Fooling Face Recognition with Infrared Light

Yet another development in the arms race between facial recognition systems and facial-recognition-system foolers.

BoingBoing post.

Posted on March 27, 2018 at 9:35 AM • 29 Comments


Not the droids you are looking forMarch 27, 2018 11:08 AM

Would a similar technique work with license plates?

meMarch 27, 2018 11:47 AM

@Not the droids you are looking for
Sure it works with license plates, but it also draws attention that you're doing something unusual, so you get extra attention by anyone watching live video feeds, like cops with instant license plate readers in their cars that throw an error when they cannot ready your plate, or security agents watching live feeds of people and notice the one person in the crowd looks like they have a supernova instead of a human face under their hat.

J. DillingerMarch 27, 2018 11:54 AM

Universal biometric identification is an essential goal of the globalist government movement. (...and eliminating cash money...)

Facial ID is the golden fleece because it has no language barriers, is readily collected and examined at a distance and presumably reliable.

So, when the "good guys" can defeat it with cheap infrared illumination I am all for it.

Since most cameras are up high, to prevent tampering, wearing a long billed baseball hat helps. I have read goggle like sunglasses can defeat facial ID systems fairly reliably. I would think beards for men and extreme makeup for women might help also.

When I moved to another state I had to get a new drivers license. Oddly, the real police officer who did the photo was very insistent I NOT smile and point my face directly at the camera without the slightest angle. Months latter I found that's because they enter all driver pics into the national criminal data base, just in case.

Grrrr. To the NWO ruling class we are all un-arrested criminals. Even police.

wumpusMarch 27, 2018 1:21 PM

Iris recognition uses IR as well (it lights up the blood vessels more). I have no idea if this is a good way to defeat it, but presumably IR-reflective glasses/contact lenses would do wonders (might even defeat some facial recognition).

WinterMarch 27, 2018 2:17 PM

@J. Dillinger
"globalist government movement"

I did not know this one. And it is a movement at that? I thought it was the Illuminati that were behind all these, or the Templars.

Bauke Jan DoumaMarch 27, 2018 3:41 PM


"I thought it was the Illuminati that were behind all these, or the Templars."

I call strawman argument.

Tin HatterMarch 27, 2018 8:16 PM


Re: Infrared filter.

A thin film of tin dioxide or silver on glass reflects infrared, but allows most of the visible light. I would start there with coated sunglasses.

There should be some effective ways to use wire mesh to reflect or disrupt an infrared sensor. Anything that conducts electricity also blocks and disrupts IR

There is already a small infra illuminator on the market that attaches to the bill of a baseball cap that will radiate a kaleidoscope of IR on your face which will really mess up any IR sensor, but is invisible to the human eye.

I think the key will be disguising the eyes and mouth to defeat facial ID and infrared disguise and disruption will be a key component.

They want our faces, but we still have a few options to keep it from them. Sometimes.

EdMarch 27, 2018 8:21 PM

If the camera maker installs an IR block filter in front of the imaging sensor then the spoofed IR light will not confuse the detection algorithm.

RonKMarch 28, 2018 2:03 AM

@ Ed

> an IR block filter

Exactly my first thought.

But even worse, when looking at the paper, I don't get the impression that the experimenters used a database which included the person wearing the device. Did I miss something?

So I don't see how this reflects on a future where, most likely, each and every individual will be in the facial recognition database which one is trying to fool.

Still, it's an interesting idea. Just really over-hyped.

Robert.Walter March 28, 2018 4:59 AM

I just scanned the paper but first thoughts:
- what’s with the discussion of injuries and sun burn from IR? Thought it was UV that did this. (Obviously there would be an upper limit to IR too but the example is unclear at best and misleading at worst.)
- what did they test this against? Again, I scanned but it did not seem they did any comprehensive testing.
- also given the Apple iPhone X is the fastest growing, FR system, did the researchers test against that, and if not, why not? Did they cherry pick a early FR system to be able to make headlines?

I’m not saying that what they propose won’t happen, I’m just wondering if it is all that much to worry about.

Who?March 28, 2018 5:58 AM

@ Ed, RonK

Why wouldn’t they use an IR block filter in front of the sensor?

Because an IR block filter would make these surveillance devices useless by night.

