Thus, any photos posted online could be scoured for identification information. And with computer vision technologies becoming more mature, it means that regular video footage of people could identify them the same way in seconds or less using a wide variety of different visual traits.
The implications of this on individual privacy are immense.
I used it on reddit to convince people it was unsafe to post any images of that sort. It seemed to work for about 6 months.
There's magic in image enhancement, but I don't know that ridges and valleys of a fingerprint are there, yet. I don't even know that "this specific person is scared that they leaked their face in a way that is recognizable to them" even scales to "never upload anything" - it could be this sort of news is programming the population that computers can tease out identities with any and all leaked information, pictures, audio, etc.
Heck, a decade and a half ago there were claims that governments could narrow a search for an audio file upload based on the deviation from 60hz on the power line noise - in an audio recording.
So who knows?
Wow. Any source for this?
So, under some assumptions this typically shouldn't be able to work. :-)
edit: extensively documented at https://tvtropes.org/pmwiki/pmwiki.php/Main/EnhanceButton, for example
It's not clear to me that they identified him that way. It might be that they arrested him due to other evidence, compared the fingerprints afterward, and told him that they could prove the fingerprints matched, whereupon he pled guilty.
It's no unheard of to elicit guilty pleas using less than scientifically robust forensic methods.
The biggest one was the money laundering.
Porras had used a money laundering service controlled by Homeland Security Investigations. Vendors sent the money launderer a certain amount of Bitcoin and the money launderer mailed cash back to the vendor. At some point in the money launderer’s career, federal agents quietly took control of the money laundering operation and used the position to identify dozens of darkweb vendors.
I reckon that'd be enough to run through a database.
https://www.newscientist.com/article/mg21228442-600-miscarri...
Then again a lot of forensic techniques seem to fall into the same category:
https://www.bbc.com/future/article/20150512-can-we-trust-for...
A lot of the problems come from human bias and over-stating the accuracy of tests at trial.
Alone it's not conclusive, but it's one more piece of evidence linking all of the activity together.
The possibility of this kind of attack has always been mathematically possible and it doesn't take machine learning or computer vision to do it. It boils down to basic linear transformation.
There's a long history of attempts to identify persons by photos of fingerprints for evidence, there's just a level of uncertainty involved which make it more suited for gathering intelligence than court-submitted evidence.
I know, that's not a bright vision of the future. I wonder where is the line where technology will switch from useful to socially dangerous and how far we are from it. To tell the truth, it already kind of switched from useful to a useless waste of time in many cases.
Or the day when FAANG & other big tech will get bored with selling those stupid ads and move on to more powerful and scary things.
Having said that it could just as obviously be a deceit to distract from an informant or other sources of information.
The article you link neither talks about a fingerprint of Angela Merkel nor about a fingerprint recovered from a photograph.
(But a CCC group did indeed years later show a politicians fingerprint recovered from a photo, but again not Merkels)