![[personal profile]](https://www.dreamwidth.org/img/silk/identity/user.png){kind=small}
Entry tags:
reverse image search
Last night Dani and I went out for dinner, as we always do after Shabbat, and chose a restaurant in Monroeville (part of the mall complex but not in the main building). We had a nice dinner and, upon leaving, found the front door locked and other people standing around. The employee at the door told us that we were on lockdown because there was a shooter in the mall. Somebody asked if we were permitted to leave and she said "I can't stop you". After conferring briefly we decided to leave, as did some of the others there. (Our car was nearby and not in the direction of the mall.) By the time I went to sleep last night they hadn't yet found the guy, so I'd say that was a reasonable call.
But that's not the main point of this post. Several news articles (here's one) report that they identified the suspect by matching store surveillance video with pictures on social media. This future contender for the Darwin award had actually posted a picture of himself on Instagram four hours earlier, wearing the same distinctive clothing he wore in the mall later, but most such searches would presumably be harder.
Image search (by keyword) is not new, of course, and more recently Google offers reverse image search (upload an image and find ones like it). I don't know how well the latter works (haven't tested it). Searching "social media" for pictures matching surveillance footage is a large task unless Google has already indexed it for you. Either way, I wonder if they are also using geo-coding information when that's embedded in photos or posts to narrow the search. (Or law-enforcement organizations might have a big, private database that includes web scrapes and lots more; they wouldn't be the first government agency to do that.)
So this all got me wondering: are local police using Google to find suspects? What kind of success rate do they get doing that?
But that's not the main point of this post. Several news articles (here's one) report that they identified the suspect by matching store surveillance video with pictures on social media. This future contender for the Darwin award had actually posted a picture of himself on Instagram four hours earlier, wearing the same distinctive clothing he wore in the mall later, but most such searches would presumably be harder.
Image search (by keyword) is not new, of course, and more recently Google offers reverse image search (upload an image and find ones like it). I don't know how well the latter works (haven't tested it). Searching "social media" for pictures matching surveillance footage is a large task unless Google has already indexed it for you. Either way, I wonder if they are also using geo-coding information when that's embedded in photos or posts to narrow the search. (Or law-enforcement organizations might have a big, private database that includes web scrapes and lots more; they wouldn't be the first government agency to do that.)
So this all got me wondering: are local police using Google to find suspects? What kind of success rate do they get doing that?
no subject
http://triblive.com/news/westmoreland/7729452-74/police-morton-mangan#axzz3RFxv3vYi
This article definitely does raise a lot of questions about how the cops are doing that, though.
no subject
no subject
It's not so much reverse-image-search (a very interesting application! See, e.g., TinEye (https://www.tineye.com/)) as face recognition, and it's gotten scarily good. Expect to have your location tracked when you're in public, if not now, then in a few years.
Google and Facebook both give you the ability to auto-recognize faces in pictures, provided the subject has given them some data by being tagged on their respective social media. Apple has also provided this capability for 6 or 8 years (you can tag people manually in iPhoto, and after a while it'll start suggesting matches).
The problem is discrimination against a very large dataset. You might have 1000 friends, and so it's easy for FB to check that list of friends and see who is the best match. The police might have a million possible matches to sort through; or ten million. I'm not sure what the state of the art is there, but even humans have trouble with this task (depending on how good the picture is). Still, the software might automatically narrow things down to a few hundred or thousand possibilities, and then a human could parse through that lot.
If Google or Facebook are providing law enforcement with this larger database, they're being quiet about it. It'd be easy for police to do on FB without FB's cooperation--at least, if the perp hasn't turned off the capability--simply upload the picture, and then seen who it offers to tag.
That being said, many times the police only have crappy pictures of people, which are much harder for the algorithms (and humans).
no subject