't be surprised if old geezers (like me) aren't under-represented, i.e. insufficient training data--which would mean all old men look alike. (Me, I've got a beard and long hair, so not so much.)
Re: (Score:1)
Why take any chances what so ever, get this https://www.securityprousa.com... [securityprousa.com] instead full face ballistic helmet with googles, don't let no virus laden bullets get by your face mask.
Not surprised (Score:4, Interesting)
Artificial neural net image recognition doesn't work the same was as the human brain (whatever that may be) as it can be fooled even by single pixels:
https://www.bbc.co.uk/news/tec... [bbc.co.uk]
Obviously t the state of the art has moved on since 2017 but the core algorithms haven't and at their heart they're just dumb statistical recognisers which work well for a lot of things but can fail badly and suddenly and with no obvious reason. The fact that no one really has a full grasp of how the weightings at the neuron level lead to successful the recognition doesn't help either.
Re: (Score:2)
You simply don't know that the human brain can not be fooled by single pixels, because we have no way of backing out the error term from the human neocortex to directly construct the pathological example.
Additionally, human recognition does have some surprising failure boundaries, but it tends not to happen at the pixel level. Human vision at the bottom is more oriented toward edges and motion.
Today science news flash: Pixel-based visual system has trivial pixel-based failure boundaries.
Future science news
Re: (Score:2)
In amongst that waffle you had a few points - yes the human brain is subject to optical illusions, but thats down to the way its evolved to produce the most information from a scene and sometimes it gets it wrong but the failure cases are irrelevant in the enviroment we evolved in. Its not a bug or hardware fault, unlike the pixel failure of ANNs which could in a safety critical situation be lethal - eg self driving car suddenly thinking that wall on the other side of the junction is actually a road.
What's next? (Score:2)
Which kind of facial recognition? (Score:2)
Re: (Score:1)
or systems that require head movement
TFA: The researchers even predict that their master faces could be animated using deepfake technology to bypass liveness detection
So? (Score:2)
Without reading tfa, this undoubtedly requires access to the database, in order to do some kind of learning to find the master faces. Without access to the database, you're SOL.
However, it does point out a root problem with most if not all learning methods: there is a lot of freedom in very high dimensional spaces, and finding one set of criteria without knowing exactly what they are doing is likely to all adversarial attack by taking advantage of the large number of dimensions.
about those Masters (Score:2)
I assume they made both a Master and a Missy face.
Re: (Score:2)
I think they said their failure rate with female faces was higher.