-1

Over the years I tried to minimize exposure people's faces in images on my internet presence, mostly to evade automated facial recognition tools. I tried to either not show them at all or blur/occlude them in some way or another.

However, what about simply dithering these images? Dithering generates images which are easily recognizable by humans, but I could think that today's machines have a hard time to recognize faces on these images. Here's an example:

enter image description here

Is dithering a viable method to evade today's facial recognition algorithms, at least those commercially available?

By a quick search, I just found this paper, which seems to actually use dithering to do the recognition.

Marcel
  • 3,494
  • 1
  • 18
  • 35
  • 1
    Image recognition (including face recognition) usually does some kind of resizing of the image. AI models are also quite good at reconstructing images from dithering or similar, similar to how a human observer does internally. So if a face is easy to recognize by a human (like in the image included in the question) then it is not really hard for modern face detection algorithms. – Steffen Ullrich Sep 21 '22 at 14:04
  • I'm not sure we can answer for all algorithms in a category ... The answer would end up being for specific algorithms. And if you found evidence to support your antithesis, what kind of answer were you wanting? The answer already appears to be "no". And why would you propose dithering in the first place? – schroeder Sep 21 '22 at 14:16
  • @schroeder My initial thoughts were that dithering removes information from the original image and adds a certain amount of randomness to it, thus I would guess that this makes harder to take further information out (such as recognized faces) – Marcel Sep 23 '22 at 05:58

0 Answers0