X

Google image search recognizes faces

How to hack your search URL to get Google to pick out pictures with people in them.

Rafe Needleman Former Editor at Large
Rafe Needleman reviews mobile apps and products for fun, and picks startups apart when he gets bored. He has evaluated thousands of new companies, most of which have since gone out of business.
Rafe Needleman
2 min read

Google Image Search can now tell the difference between a picture that has a face in it and one that doesn't. But don't get too excited. it can't tell one face from another.

Step one: Do an image search

Technically, what Google can now do is similar to what some digital cameras do: Recognize faceness. Cameras do it to lock in focus; Google does it to separate pictures into "people" and "not people" buckets.

This feature is not in general release yet, but you can play with it. Here's how: Do an image search. For example, try a search on "CNET." You'll see a lot of pictures of products, buildings, and screenshots. Now go into the address bar and stick this on the end of the URL: &imgtype=face. Go to that URL and you'll see just CNET image search results of people.

You can also search for news results: Append &imgtype=news.

This isn't an official Google search feature yet, which is why you need to hack the URL to see it. Regarding the future for the capability, Google P.R. people only have this to say: "We're always working to improve the search user experience. In the near future, we will be launching a new search feature that will enable users to refine their searches for images pertaining to faces and news articles."

Step two: append the search URL with "&imgtype=face"
Result: Only pictures of people.

Historical note: A Web 2.0 startup, Riya, was pitching its facial recognition engine last year. It never worked well enough, and the company withdrew a lot from that product, launching instead a shopping search site, Like.com (preview) where you can select goods that look similar to the things you like.

Found on Google BlogScoped via Ars Technica.