Web users have become used to the idea that most of what they read online—whether it’s Facebook comments or personal e-mails—is scanned by software that tries to serve up relevant ads. But soon online advertising companies may start serving up ads based on the age of people in photos that you’re viewing on a page.
That’s thanks to startup Face.com, which already offers a face-recognition service that websites or apps can use to count the number of faces in a photo, tell their gender, or match them to known individuals. Starting this week, that service will also guess the age of the faces it spots in photos and ad networks and other Web and mobile companies already have plans to use it.
“You send us a photo with a face in it, and it’ll send back an estimate of their age,” says Gil Hersch, CEO and cofounder of Face.com, which is based in Tel Aviv, Israel. To use the service, programmers have their software send photos to Face.com over the Internet and receive back the results of the analysis. Face.com returns an upper and lower range on the age of the face, a specific estimate, and a confidence score. A demonstration site shows the information that Face.com calculates from a photo.
“We heard from a bunch of clients that they’re interested in adding age detection for a variety of applications,” says Hersch. “Ad services is one.” He says some ad services companies are already using the gender-detection capabilities of Face.com’s technology to help choose which ads to display next to a photo.
The operators of video chat sites that pair up strangers have also expressed an interest in age detection, says Hersch. They already use Face.com’s service as a kind of safety feature to ensure that people are showing video of their faces. “They’re trying to match you with other chatters, and age detection could help with that,” says Hersch.
Roughly 45,000 software developers are currently registered to use Face.com’s service, which Hersch says processes “a few billion” photos every month.
Face.com’s ability to guess age comes from training software on a collection of hundreds of thousands of photos that had been labeled by people who made their own attempts to judge the age of people in them. Face.com’s software matches human guesses of the age of a face in a photo about 90 percent of the time, but the company has not compared its accuracy against the true age of people.
Certain types of photograph are challenging for Face.com’s age detection. “Women who are stars tend to look younger than they really are,” says Hersch, “but this is consistent with how people judge age, because we relied on them to average the truth when we trained the system.”
Face.com’s technology may never be as good at determining age as a person, but Hersch says that in many cases, just knowing the approximate age is useful enough. “Marketers, 90 percent of them are only into the range anyway,” says Hersch.
One possible way to improve the accuracy would be to teach the system to recognize certain lighting conditions that affect how it gauges age, says Hersch. For example, strong light from above accentuates shadows around a person’s eyes and may lead to their being guessed as older than they really are.
“I think this has tremendous potential for contextual advertising, as you can pull out the right advertisements for that category of users,” says Qiang Yang, a professor at Hong Kong University of Science and Technology. Yang previously developed a way to serve ads based on what the images on a page depict. “These features coupled with textual features on a page could paint a very good picture of the person [viewing a page],” says Yang. “If this is a teenage girl, we will be able to show different ads from those for a middle-aged man.”
Facial-recognition technology often raises privacy concerns, although it has become a standard feature of Facebook, which uses it to make it easier for users to tag their friends in photos, and in photo management software such as Google’s Picasa and Apple’s iPhoto.
Hersch says the service he offers simply allows websites and mobile apps to offer more to their users. “This provides more value out of photos,” says Hersch. “We’re unearthing data that’s really there, but just not available.”
A Roomba recorded a woman on the toilet. How did screenshots end up on Facebook?
Robot vacuum companies say your images are safe, but a sprawling global supply chain for data from our devices creates risk.
The viral AI avatar app Lensa undressed me—without my consent
My avatars were cartoonishly pornified, while my male colleagues got to be astronauts, explorers, and inventors.
Roomba testers feel misled after intimate images ended up on Facebook
An MIT Technology Review investigation recently revealed how images of a minor and a tester on the toilet ended up on social media. iRobot said it had consent to collect this kind of data from inside homes—but participants say otherwise.
How to spot AI-generated text
The internet is increasingly awash with text written by AI software. We need new tools to detect it.
Get the latest updates from
MIT Technology Review
Discover special offers, top stories, upcoming events, and more.