Academia does not stop surprising me with the low quality of its current research. We should seriously consider stopping this game based on "publishing at any price" before it is too late (if it isn't too late right now for that community of oligophrenic egotists).

I had been using webcams to check TV IR remote controls for years. Using infrared light to hide faces had been a well-known technique for more than a decade. I am not talking about the IC, using infrared light to block surveillance cameras has been known to any guy that is not technologically blinded.

wiredogMarch 28, 2018 6:19 AM

@@Not the droids you are looking for @me

In Virginia it's been illegal to obscure the license plate for decades, and any cover, even a clear one, over a license plate counts as an obscuration. In 1988 I was riding with a guy when he got pulled over for "license plate obscured" when he had a clear plastic cover over it.

RonKMarch 28, 2018 7:43 AM

@ Who?

> using infrared light to block surveillance cameras has been known to
> any guy that is not technologically blinded

Yes, and it has been previously posted on this blog.

The research in question, however, supposedly has more to do with fooling the algorithm into either not recognizing one's face as a face, or into recognizing one's face as someone else's face. This could be interesting research. However, the methods used to check it seem to me to not reflect the reality of a dystopian future where everyone's face is in the database being used for recognition.

And in addition, ASAICS the paper did not even try to address how easily mitigation techniques could be introduced into the algorithm(s) and how effective such techniques could be. (My guess is that this is easily mitigated.)

VinnyGMarch 28, 2018 8:05 AM

Regarding the "how" of their testing, I got the impression that the authors compared the output of algorithms presumed to be used by the FR target system with and without IR masking (hence "white box".) I would prefer a full-scale, "through the system" audit, but imo this is a good start.
Re potential weaknesses cited by others, I think it important to distinguish between the scenarios they posited: dodging & impersonation. I think most of us here who would like to find ways to avoid successful FR are most interested in dodging. "Dodging" was further separated into two domains, termed somewhat obscurely as: (1) failing the landmarking module; (2) avoiding the preprocessing model. I think the first means that the system does not recognize the presence of a face in the image at all; and the second that it does recognize a face, but cannot match it to any person in the database (except possibly by a random match?) using the facial measurement markers. They mention a "Yamada method" (not much elaborated upon) to accomplish "1." This seems to entail (assumption on my part) overwhelming the cam sensor with IR. Possibly doable with a sufficiently intense IR source aimed directly at the sensor? The method proposed by the researchers seems to be altering the facial measurements "seen" by the system to be unrecognizable as a face.
As @Who? mentioned, an IR filter would render the cam useless in low light conditions. Realistic solutions to that issue would appear to require wholesale replacement of cams, which would be quite expensive at this point. I believe there is going to be an ongoing, escalating, iterative battle between the forces who want ubitquitous FR as part of a pervasive surveillance society, and those who are diametrically opposed to it, and intend at the least to avoid it personally. Research of this kind is imo invaluable in obstructing said surveillance state. If it is indeed a losing battle, perhaps it can be made a protracted and expensive one for the victors. Who knows, maybe they will run out of discretionary funds and/or motivation for this kind of thing somewhere along the way...

PeteMarch 28, 2018 8:43 AM

Thanks for the reminder.

My DL expires next year and I need to grow a beard for the new RealID photo. My current DL has a photo from 1996 on it, since renewal via snail-mail was pretty easy.

echoMarch 28, 2018 9:12 AM


If we assume survellience is about power perhaps a way to counter this is to create a legislative framework of openess such as mandatory publication of tax records and business ownership and perhaps more importantly mandate establishment types have their image available. I mention this because the kinds of people who wield influence often like to keep a low profile. The odd financial scandal can flush them out and they hate this. What's sauce for the goose is sauce for the gander so if they hate this exposure then reframe discussion to put them under the spotlight and see how long pervasive face recognition lasts.

Not the droids you are looking forMarch 28, 2018 11:36 AM

@Wiredog: I was not talking about anything covering the license plate, rather creating a frame that goes around the plate, similar to the one installed by the dealer when you buy a new car or by people who want to brag whatever.

Instead of having text, the frame would have a series of LEDs to blind the sensor.

vas pupMarch 28, 2018 2:50 PM

What about full face latex mask similar to those done by Tony Mendes (CIA) to take home US diplomats out of Iran after hostage situation with fake papers/pasports matching photos with those masks?
Is any face recognition technology could recognize those masks?

