With use cases in chat, advertising and photo sharing, the Face.com API now estimates the age of the faces it finds in photos. The service released mood detection last July and has seen the number of developers registered on its platform double in less than a year. The age estimation is hardly perfect, but gives a guideline that joins a number of attributes the service can return from just a snapshot.
Where the system is able, it returns a minimum, maximum and estimated age. For each attribute, it also supplies a confidence rating. The age factors are returned along with the many other attributes available, including gender, rotation of face, lip position, mood and whether the person is smiling or wearing glasses.
In the photo above, Mark Zuckerberg is pegged for a 28 year old female. When the photo was taken, he was 24, within the given range. The system only returns a minimum age for Robert Scoble and it’s much greater than his actual age.
Rob Spectre posted a collection of photos to show the results he calls “broken.”
Face.com’s Gil Hirsch admits the age detection isn’t perfect. “Jennifer Anniston would probably get the wrong age at every stage of her career,” he said.
In my tests, Face.com did particularly well identifying children. From time to time someone in their twenties would be estimated in the upper teens. The system performs best with front-facing shots, according to Hirsch.
To create the age estimation, Hirsch says the company trained its algorithm with millions of photos. Along with the new feature, Face.com also rolled out new algorithms that it claims improve its pre-existing detection methods by 30%. “We’re used to increments of 3% per year,” Hirsch said.
Potentially this technology could be used for ad targeting by gender and age. Hirsch also expects video chat services might use it for security (make sure there’s a face or to screen out young children), as well as pairing up chatters to make a smarter Chat Roulette sort of experience.
There are now 45,000 developers signed up for Face.com, which has been active in the hackathon scene. At Photo HackDay 2 about 40% of the submitted hacks used Face.com.