Landmark Recognition Coming to Google Image Search
Google is adding the ability for users to search photos of more than 50,000 landmarks such as the Golden Gate Bridge, the Eiffel Tower and other special constructions around the world.
Though still in a proof-of-concept stage (presented at a "computer vision" conference in Miami today, June 22), the project makes use of 40 million GPS-tagged landmark photos from Google's Picasa and Panoramio and tour guide Web pages.
Jay Yagnik, head of computer vision research at Google, noted in a blog post:
Next, we found candidate images for each landmark using these sources and Google Image Search, which we then "pruned" using efficient image matching and unsupervised clustering techniques. Finally, we developed a highly efficient indexing system for fast image recognition.
Here is a visual representation of Google's clustered recognition model rendering the Acropolis in Athens, Greece. This allows for a more efficient image matching system, Yagnik claims.
To get this cluster, Yagnik said Google entered the Web address of an untagged picture of a landmark into the recognition engine. The computer then identified and named it: "Recognized Landmark: Acropolis, Athens, Greece."
Google says the recognition engine is 80 percent accurate, which is pretty inaccurate by Google's standards. I'd guess Google will get it down to 98 percent accuracy if it can. Sergey Brin and Larry Page are notorious sticklers for speed and quality. Yagnik adds:
While we've gone a long way towards unlocking the information stored in text on the Web, there's still much work to be done unlocking the information stored in pixels. This research demonstrates the feasibility of efficient computer vision techniques based on large, noisy datasets.
Perhaps this is a fine exercise in separating the photographic wheat from the chaff. I see this landmark recognition image search feature eventually aiding middle schoolers' research projects for history.
I'm not sure where else this might be useful. Perhaps, as Cnet's Tom Krazit noted, for vacationers planning a trip abroad.
Where else would one use this image search feature? Thoughts?