Security SamMarch 28, 2018 11:14 PM

Who would have thought it possible
That face recognition one could bypass
Merely by illuminating a given object
With a sealed tube full of ionized gas.

VinnyGMarch 29, 2018 7:21 AM

@Not the droids you are looking for re: dodging license plate recognition systems
I see some difficulties in what you propose. In regard to one of wiredog's legitimate objections, you would need to identify an illumination technique that would disable recognition by some systems (e.g. toll booth monitors) but not others (patrol car live camera & plate look-up systems.) You might try illumination that is only powered up when in range of the first type of system, but both manual & automated triggering of the "dodging" system present problems. Most stationary cameras of the first type seem to be triggered by radar - perhaps tying in to a modifiable detector (e.g. Valentine products) could identify nearly all of those. But the cop in the patrol car who wants to run your plate number is no doubt also using radar to monitor your speed. Manual triggering would require previous knowledge of the locations of cameras you wish to avoid, and activating the dodge in time while driving. Navigating toll approaches in heavy traffic on major interstate exchanges presents enough of a safety issue for me as is, thank you :) It would probably be at least as easy to find someone with a vehicle of the same DMV description and forge a copy of their plate. Designing a mechanical system that quickly swaps plates while the vehicle is moving should not present any absurdly difficult obstacles, if that proved advantageous. There is the issue of the reflective "watermarking" that is featured on most if not all currently issued plates in the US, but I'm pretty confident that could be duplicated with significantly less difficulty than the use of IR to selectively confuse plate readers. I wouldn't be at all surprised if there was already an underworld cottage industry supplying that material.

VinnyGMarch 29, 2018 7:30 AM

@vas pup re: latex face masks - recognition. The answer probably depends on the extent to which the mask alters the points of the face that make up the matrix on which the recognition system relies. I am doubtful that the result would allow identification of the underlying face, but no guarantees. The technique presents some serious practical issues. Full latex face mask - ever worn one? Damned things tend to be very, very hot. Better not try this outdoors in a troplical climate, and better hope the A/C is turned way up at the airport. Also, I'm not familiar with the details of the incident you cite (link?,) but a typical latex mask would be detectable by a human within a few yards under most circumstances.

MMarch 29, 2018 7:54 AM

"Filtering out infrared at lens is not a feasible defense, though it is done by a lot of high end DSLRs"

Not necessarily relevant, but almost all DSLRs have an infra-red filter at the sensor

vas pupMarch 29, 2018 8:51 AM

@VinnyG • March 29, 2018 7:30 AM
That was in Tony(Antonio)Mendes book 'My Life in CIA'(kind of - you may do search on Amazon) I read it many years ago, then it was movie related 'Argo' - I did not see the movie, but I trusted story in Tony's autobiographic book. he is the master of all disguise techniques.

vas pupMarch 29, 2018 9:00 AM

I guess the prospective is utilization of AI to better/reliable biometrics evaluation including face recognition and preventing attempts to bypass it. I mean several biometrics should be analyzed by AI simultaneously. E.g. in example related to Tony Mendes, I recall that one of US diplomat has special shoes which increase his height to match bio in a passport, but that was old era.

Kind of related new AI academic research (important):

The brain learns completely differently than we've assumed since the 20th century:

"The new learning scenario occurs in different sites of the brain and therefore calls for a reevaluation of current treatments for disordered brain functionality. Hence, the popular phrase "neurons that fire together wire together," summarizing Donald Hebb's 70-year-old hypothesis, must now be rephrased. In addition, the learning mechanism is at the basis of recent advanced machine learning and deep learning achievements.
[!] The change in the learning paradigm opens new horizons for different types of deep learning algorithms and artificial intelligence based applications imitating our brain functions, but with advanced features and at a much faster speed."

Leave a comment

Allowed HTML: <a href="URL"> • <em> <cite> <i> • <strong> <b> • <sub> <sup> • <ul> <ol> <li> • <blockquote> <pre>

Photo of Bruce Schneier by Per Ervland.

Schneier on Security is a personal website. Opinions expressed are not necessarily those of IBM Resilient.