Well, digital zoom is doesn't add anything. Still a questionable defense but if you search on Google for "pinch to zoom interpolation" you will find countless news articles to exactly this topic.
I ask myself why do I even find these articles? Has nobody ever searched for interpolation algorithm for Apple? Results like this make me feel I live in FakeWorld®. Arstechnica has become abysmal, but a lot of other tech sites too.
Not that this is relevant to the argument of the defense at all. It is complete bollocks, but I still would like to know what interpolation method the zoom of the standard image app on iOS uses (standard, not multiple lens magic). Is this some corporate secret from Apple and nobody knows? At least the articles could have looked it up.
Aside from that there is merit to not resample an image and look at the "raw" data. Image forensics is regulated I believe and I don't know if they would accept Apple zoom either to be honest. That would be interesting to ask a professional, perhaps a real journalist could help here?
I think the case if high profile enough that you can send someone to buy the largest TV one could find.
In general you are right, it is not relevant, except in this case the area in question was about 30 pixels square in the original badly lit night footage.
With an area that small in those conditions, you'd rather not use any algorithmic interpolation or machine learning in a judicial setting.
Quoting another commenter here : "Expert witness said he had no way to know and never compared his work to original, [...] if he ever held his enhanced image next to original to see if they matched. Expert said no I never did that."
It was not the zoom feature it was the image enhancement feature that the expert failed to guarantee didnt change key pixels to one color or another. Expert witness said he had no way to know and never compared his work to original. Defense asked expert if he ever held his enhanced image next to original to see if they matched. Expert said no I never did that. So the unverified forensic photoshoped image was not admissible.
> The discussion offers a glimpse into how criminal trials are affected by a judge's unfamiliarity with technology—even when that technology is a common consumer feature that's grasped intuitively by millions of people of all ages.
Is Ars Technica, supposedly a tech-focused news site, really so clueless as to confuse using a technology, with understanding it?
While the judge might be overly cautious in this instance, it's definitely plausible that Apple, in their closed-source software, is using machine learning-enhanced zoom.
There's a fascinating part of a Vinge novel where the fact that the AI's high end features aren't available causes a most-likely interpretation of a collection of sensor data to lead to the wrong conclusion.
5 comments
[ 3.4 ms ] story [ 21.0 ms ] threadI ask myself why do I even find these articles? Has nobody ever searched for interpolation algorithm for Apple? Results like this make me feel I live in FakeWorld®. Arstechnica has become abysmal, but a lot of other tech sites too.
Not that this is relevant to the argument of the defense at all. It is complete bollocks, but I still would like to know what interpolation method the zoom of the standard image app on iOS uses (standard, not multiple lens magic). Is this some corporate secret from Apple and nobody knows? At least the articles could have looked it up.
Aside from that there is merit to not resample an image and look at the "raw" data. Image forensics is regulated I believe and I don't know if they would accept Apple zoom either to be honest. That would be interesting to ask a professional, perhaps a real journalist could help here?
I think the case if high profile enough that you can send someone to buy the largest TV one could find.
Here are the picture sin question : https://preview.redd.it/ik22sbz2e6z71.png?width=2170&format=...
With an area that small in those conditions, you'd rather not use any algorithmic interpolation or machine learning in a judicial setting.
Quoting another commenter here : "Expert witness said he had no way to know and never compared his work to original, [...] if he ever held his enhanced image next to original to see if they matched. Expert said no I never did that."
Is Ars Technica, supposedly a tech-focused news site, really so clueless as to confuse using a technology, with understanding it?
While the judge might be overly cautious in this instance, it's definitely plausible that Apple, in their closed-source software, is using machine learning-enhanced zoom.