2
$\begingroup$

Is it possible to detect image manipulation, for example deep fakes?

I will publish some photos on internet, and want others to be able to verify if photos are really from me, and also to detect if image has been modified or not.

$\endgroup$
2
  • 2
    $\begingroup$ Why have you posted this question shortly before posting an answer that claims "Digital signatures can be used to detect of deepfakes." and recommends an app to do that? I find the juxtaposition puzzling. $\endgroup$
    – D.W.
    Mar 2 at 6:58
  • 3
    $\begingroup$ Except digital signatures don't detect deep fakes at all, they set a potentially provable level of provenance (limited by the trust in the publishing/storage method of the signatures). The signing app doesn't know whether I have just signed a raw capture from a sensor, or a modified photo, or an AI generation. Instead, it sets a point after which further mods are reliably detectable (contingent on publish method for the signature) $\endgroup$ Mar 2 at 13:40

1 Answer 1

4
$\begingroup$

This is what the discriminator in a GAN does. The generator produces synthetic images. The discriminator tries to tell the real from the synthetic images. Eventually, perhaps, the generator gets so good that the discriminator cannot tell the difference. Perhaps an even more capable discriminator could distinguish between the real and synthetic images, however.

You really just have a binary problem: real or synthetic. The problem might be very difficult, depending on how good the deepfake generation is, but that is not inherently different from any other situation where classes are quite similar in the feature space.

$\endgroup$
1
  • $\begingroup$ How the state of the art when it comes to deepfake generation vs detection compare goes beyond what I know. $\endgroup$
    – Dave
    Mar 2 at 14:12

You must log in to answer this question.

Not the answer you're looking for? Browse other questions tagged .