Hasn't this crossed over into being a philosophical question? You want to attest to reality or provenance. But then you start making lists of "acceptable" changes: file format; compression; color-correction; size; taking "medium" into account... It then slowly slides into "average human perceptibility"; "irrelevant details"; keeping the "spirit" of the image intact; judging the intention or motive of the user or of the viewer; aligning the image with "values"; appropriate/legal use... etc. Not sure what the end-game is?
We are trying to make images in the AI era be as reliable as they were pre-AI? But they were not reliable pre-AI, it was just more difficult to intentionally doctor them.
> our tool can verify a large family of image transformations
Is this family large enough to transform real image A into arbitrary fake image B?
> ZK-JPEG can merge transparent or translucent layers into an image, useful for placing visual watermarks,
creating double exposures, or merging visual layers, potentially AI generated over portions of the image. In
our tool, the transparent layer is revealed, but anything beneath an opaque portion of the layer becomes
secret. Note that in the case of an AI generated layer, the layer itself is revealed, minimizing the dishonesty
in using generative AI (but not minimizing the dishonesty of stealing artwork to train the AI)
Seems like the answer to my question is "yes", so you have to carefully inspect the transformations to see if they look legit. (The parenthetical was a surprising inclusion in an academic context)
If you're preserving the original, then you don't need the ZKP. Part of the appeal of this approach is that you could apply redactions or make the file smaller, without invalidating the signature.
Hasn't this crossed over into being a philosophical question? You want to attest to reality or provenance. But then you start making lists of "acceptable" changes: file format; compression; color-correction; size; taking "medium" into account... It then slowly slides into "average human perceptibility"; "irrelevant details"; keeping the "spirit" of the image intact; judging the intention or motive of the user or of the viewer; aligning the image with "values"; appropriate/legal use... etc. Not sure what the end-game is?
We are trying to make images in the AI era be as reliable as they were pre-AI? But they were not reliable pre-AI, it was just more difficult to intentionally doctor them.
Tie this into the Apple/Android/Sony/Leica signed photos, and you get provenance from capture to publish
Interesting work.
> our tool can verify a large family of image transformations
Is this family large enough to transform real image A into arbitrary fake image B?
> ZK-JPEG can merge transparent or translucent layers into an image, useful for placing visual watermarks, creating double exposures, or merging visual layers, potentially AI generated over portions of the image. In our tool, the transparent layer is revealed, but anything beneath an opaque portion of the layer becomes secret. Note that in the case of an AI generated layer, the layer itself is revealed, minimizing the dishonesty in using generative AI (but not minimizing the dishonesty of stealing artwork to train the AI)
Seems like the answer to my question is "yes", so you have to carefully inspect the transformations to see if they look legit. (The parenthetical was a surprising inclusion in an academic context)
Viewers could provide a way to see the original before transforms, and perhaps apply each one at a time to see how the end result is arrived at.
If you're preserving the original, then you don't need the ZKP. Part of the appeal of this approach is that you could apply redactions or make the file smaller, without invalidating the signature.