Face recognition systems are getting very good at spotting a face in a crowd, whether using cameras in public places or software scanning millions of images posted to social media.
Facebook's DeepFace system, for example, uses a neural-net based machine learning approach to identify any person with 97 percent accuracy.
|Faces identified in red. Green square indicates no face detected. Via CVDazzle|
In a new confluence of couture fashion and privacy activism, a group of hackers is adopting the methods of dazzle camouflage to disrupt face-detection technology.
Contrasting patterns of light and dark cross the contours of the face and overlap the features, making it hard to recognize the shape of the head, and interfering with the edges of the facial features.
Note to concept artists: these styles would fit well into a futuristic cyberpunk world.
Random patterns are painted onto the face. Hair is alternately curly and straight.
Whether these fashions are accepted or ultimately banned in public places is anyone's guess, since authorities will argue that terrorists can use them too.
Such methods may only be effective temporarily, since machine learning systems are now being applied to individual movements, such as gesture and gait recognition, with comparable levels of accuracy.
Yet such advances in technology—and countermeasures to that technology—call into question our basic human assumptions about the expectation of privacy and anonymity in public places.----
Read more: Camouflage from Face Recognition Technology
Scientific papers on deep learning systems for movement recognition
Previous post: Dazzle Camouflage
Wikipedia on Facebook's DeepFace system