> e.g. inventing a cleanly readable but hallucinated license plate number on a car, or financial figure, where the original is blurry?
You can use enhanced images in trials.
In general, Courts rely on expert witnesses so if you can get one that says X method is applicable and used in that field then it's a go. It then becomes the other side's job to find an expert witness to say X actually has these flaws and is untrustable. i.e. the Court is not bothered by confidence intervals; if your profession makes decisions by flipping coins its admissible in court. The Court only cares if somebody is willing to certify it as true.
> [1]: Super-resolution imaging (SR) is a class of techniques that enhance (increase) the resolution of an imaging system.
It looks pretty clear to me that a Super Resolution Algorithm enhances images (aka Image Enhancement).
Although I think you missed the point of my post. As long as you can get an expert witness to certify the accurate you can use a technique in court.
Somebody else [2] linked the Rittenhouse case as an example as to this being disallowed but that's not what the Rittenhouse case shows. The Prosecution wanted to use pinch-to-zoom; the defense challenged that zooming wouldn't be an accurate representative; the Judge asked the Prosecution if/how they knew that zooming would be an accurate representation; the Prosecution admitted they had no idea how pinch-to-zoom works; the Judge (very reasonable) didn't allow an unknown method to be used in court. The lack of an expert witness to explain how Zooming wouldn't've changed the image was the problem.
You linked to an actual imaging technique called super-resolution imaging which is for imaging. It's a technique applied to the capture of an image, and cannot be applied to an already captured image.
Machine learning super resolution adds detail to image that does not exist, in the same way that image inpainting (or even outpainting) do.
"It's important to note that the computed super-resolution image is not real. The added details—known as "hallucinations" in image processing jargon—are a best guess and nothing more."
If you go and research the literature on deep learning super resolution techniques, they also carefully use the term "hallucinate" as well.
You can use enhanced images in trials.
In general, Courts rely on expert witnesses so if you can get one that says X method is applicable and used in that field then it's a go. It then becomes the other side's job to find an expert witness to say X actually has these flaws and is untrustable. i.e. the Court is not bothered by confidence intervals; if your profession makes decisions by flipping coins its admissible in court. The Court only cares if somebody is willing to certify it as true.
http://www.curtisreevestrial.com/files/1187.pdf
https://www.jonathanhak.com/2018/02/17/image-clarification-n...