Stories
Slash Boxes
Comments

SoylentNews is people

posted by janrinok on Wednesday January 10 2018, @04:34PM   Printer-friendly
from the do-you-see-what-I-see? dept.

Image recognition technology may be sophisticated, but it is also easily duped. Researchers have fooled algorithms into confusing two skiers for a dog, a baseball for espresso, and a turtle for a rifle. But a new method of deceiving the machines is simple and far-reaching, involving just a humble sticker.

Google researchers developed a psychedelic sticker that, when placed in an unrelated image, tricks deep learning systems into classifying the image as a toaster. According to a recently submitted research paper about the attack, this adversarial patch is "scene-independent," meaning someone could deploy it "without prior knowledge of the lighting conditions, camera angle, type of classifier being attacked, or even the other items within the scene." It's also easily accessible, given it can be shared and printed from the internet.


Original Submission

 
This discussion has been archived. No new comments can be posted.
Display Options Threshold/Breakthrough Mark All as Read Mark All as Unread
The Fine Print: The following comments are owned by whoever posted them. We are not responsible for them in any way.
  • (Score: 2) by Kromagv0 on Friday January 12 2018, @05:00PM (1 child)

    by Kromagv0 (1825) on Friday January 12 2018, @05:00PM (#621445) Homepage

    For things like that I would think that IR may not be the best choice as a lot of personal cameras (cellphones mostly) have an IR filter over the sensor. For security cameras were they likely don't have an IR filter it would work and one could probably just put out a bunch of IR to throw the exposure off. By a bunch here I am thinking the output of a few IR LEDs in an earring or as decoration on a hat so a couple of watts in a broad pattern. The problems with lasers is that they don't have a spread and instead are a point while LEDs have a spread to them which is something you want to take advantage of here. Lasers are fun for damaging sensors as while it isn't a lot of total power it is in a very tight beam so the watts per illuminated area is huge which is what causes the damage. I should try this some day and install a couple of high output IR LEDs in my hat and go enter the local wal*mart where they have the security cam and TV at the entrance to see how it works.

    --
    T-Shirts and bumper stickers [zazzle.com] to offend someone
    Starting Score:    1  point
    Karma-Bonus Modifier   +1  

    Total Score:   2  
  • (Score: 2) by etherscythe on Monday January 15 2018, @05:28PM

    by etherscythe (937) on Monday January 15 2018, @05:28PM (#622628) Journal

    Where I work we have a security demo station that I have done some testing with. It may depend on the quality of the cameras, and whether they are designed to supplement visible spectrum with IR for night vision as some of them are equipped with, but even my "up to 1W" blue beam was unable to permanently damage any of the cameras I attacked (with authorization of management). This suggests that the output, even in wide "unfocused" spread would have to be significant enough that injury to the user is a real possibility. As I said, you don't see the IR, so unlike having a bright flashlight clipped to the front of your shirt where you can tell that you need to aim it away or squint a little bit sometimes and be fine, the safety margin is too thin for my comfort.

    Try taking an average TV remote. Most of them are NIR, and can demonstrate washout pretty well (or lack thereof, as the case may be). I found them to be mildly annoying at best and not significantly impairing recognition in the video stream, but again, haven't worked up into stronger LEDs for the above reasons. Not to mention, your power source would need to be bulkier and/or more often refreshed when your power output is boosted.

    --
    "Fake News: anything reported outside of my own personally chosen echo chamber"