'Privacy Visor' Can Fool Face-Recognition Cameras

itwbennett writes: Dark shades aren't enough to go incognito in the age of face recognition camera systems. For that you need the Privacy Visor developed at Japan's National Institute of Informatics. The visor consists of a lightweight, wraparound, semitransparent plastic sheet fitted over eyewear frames. It works by reflecting overhead light into the camera lens, causing the area around the eyes to appear much brighter than normal.

Comments

Kind of self-defeating

[timrod]

These glasses seem kind of pointless in that from what the article says, they pretty much have to be that big in order to actually work - the earlier model by the same company was even bigger. With something like this, the goal should be to make them as surreptitious as possible so that the person wearing them doesn't stand out in a crowd and thus draw attention from whatever security organization is likely monitoring the cameras. $250 (at current exchange rates anyway) is also far too high of a price tag for a pair of what are basically glorified sunglasses.

Now, if they looked like normal sunglasses (or better yet could be built into prescription glasses) and were under $100, I could see myself getting a pair of these if I planned to be in an area with heavy CCTV usage.

Re:Kind of self-defeating

[Darinbob]

The basic reason they work is probably that current face recognition programs are confused by them. The light being shined back is a minor influence at best, it does not blind the cameras. A few tweaks and the system will recognize faces again.

I'll give it a week

[penguinoid]

Within the week, a new update will allow
a) recognition of people wearing the "Privacy Visor"
b) selling their name to people advertizing privacy products

Like All Defenses...

[The Raven]

... this will last only until the facial recognition algorithm is trained to ignore it. If it won't fool a human, it won't fool an algorithm for long. Better fixes are ones that exploit the weaknesses of the sensors rather than attacking the algorithm. The other example, cited right in TFA, uses a more effective long term strategy of hampering the sensors.

Upgrading the algorithm? Cheap, and only needs to be done once. Upgrading every sensor to filter IR? Not impossible, but much more expensive and thus likely to be skipped by businesses.