Hacker News Board Hacker News, browsed like a classic message board
Live data from Hacker News

Russian photographer matches random people with social network photos

rbth.com

Post reply on HN
Print view View original

31–40 of 135 posts

Re: Russian photographer matches random people with social network photos

#31

Post by WireWrap »

Karlozkiller wrote: post #25
noja wrote: post #21

Earlier quoted context omitted.

They're not surprised by that. They are surprised when one of the rules that has held true for the eternity of cities: that you can be semi-anonymous in a crowd, is no longer true.

In my opinion you still can. What makes this possible? You need a database linking images to names, as well as the face-recognition data. People are freely giving away the images for the database linking it all to further personal data. If they did not do this, creating such a database would still be a massive surveillance effort, albeit probably still possible if someone really wanted to do it and had the sufficient…

Various businesses have a "take pictures of customers" policy. Medical offices are one example, due to insurance fraud they say. They typically don't ask. They simply tell the customer that they are about to take a picture and it is over before they know it.

When I spot such a camera I stand to the side so they have to ask me to move in front of the camera. Which is when I politely refuse to participate. They don't like that very much. Last time, the person made a "well you might not be able to refuse next time" remark. He may be correct, or not, we'll see.

These are cases where the images should never be uploaded to a public database or shared with any other parties. However, given the increasing use of SaaS and cloud providers, exposure to some third-parties is highly likely. Any consumer information hitting the cloud is at high risk of abuse.

In addition to those scenarios you have cameras behind ATMs, cameras embedded within grocery store self-checkout lines, etc and so forth. It has already become difficult to control your image and, in some cases, prevent that image from being combined with other personally identifiable information.

Re: Russian photographer matches random people with social network photos

#32

Post by foltz »

justsaysmthng wrote: post #10

I don't see why this can't be done in real time. Point your phone towards a person in a cafe and immediately figure out her name, age, maybe relationship status, social class, augmented reality style. 5 minutes later accidentally bump into her .. and start a conversation, armed with lots of private info about the person. Or even more basic - thieves picking their targets - is this woman's purse likely to contain a lo…

How long before anti-facial recognition makeup, hair, and accessories becomes fashionable? https://cvdazzle.com

Re: Russian photographer matches random people with social network photos

#33

Post by sp332 »

larrik wrote: post #2

The artist (or the article) suggests that the pictures he took on the metro are the 'real' person, whereas the pictures on the social site are the 'ideal' person. I'm not so sure. I think its very possible that the person in the social network photographs is actually far closer to the person they really are than the photo of them riding alone on the subway.

I thought exactly the same - and the photos chosen for the article don't really back up the point either. The pairs look like the same "person" to me, not one real and one fake.

Anyway I don't know anybody who busts out their truest expression of self to ride the subway.

Re: Russian photographer matches random people with social network photos

#34

Post by gwern »

I am not sure this works or that success rates like claimed are possible:

> In 70 percent of cases, he was quite easily able to identify the people he had photographed.

VKontakte has 280m+ accounts and is the second most popular website in Russia (https://en.wikipedia.org/wiki/VK_%28social_networking%29); the Russian population is ~144m. I can't find any decent estimate of how many Russians use VK, but it seems to be at least as dominate in Russia as Facebook in the USA, which is to say, ~71% (http://www.pewinternet.org/fact-sheets/social-networking-fac...). So from non-VK usage alone, he should be able to identify no more than 70-80 percent since you can't identify those who aren't on VK in the first place.

Facial recognition is a notoriously difficult machine learning task, and doing facial recognition on the VK population of St Petersburg (>249688 * 0.7 = >174781) will be challenging even with extremely good facial recognition. I don't know what this 'FindFace' application is, but I'll be charitable and assume it matches the 2015 state of the art, "Facenet: A unified embedding for face recognition and clustering" http://arxiv.org/abs/1503.03832 AFAIK, with a best-case binary classification accuracy of 99.63% (https://en.wikipedia.org/wiki/Accuracy_and_precision#In_bina...):

0.9963 = true positives + true negatives / true positives + true negatives + false negatives + true negatives

If each person = 1 profile photo, then to classify a photo of someone there would be 1 true positive, 174781-1 true negatives, and unknown number of false negatives.

0.9963 = (1 + (174781-1)) / 1 + (174781-1) + x + (174781-1) 0.9963 = 174781 / 349561 + x 0.9963349561(1/174781) = x 1.9925943 = x

There would be ~2 false negatives for every true positive. So there would be a There is no mention of any attempt to validate the findings, so I think for many of his results, he may have merely found a similar-looking person (possibly in some cases, look-alikes http://www.nytimes.com/2014/08/26/science/looking-at-twin-pe... https://en.wikipedia.org/wiki/Look-alike ). Look-alikes are far from identical twins, so it's not necessarily surprising that

> Often, there was a striking difference between a person’s real look and the image they projected on their social network profile: a shy and grim young man might appear to be the life and soul of the party and a lover of extreme sports on his VKontakte page.

