All true, although some libraries are better at handling some of those issues than others. But you’re absolutely correct that a truly candid photo is a lot less likely to get matched than a posed frontal likeness.
A study from NIST NIST Study Evaluates Effects of Race, Age, Sex on Face Recognition Software | NIST
points out the different problems with the programs:
- For one-to-one matching, the team saw higher rates of false positives for Asian and African American faces relative to images of Caucasians.
- Among U.S.-developed algorithms, there were similar high rates of false positives in one-to-one matching for Asians, African Americans and native groups
- However, a notable exception was for some algorithms developed in Asian countries. There was no such dramatic difference in false positives in one-to-one matching between Asian and Caucasian faces for algorithms developed in Asia.
- For one-to-many matching, the team saw higher rates of false positives for African American females.
- However, not all algorithms give this high rate of false positives across demographics in one-to-many matching, and those that are the most equitable also rank among the most accurate. This last point underscores one overall message of the report: Different algorithms perform differently.
The Society For Modeling and Simulation International? I’m just guessing.
“SCS” is not familiar to me. Is it familiar to most other people? I’m always curious whether Dopers think we should have to look up unknown words and initialisms used in posts or whether the Poster needs to explain it or not use it at all as would be the case with using foreign words.
yes, the problem is the ability to aggregate data -
I forget who it was once upon a time years ago had a thread about “should I send an anonymous email to my boss about X that he’s doing in violation of Labor Standards” or some such. I pointed out that based on a quick perusal of his StraightDope posts, he’d mentioned he was in eastern PA, was a photo hobbyist, had recently had foundation problems, and been on a vacation to Japan recently. If his boss or a co-worker googled the type of offence and “report boss” and found his SD thread, there’d be no problem figuring out which employee sent the email.
For face recognition - there are several approaches.
The first is the facebook account process, where you build a library of people you know or encounter and tag them. Maybe you have a lapel cam like the police bodycams, and at the end of the day it flashes photos and you ID them.
Another is the media approach - any time media mentions someone with photo, it’s added to your personal database.
The third is the Clearview approach - someone pulls together a huge collection, and by subscription you have access to some or all of it. (Let’s say, I don’t need generally the European or Asian data if I’m in North America)
Another option is crowd sourcing, but then you run into the Wikiality issue - as Colbert pointed out, a crowd-sourced media like Wikipedia was vulnerable to error or deliberate sabotage (such as editing Wikipedia to note that the elephant population had tripled in six months).
but yes, I await the day when we’ll all wander around like Secret Service with an earpiece, plus bodycam connected to our pocket computer-phone; every person who lingers in camera range for more than a few seconds, Alexa or Siri will whisper to you who that is, why they are important to you, their occupation, criminal record, net worth, and shoe size… and what the last 5 items they googled were. Oh, and there’s a sale going on in that clothing store across the street. Also, our phone will take note of their phone number and email by asking their phone automatically. (Meanwhile, they do the same to us)
We no longer remember phone numbers, because our iPhone does; nor do we remember appointments or special events, iPhone or facebook can do those. Soon, we will no longer remember anyone except our significant other, because it’s not longer necessary, our phone will tell us who they are.
It would be extremely useful for the faceblind.
I’m not sure that the downsides don’t overwhelm that. Maybe we could just get everybody to keep wearing masks, so that when I don’t know who they are they’ll blame it on the mask.
Or everyone at the party but me could wear a nametag, along with a brief CV.
In large print, please.
I’ll wear one too. After all, maybe we’re at the same party.
i think they’d be extremely useful for some businesses. Recognize repeat customers (can you tag them to your customer database?) or problem customers. Like license plate readers or cellphone tower pings, they could become the equivalent, building traffic patterns for future analysis. Even internally, your cameras would list for you who went into the storage room what time, who came/left the office what time, etc. Lobby cams for apartment buildings, tracking undesirables or frequent visitors. Malls could log and track visitors, to help identify shoplifters, etc.
Note the rich media available to track rioters on Jan 6th - saw something on the news similar for a riot in Vancouver years ago, they could track someone by clothing from place to place - “the guy with the red hat and blue hoody smashes a store window, but here he is two blocks away a bit later with full facial shot”. Add face recognition.
Information is always useful - more useful than no information. Do you notice in cop shows, nowadays, it’s not “bug this guy’s phone” it’s “who did this guy call on his phone, when, how often?”
Also should note that facial recognition is by no means a perfect science, and the problem seems to be that it is less reliable for Asians, and extremely unreliable for Blacks.
The Special Collection Service. Please excuse me. I fear my writing in simply not very good.