Algorithmically evolved masks that appear as faces to facial-recognition software




This is fascinating…


What freaks me out is that I think I see Stephen Harper’s face in there… shudder


I see Trey Anastasio:

(admittedly, it’s all pareidolia…)



conceptually similar to reverse ocr by tiny subversions, although that one is more random


Also similar to Deep Neural Networks are Easily Fooled:
High Confidence Predictions for Unrecognizable Images


After glancing at the image I first read the headline as “…appear as feces,” and agreed.


Now we need T-shirts and fabric with that pattern all over.
Urban camouflage.


A thousand likes!


Except there’s no way to fool gait-recognition - that’s based on too deeply ingrained biometrics. You can’t just put a pebble in your shoe and expect the mandated wearable always-on cop-cams to not know who you are, and how many parking tickets you have.

We would only think we are camouflaged, acting like we’ve got uncrackable crypto, when instead we’re just using rot13 from Urban Outfitters.


Everything has vulnerabilities.

What we need is some opensource version of a gait recog system (preferably, but a copy of a commercial, closed one will do too) and a bunch of biometrics hackers having fun with it. Then we will know more.


my jake the peg outfit has them fooled every time


I’d be a bit cautious with that outfit now, though. Might end up getting more police attention than you bargained for.


Keyser Soze could pull it off, and he doesn’t need your stinkin pebble,.


Of course that pixellated khaki worn by soldiers is not an invisibility cloak, either.

For the gait-recognition, you could ride around on a unicycle playing a flame-spouting bagpipe. They’d be confused whether it’s you or, uh, the other guy who does that.


I wish there were a way to provide feedback that matches our own recognition filters. I’d love to hear an evolved stream of sound that my brain recognizes as archetypal music.


Consider the fusion of these two approaches: (though possibly here not as a control system per se but as merely an instant feedback)
and maybe spice it up with some genetic algorithms.

Beware, most of the EEG/music generation papers use EEG as source for generation of the performance. What I mean is using EEG (or fMRI, or near-IR tomography, or any other suitable modality) for the fitness selection part of the genetic algorithm.


This topic was automatically closed after 5 days. New replies are no longer allowed.