This could be due to care in social presentation. Or it could just be that they're not the same person in the first place. (The photos are not that persuasive. I'm sure there's more than one grim young men with widow's peaks in St Petersburg, or young women with curly black hair.)

Re: Russian photographer matches random people with social network photos

#35

Post by anotheryou »

blakesterz wrote: post #4
larrik wrote: post #2

The artist (or the article) suggests that the pictures he took on the metro are the 'real' person, whereas the pictures on the social site are the 'ideal' person. I'm not so sure. I think its very possible that the person in the social network photographs is actually far closer to the person they really are than the photo of them riding alone on the subway.

>> whereas the pictures on the social site are the 'ideal' person. Maybe it's that the picture on the web was chosen . Someone put thought into that, they said "oh hey I look good here", it's kinda like an ideal, or at least an above average view of themselves.

Of course my profile pic is a chosen ideal of myself. But you know i chose it and it depicts how I want to be seen. This says much more about me, than me monday morning or after a long of work, bored, alone, with the worst lighting conditions (blueish light from my mobile screen from below and neon lights from too far above).

Re: Russian photographer matches random people with social network photos

#36

Post by joe_the_user »

blakesterz wrote: post #3

I keep trying to figure out why this feels so wrong and creepy. You're in public, you can't avoid some random person from taking your picture. You have a public profile, you post pictures of yourself there, in public. But some how this feels... wrong, why is that? Something about the surprise of it? It feels like a surprise, it feels like something that shouldn't be so easy.

It makes glad about my zero-photos-of-me-on-the-Internet policy.

Also, the neural net may return a certain percent of false positives and there may be no particular way for this guy to tell. The impact is small here and would be larger in other circumstances.

Re: Russian photographer matches random people with social network photos

#37

Post by cryptoz »

foltz wrote: post #32
justsaysmthng wrote: post #10

I don't see why this can't be done in real time. Point your phone towards a person in a cafe and immediately figure out her name, age, maybe relationship status, social class, augmented reality style. 5 minutes later accidentally bump into her .. and start a conversation, armed with lots of private info about the person. Or even more basic - thieves picking their targets - is this woman's purse likely to contain a lo…

How long before anti-facial recognition makeup, hair, and accessories becomes fashionable? https://cvdazzle.com

That doesn't seem like it would work at all. The makeup and hair styles suggested would make you stand out even more! Sure, 1990s-style facial recognition won't work as reliably, but surely today's systems would flag you right away!

Re: Russian photographer matches random people with social network photos

#38

Post by brohee »

Karlozkiller wrote: post #13
prohor wrote: post #7

So, you can no longer assume being anonymous when walking down the street. If your random photo is matched with your profile, you can be basically fully identified anywhere. With all those cameras everywhere, you could be fully tracked.

Given that there are public images with your face in them and name connected to them, that can be searched by a neural net... Yes.

No need for the images to be public. A police state could match images taken by street CCTV to e.g. the drivers license database and track the citizenry in real time.

You could easily reduce the search space to match a face taken by a CCTV camera by using cell phone location data to limit it to people in the general area...

Re: Russian photographer matches random people with social network photos

#39

Post by AlexC04 »

The first thing that struck me (After I got over the whole, creepy/horror/revulsion thing) was how much more fun people seemed in their social media lives.

Rather than seeing this as an exposition of "people on social media are so much more bland than they appear" I found myself feeling "people on the subway are so much more beautiful and interesting in their 'real life'"

I feel like that might be backwards from how it gets presented elsewhere.

But, if you look for example at the blonde woman snarling a piece of cake https://birdinflight.com/ru/vdohnovenie/fotoproect/06042016-... (second last) I thought, that was charming and funny... But you'd never know that from under they grey fluorescent lights of the subway car.

It's really interesting to be confronted with the realization that every single person filling up space on the subway around you has a rich and lustrous story that brought them to that place and time.

Still, super creepy.

Re: Russian photographer matches random people with social network photos

#40

Post by type0 »

justsaysmthng wrote: post #10

I don't see why this can't be done in real time. Point your phone towards a person in a cafe and immediately figure out her name, age, maybe relationship status, social class, augmented reality style. 5 minutes later accidentally bump into her .. and start a conversation, armed with lots of private info about the person. Or even more basic - thieves picking their targets - is this woman's purse likely to contain a lo…

> But this is unavoidable, right ? We just have to accept it as a fact of life and adapt to it..

For a woman with a lot of diamonds in her purse i would suggest wearing a burqa, it would prevent most biometrics from being spotted by the cam. On the more serious note, what methods did he really use?

Post reply on HN