There are many face detection and recognition algorithms and APIs out there; however, most of them can produce same results for paintings and avatars as for human face. So, is there any specific algorithm or tool to distinguish real human face and a simulated/fabricated one?
$\begingroup$
$\endgroup$
2
-
1$\begingroup$ Yeah, sure, you could just train something to tell human faces from painted ones; there's actually quite a lot of literature of making and breaking (video download contains english translated audio track) biometric systems. Can you be more specific? $\endgroup$– Marcus MüllerCommented Jan 2, 2019 at 12:24
-
1$\begingroup$ Good question. You're basically asking for the error function of face generation. If you can reduce that enough, the generated faces will be indistinguishable from real ones. $\endgroup$– MisterGeekyCommented Jan 3, 2019 at 13:18
Add a comment
|
1 Answer
$\begingroup$
$\endgroup$
Actually the need to distinguish is part of the training phase of some models.
For example look at Generating Photorealistic Images of Fake Celebrities with Artificial Intelligence.
The GAN Models used is basically a "fight" between 2 nets. One is trying to generate images and the other is trying to classify them to be either real or generated.
Once this "fight" is over in the training, for production only the generator is used.
But nothing stops you from taking the discriminator and use it for the task it was trained for, doing what you want.