Photographs by Alexa Vachon for POLITICO
Meet Janosch.
Around the world, governments and corporations are rolling out technology that scans faces for unique features and allows them to identify individuals based on that. As facial recognition technology spreads, it seems theres no hiding from it. Or is there?
We teamed up with a photographer and a makeup artist to try to camouflage POLITICOs AI correspondent, Janosch Delcker, from the all-seeing eyes of a facial recognition system, with disguises ranging from an old-school balaclava to makeup designed to distort his key facial features. Then we used a free version of Amazon Web Services facial recognition software, Rekognition, to compare the photos to a reference.
The results provide striking insights into what todays facial recognition technology is capable of: Different beards, or the lack thereof, did nothing to conceal Janoschs identity; neither did dark sunglasses or a bandana that covered the complete lower half of his face. Camouflage makeup, which distorted the proportions and symmetry of his face, made it slightly more difficult for the system to identify him, but it still detected enough common features to conclude that both photos show the same person.
The only disguises that worked were masks that fully or almost fully covered Janoschs face — or, surprisingly, Groucho glasses. Dont blame us if those become the new face of the revolution.
99.8 percent match
99.1 percent match
No match
98.6 percent match
96.3 percent match
No match
True to form, an Anonymous mask does make the wearer, well, anonymous. Distort all facial features save for the eyes, and theres no match through Rekognition.
No match
99.7 percent match
99.8 percent match
No match
99.6 percent match
99.8 percent match
The sacrifices of an AI correspondent: From beard to mustache to clean-shaven, nothing below a 99 percent match.
93.1 percent match
Makeup applied to distort Janoschs most recognizable facRead More – Source
The post Want to dodge AI’s prying eyes? appeared first on News Wire Now.
