This is a very cool use of facial recognition. It could also be expanded to the complete history of photographs. So lets say trying to identify all people on all historical photographs. Wouldn't it be funny to see some of the same people pop up in different pictures around the world in events that were previously thought unrelated?
For privacy issues you'd want to limit this to periods in time that are far back enough for most people to have passed though. So I could imagine the Second World War to be a (somewhat arbitrary) limit for now.
This limit would eventually need to be expanded, because I wouldn't want my photographs to be examined like this in 70 years time.
Given the likelihood of coincidences like that actually happening, you'd surely end up finding lots and lots of pairs of completely unrelated people that look remarkably similar, with no way of knowing if any of them are actually the same person.
I've always found it slightly amusing that we've now reached the time in the history of civilization where immortality, even were it possible, has become a massive liability. One can no longer simply assume a new identity of proper age, and attempting to do so is liable to get you jailed, or worse still, in some shady government lab.
It's even getting to be the case now that you can't just move to a new country and start over. Unless you've got skilled forgers close at hand, you're going to have a bad time immediately, and with facial recognition and biometrics, you'll have a worse time still.
It's perhaps the definition of a low-probability corner case, but if any alchemists happened to succeed, they're no doubt having a pretty depressing century.
I think if you've lived a few centuries, you might've picked up the skills, experience and connections required to make it work. There are already plenty of regular people living with false identities, so you wouldn't stick out like a sore thumb. Plastic surgery solves the rest.
> I think if you've lived a few centuries, you might've picked up the skills, experience and connections required to make it work.
This is how the immortals from Ilaria Corporation handled it in Helix. The show was promising, it's really sad it quickly went from interesting to crazy to utter gross.
Could you not use metadata to create a confidence score? For example, if two people look similar but the dates of the photos are 100+ years apart, it’s unlikely they’re the same person.
I think this is key. Figuring out a name from a photo with no additional data would give a high probability of incorrect identification. But any metadata attached to the photos could be useful.
Date the photo was taken.
Where the photo was taken (e.g. the annual family reunion in SmallTown, Illinois)
Photographer. They're likely taking photos of the same general set of people or in the same geographic area.
Time+location of other similar photos. Back in the days of the civil war, it wouldn't be possible for someone to be in to locations 1,000 miles apart in 2 days.
Side note: I recently saw a video from ~1910 in New York City. Everybody in the video had a similar physical build and would make the "average" person today look fat. I wonder how this would affect the confidence of photo grouping? Overall, did people look more homogeneous back then?
Cynically, I'm going to say you either find unsurprising matches (eg. photos of the same group of people taken on different days) or even more convincing and hard-to-detect false positives.
If your goal is explicitly to find something extremely rare - unexpected face matches in different places - you will be flooded with false positives even if the false positive rate is fairly low.
You have a lot of context/metadata like date taken (or at the worst, date acquired for a historical collection, giving a terminus ante quem) and can easily figure out the false positive rate. You may not know if two matches are the same person, but if one photo is dated to 1850-1860 and another photo is dated to 1931, you can safely infer that you have a false positive. Over a dataset of matches, that gives you an idea of how reliable each one.
Give people access to all the other clues like "insignia and inscription" and probably a lot of reliable identifications can be made, and the remainder flagged as 'possible but uncertain'.
> This limit would eventually need to be expanded, because I wouldn't want my photographs to be examined like this in 70 years time.
Wait, why are you fine with doing it with other people’s 70 year old pictures, but you don’t want people 70 years from now to do it with your pictures?
No sarcasm and I know it sounds dubious, but these are my thoughts:
- live expectancy has gone up, so in 70 years it will be more likely that I am still alive
- there are and will be a lot more photographs of me than of the average person in the 1940s, probably in the thousands of times more, so the impact on my privacy of studying all my photographs is far greater than it would be for the off picture of someone in the 40s. "Here he is, dressed up, at his wedding and 10 years earlier as a child" vs "Here is a map with all the places where mosselman was seen with a social graph of all the people he was ever seen with"
I don't want Mickey Mouse to stay copyrighted forever, and I don't want your - and by extension, my - right to privacy to extend too much further past my death.
What 'too much' would mean is the question though. Even my original specification of 70 years might, in some cases, be too short a time. Take the holocaust for example: The nazis used privacy sensitive information like notes from jewish church meetings to find people to deport to concentration camps.
