Microsoft tries to guess relatives with 'Twins or Not'

The service, launched Friday, lets people upload a pair of photos to the web and get back a similarity score that ranges from completely unrelated to "OMG, clones!!!" It's powered by Microsoft's Project Oxford Face API, which lets developers handle image recognition problems, like determining whether people are related, by using the company's Azure cloud platform. This is another novel use of Microsoft's image recognition technology, which the company first put on display with the "How Old do I Look" website that it launched at its Build developer conference in late April.

That website went viral after it became clear that its predictions were often inaccurate -- at times, hilariously so. Some limited tests of Twins or Not turned up interesting results, like one claim that Microsoft's Operating Systems Group Corporate Vice President Joe Belfiore and HoloLens mastermind Alex Kipman were "definitely related." Twins or Not said that Mary-Kate and Ashley Olsen were a perfect match -- hardly surprising, since the identical twins played one character on the TV show "Full House."

In a blog post explaining Twins or Not, Microsoft Senior Software Development Engineer Mat Velloso said that he coded the service in four hours based on an example included in the Face API software development kit. Technically speaking, the name of the site is a bit of a misnomer, since Twins or Not seeks similarities between the faces it's asked to compare but doesn't seem honed to specifically detect twins, except insofar as they look similar.

Twins or Not was designed as a demo for Microsoft's Build Tour stop in Prague to show how easy it is for developers to build applications using the Face APIs and publish them to the web. It's part of the company's broader push to drive adoption of its cloud services, especially the designated machine learning offerings that Microsoft has built. Of course, if it gets people talking about Microsoft, so much the better for the company.