Antisurveillance face camouflage

[WearySky]

@SamSam

I really don’t think that’s the issue here. Yes, these manage to prevent consumer software from seeing that there’s a face at all, but the important thing they do is prevent the software from recognizing your face.

Recognizing the existence of a face isn’t really all that useful to the NSA, and anyway is pretty trivial to do even with this make-up: you see that body over there with the swinging arms and legs? Just assume there’s a face over it.

Well, I was going off this quote in the article:

When it works, CV dazzle keeps facial-recognition algorithms from seeing a face.

(Emphasis mine)

@imaguid

two things:

1. facial detection and facial recognition aren’t quite the same thing.
   one is a precursor to the other.
2. depending on the angle, the brim of
   the baseball cap can obscure just as much of the face from the
   camera’s field of view as bangs can. just keep looking down.

See my above comment for #1. For #2, this is true, but also requires you to keep looking down at all times. Look up for one second and the computers have caught ya. Of course, that also applies if it’s a windy day and the wind blows your silly obscuring bangs look out of your face with the dazzle method… But I guess if you paste em down real good you’ll be fine.

[SamSam]

@WearySky

Well, I was going off this quote in the article:

When it works, CV dazzle keeps facial-recognition algorithms from seeing a face.

Sure, that’s what the article says, but it’s not the main point of CV dazzle. Again, if the NSA just wanted to know if faces appeared at all, without caring who they were, there are much better algorithms they could write that wouldn’t be fooled by dazzle, like finding a whole body.

[WearySky]

@SamSam

Sure, that’s what the article says, but it’s not the main point of CV dazzle.

Going by the CVDazzle homepage, it does seem to be the main point of CV dazzle:

CV Dazzle explores how fashion can be used as camouflage from face-detection technology, the first step in automated face recognition.

@SamSam

Again, if the NSA just wanted to know if faces appeared at all, without caring who they were, there are much better algorithms they could write that wouldn’t be fooled by dazzle, like finding a whole body.

That requires a whole body to be there. In a crowd of hundreds of people where all you have is faces, rudimentary facial detection algorithms are pretty much all you can go on.

[billstewart]

The way they deal with the issue is to keep track of other parts of the face. Yeah, they can’t see the parts you cover up, so they’re not much use for people wearing balaclavas and sunglasses, but if they’ve got good picture resolution to work with, they can still get some results with what they can see. (Maybe not as good results as the vendors brag about, or as the cops brag about when they’re trying to get funding to blanket your local stadium with surveillance cameras to prevent the “going dark” problem caused by people wearing baseball hats to baseball games, but still, it’s something they can work with.)

[Drew_G]

Yeah it is. This (for the moment) fools facial recognition no worse when 100 people are wearing it than with 1. It’s ‘camouflage’ based on making your face not register as a face to recognition software, and that will work whether you’re the only one on the street wearing it or everyone it.

[Nelsie]

How well do they do with persons in long skirts and robes?

[Elusis]

@murrayhenson

Being a morris dancer is going to introduce so many more problems than it would solve.

AKA “Now you have two problems…”

[goretsky]

Hello,

I wonder if pairing the facepainting with small flashing IR LED emitters worn in proximity to the face (glasses, jewelry, etc.) would further confuse camera focus systems.

[shaddack]

I am not certain. The best answer would be rigging up a test system with several face-similarity processing algorithms, and running tests and seeing the degradation of the recognition ability in dependence to the countermeasures used.

That way, we can evolve the scheme step by step, seeing which step was in the right direction, and backing off those deemed inefficient or counterproductive.

[noahdjango]

heeeyyy, nice dingo balls, man!

[shaddack]

Relevant video. Steve Rambam from HOPE X. Addresses lots of issues, including face and gait recognition.

[imaguid]

@billstewart

The way they deal with the issue is to keep track of other parts of the face.

facial recognition systems already look at multiple parts of the face, but the fewer parts there are available, the less unique the resulting information is, and the less accurate (to a significant degree) any kind of recognition will be - and that has nothing to do with how sophisticated the system is because we humans would have problems too. when was the last time you accurately identified someone after seeing only their lips and chin, for example?

[crenquis]

Here’s an interesting test suite:

[jandrese]

I’m going to believe that those are pictures of people with bee allergies who were recently stung. The alternative is too depressing to contemplate.

[daneel]

I think they’re photos from Aerosmith’s most recent gig.

[davide405]

That, or they are actually splicers.