Re:Weaponizing image scaling against production AI systems
by canjobear on Thursday August 21, 2025 @03:47PM
(#44977193)
Could this be solved by applying some small amount of noise to the image before downsampling?
[ Parent
| on HN ]
Re:Weaponizing image scaling against production AI systems
by Sebb767 on Thursday August 21, 2025 @04:02PM
(#44977367)
It could be made harder, yes. This depends a lot on how the text is hidden and what kind of noise you use, though. Also, this would quite likely also impact legit usecases - you'll obscure intended text and details, as well.
Re:Weaponizing image scaling against production AI systems
by grumbelbart2 on Friday August 22, 2025 @02:08AM
(#44981481)
It should be solved by smoothing the image to remove high frequencies that are close to the sampling rate, to avoid aliasing effects during downsampling.
The term to search for is Nyquist–Shannon sampling theorem.
This is a quite well understood part of digital signal processing.