about
Practical Attacks on Face Recognition with Infrared [pdf] (arxiv.org)
63 points by alexanderdmitri on Jul 8, 2018 | hide | past | pdf | 11 comments on HN

In plain words: Shining carefully shaped infrared light on a face tricks face recognition into seeing someone else, while the light stays invisible to people nearby. On a large photo collection, the team found a working light pattern for over 70% of attempts.

Abstract · Invisible Mask: Practical Attacks on Face Recognition with Infrared

Accurate face recognition techniques make a series of critical applications possible: policemen could employ it to retrieve criminals' faces from surveillance video streams; cross boarder travelers could pass a face authentication inspection line without the involvement of officers. Nonetheless, when public security heavily relies on such intelligent systems, the designers should deliberately consider the emerging attacks aiming at misleading those systems employing face recognition. We propose a kind of brand new attack against face recognition systems, which is realized by illuminating the subject using infrared according to the adversarial examples worked out by our algorithm, thus face recognition systems can be bypassed or misled while simultaneously the infrared perturbations cannot be observed by raw eyes. Through launching this kind of attack, an attacker not only can dodge surveillance cameras. More importantly, he can impersonate his target victim and pass the face authentication system, if only the victim's photo is acquired by the attacker. Again, the attack is totally unobservable by nearby people, because not only the light is invisible, but also the device we made to launch the attack is small enough. According to our study on a large dataset, attackers have a very high success rate with a over 70\% success rate for finding such an adversarial example that can be implemented by infrared. To the best of our knowledge, our work is the first one to shed light on the severity of threat resulted from infrared adversarial examples against face recognition.

Zhe Zhou, Di Tang, Xiaofeng Wang, Weili Han, Xiangyu Liu, Kehuan Zhang
arXiv:1803.04683 · cs.CR · submitted Mar 13, 2018
abstract · pdf · html

add comment on HN
Also discussed: Mar 2018 (3 points, 0 comments)

This does not work for FR that does not observe IR. Which is most FR systems, because most FR systems have to use legacy video hardware several years, often a few camera generations, back. Including wholly optical CCD systems.

However, a similar fashion accessory does an even better job: some fashion hats have blinky LEDs on them; blinky LEDs cause a camera's components to adjustment for the unusually bright light, which are then gone, than back again. The mechanism in most security cameras cannot adjust for the lighting changes as fast as the LEDs are blinking, causing no good captures of the face beneath the blinkly LEDs.

Don't a lot of image sensors pick up infrared? I don't have any IR remotes to test modern stuff with, but I know my Galaxy S5 and every phone and point and shoot camera I had before that would register the light from IR remotes.

I guess that's a cheap filter away from being fixed anyway.

Sure, lots of image sensors pick up IR. But they are not in the millions of legacy security camera systems at banks, grocery stores, college, corporate and campus security systems.

The majority of the surveillance systems of the world are 10-20 years old. they work fine, for their original purpose, and the owners want to add FR, not purchase an entire new system requiring the loss of their previous investment.

Say what?

All those legacy systems (probably) shipped with a weak IR filter, but the CCD on the the devices will more than happily react to a bright IR light.

This is even how the majority of "night" cameras operate. They ditch the IR filter, then illuminate the scene with IR.

Yeah this is what I was getting at. How many old cheapo security cameras have no or bad IR filters if that was still the standard for ~2012 flagship Android devices...
> Health concerns. It is unclear if an attacker would get his eyes and face skins hurt, under the exposure of such a large amount of infrared in a long period of time. If this is true after confirmation from physicians, attackers may not dare to launch attacks with this method, worrying about their health.

As a physician, I don't understand this to be a serious concern. UV light is the opposite side of the viable spectrum from IR

This is fascinating. What a great idea.

Intense IR can damage the eye; more specifically, it can lead to cataracts.

Prior to learning to arc weld, I always thought the primary danger of looking at the arc without protection was UV exposure. However, it turns out that IR exposure is also problematic. See, e.g., https://en.wikipedia.org/wiki/Glassblower%27s_cataract

more specifically, it can lead to cataracts.

Not to downplay it, but cataracts are probably the easiest operational procedure on the human body. At least in the rich world.

Most people will have to deal with that at one point in their life.

Source: Myself; I got both of my eyes fixed last year. While I can appreciate the horror of your eyes cut up, the actual operation is really minimally invasive.

Mind you, I'm not arguing that excessive IR - or UV exposure is generally a good idea.

You can be burned by IR light.
Unlike UV which can cause skin inflammation later, you will certainly notice that your skin is being burned by IR. Have you ever tried to keep your hand close to burning coals?
This is true. Heat is given off as IR (more broad spectrum)

But also you can not notice when you are burned. Or notice it only after the damage is done. Which happens in places like laser labs (very narrow spectrum).