pull down to refresh

Camera with content credentialing support stops deepfakes at the shutterspectrum.ieee.org/leica-camera-content-credentials
152 sats \ 3 comments \ @0fje0 22 Nov 2023 tech
Cameras with content credentialing support, essentially encrypted EXIF?
Canon and Nikon already have prototype cameras with content credentialing support. Smartphones will also get in on the action. Truepic, a startup that builds “authenticity infrastructure,” has partnered with Qualcomm to make Qualcomm’s Snapdragon 8 Gen 3 chips support content credentials; those chips will power flagship Android smartphones next year.
write
preview
related posts
21 sats \ 2 replies \ @Bitman 22 Nov 2023
That's pretty good - so, as you say, an encrypted EXIF that gets more data added whenever another edit occurs.
This pioneering camera will only be in the hands of the rich and professionals.
Most professionals shoot and work with RAW images1 already - they're 'digital negatives' and so can't be changed - wouldn't these RAW images be enough to prove that someone has maliciously changed an image?
Given the high price tag of this, the only camera that incorporates this technology, it would need to be adopted by cheaper cameras to make a difference.
Unless there's some serious privacy problems, I'm all for it.

Footnotes

  1. RAW images are great - they're akin to giving the photographer the ability to change the camera settings within the image AFTER the image has been taken. When happy with the result, the image would be saved as a regular image. ↩
reply
21 sats \ 1 reply \ @0fje0 OP 22 Nov 2023
Great reply, thanks.
Unless there's some serious privacy problems
This was my first thought, as well.
Until today, I wasn't aware of the Content Authority Initiative (CAI) nor their C2PA standard, so I'll be keeping an eye on developments.
it would need to be adopted by cheaper cameras
Seems like it's already on its way. Possibly paving the way for more people giving up privacy for convenience?
reply
0 sats \ 0 replies \ @Bitman 22 Nov 2023
It sure is an interesting area of research.
You might be interested in hearing about a parallel move - for creators to (invisibly) 'watermark' images so that people identify if their images have been adapted used by others - I'm not 100% sure but it might even be apparent if AI has utilized it too. I do remember that it's FOSS and up to the creator to do to protect their copyright.
There's also a seperate move to 'taint' AI's well of images that they might use.
When these purposely 'tainted' images are used by AI some objects are transformed into things like dogs - against both the AI algorithm's and end-user's wish. The idea, I believe, is to really mess with the AI's output. A 21st century sabot (clog) in AI's wheels as it were...
I'll try and find a source on it and write a post - but as you can imagine, unless I've bookmarked it, there'll be many more people separately writing about putting dogs in images on purpose too.
reply