New Google Image Search
“The next big thing for image search would be the ability to search based on visual concepts, such as a picture of a house on a mountain with a river in front of it.” And now, Google Images allows you to restrict your search to a specific category – albeit in an “unofficial” mode only – and one of these categories may well be powered by actual image recognition (as opposed to textual keyword analysis). Right now, the available modes are (at least) the following:
show everything (the default old search)
show faces
show news images
But there doesn’t seem to be anything in the interface to trigger this – you have to resort to appending a parameter named “imgtype” to the result URL, with the values “face” or “news”. A normal result URL when searching for the W3C (the World Wide Web Consortium) for instance looks like the following:
>> http://images.google.com/images?hl=e...c&imgtype=face
Google watcher Ionut Alex. Chitu comments that this type of search result “may be the first visible result of the Neven Vision acquisition” from August 2006; Neven Vision’s speciality was image object recognition. The “news” parameter on the other hand triggers the following search result, and it may well be that it’s based on the much simpler algorithm of Google just taking into account images from sources they list as news:
>> http://images.google.com/images?hl=e...c&imgtype=news
It’s unclear what exactly Google defines as news. Whereas for instance the second result, Sci-tech-today.com, is also one of the ~10,000 Google News USA sources – we can verify this using the “site:” operator – the first one, Counter-smap.com, isn’t (it may be a source in a non-US Google News, or it may be that Google uses certain keywords found on a page to determine its “news” status).
It would be incredibly neat to see this being rolled out for many other categories, from “animal” to “Creative Commons-licensed” and what-not. And it’s actually possible that more undiscovered image types are already supported. Which image categories would you like to see?
Try it yourself...