Is anyone aware of any reputable software, service, or techniques to pollute or obstruct facial recognition in still images (photos).
I’m being asked to submit a short bio + headshot for a University program I’m involved with. I try not to post photos of myself on the open internet, but can’t easily avoid it in this case.
I see there are services out there that claim to prevent facial recognition through subtle and mostly invisible changes which fool AI, but I don’t know if these services are legit, and I especially don’t know if they are trustworthy.
Do any of you have experience or recommendations with this sort of software, or effective DIY techniques. I’d really prefer to keep my face off the open internet if possible. All ideas are welcome.
Realistically, if you know you already have photos of you online and public (likely from something public with someone else’s photos), you could try checking what you have on PimEyes, then using any of the images that come up, changing the background, and running it through apps from that article or the dface app suggested. If any of them don’t ping back for the same images, then you’re good. If not, Just tell them you don’t consent to having an image used. It’s a university, they detest liability and I’m sure have a policy or legal standing to let you not include and image.
I don’t know of any software that can reliably do this (or could be reliably trusted to protect against future techniques).
But if this is somewhat low stakes (just wanting to keep your face off the open internet or prevent automated and untargeted scans), I’d guess there are probably some interesting creative solutions.
I wonder about the effectiveness of strategies like:
Using some combination of ‘beauty’ filters (because they often ‘smooth’ skin / remove ‘blemishes’ and sometimes subtly (or not so subtly) alter face geometry (eyes, lips, cheeks, jawline). This seems like a ‘low confidence’ but possibly effective approach.
Take a real photo of yourself, reduce it to a pretty low resolution, then ‘upscale it’ back up to a higher resolution.
Train a (local/trusted) image generation AI on your likeness, possibly introducing some polluted data into the mix (celeb lookalike, closeups of faces that look somewhat like you but aren’t you) and have it generate some photos for you to use.
Any thoughts or obvious flaws with the above possibilities?