So let's say that Apple, on who's servers I store my photographs, may release my images to the public once I die (assuming they can). What if there is some other totalitarian regime by then that will use those images to create a profile of me. They can then use that profile to by extension profile my children or other relatives and use that information to harm them.
So I don't really agree with the no-privacy beyond death story. Since maybe you have to say that the privacy rights of my data will be passed on to my children or others who are still alive that I've had relationships with of any kind.
The images in this article are probably far enough back not to be of much importance though, but maybe my idea to make a bigger study ranging to WWII was maybe a bad one with regards to privacy issues of the people in the pictures and the privacy of their relatives.
I will disagree. Being alive or dead should not impact what others can do as far as your privacy goes.
I would also extend this to any technology available in the future. Facial recognition is interesting, but let’s think about more advanced biological things that could be available in 50-100 years.
It could be argued rights are something a person has. Once you die, you are no longer a person. Dead people don't have a right to vote.
I think a much stronger argument is to be made that your surviving family/friends have a right to privacy that somehow includes aspects of deceased relatives lives staying secret. But that makes it a right for the living, not the dead.
I’m not suggesting the dead should vote. I’m saying the living consider the consequences of their vote even if they don’t think they will live to see it.
> This is a very cool use of facial recognition. It could also be expanded to the complete history of photographs. So lets say trying to identify all people on all historical photographs. Wouldn't it be funny to see some of the same people pop up in different pictures around the world in events that were previously thought unrelated?
My issue with that is faces aren't that unique, so you might get unrelated people identified as the same individual.
The optics are not typically the biggest issue. It's much more that film was very slow and grainy. We've been able to make good lenses for at least 100 years and probably longer.
Solving some of the quality issues would be interesting as well as studying the similarities of faces over time across the world. All in all it would be an interesting field of study for the hobbyist.
I don't think your paraphrase version is any easier to parse than the original. However I have taken your advice onboard and tweaked my original comment.
So you'd rely on social sanction? With the hope that people with the resources to do this would have enough of a reputation to lose? Otherwise I see no way to stop it.
I have seen 100 years commonly used as a limit for census data and the like. It seems like a good number as any individuals are very unlikely to still be alive. (Those few that are would have been very young).
Do you think there's no risk of harm (or at least inconvenience) for people who are suddenly discovered to be descendents from baddies of History?
I'm sure the Hitlers and Görings of this world changed their names in due time and took steps to insulate themselves and their offspring from their forefathers' bad rep, but what if it turns out your pop-pop was a guard at Auschwitz?
I always though that this was gonna happen in the future. That someone 100 years from now will check zettabytes of videos, pictures, posts and is going to create complete profiles of everyone that has lived in our time.
Each time that you are in the background of a picture of an stranger. In the broad panoramic videos from sport stadiums. From fingerprinting you writing in all social media. From government documentation. From text of other people talking about you... all your digital traces to know who were you.
Now is happening with past pictures, even that there is no so much to mine compared with the present.
Well here's the actual link to the website https://www.civilwarphotosleuth.com/
What I'm curious about is what image recognition system they used to map the faces in the picture?
I thought it would be dlib since that's open-source and very easy to use, but it's actually the Microsoft face recognition API, according to their paper: "Photo Sleuth: Combining Collective Intelligence and
Computer Vision to Identify Historical Portrait", Mohanty et al 2018 http://crowd.cs.vt.edu/wordpress/wp-content/uploads/2018/04/...
The pictures that jumped to mind for this as a potential social issue aren't the Civil War ones - they're the Jim Crow era lynch mob pictures.
I'd say there's a huge difference between knowing that your (or someone else's) ancestors were terribly racist and knowing that they were involved in murders.
Based on numbers from http://law2.umkc.edu/Faculty/projects/ftrials/shipp/lynching... lynchings dropped off pretty rapidly after 1922, and were in the single digits annually by the mid 1930s. I suspect that there's also a drop in pictures of crowds and a social change that made them considered if not wrong at least "not something that people like US do."
There are very few people still alive who were adults when there were large numbers of lynchings happening in the US.
For privacy issues you'd want to limit this to periods in time that are far back enough for most people to have passed though. So I could imagine the Second World War to be a (somewhat arbitrary) limit for now.
This limit would eventually need to be expanded, because I wouldn't want my photographs to be examined like this in 70 years time.