Software spots a city's visual DNA

Every city has features that make it unique. You might be able to tell the difference between Paris and Chicago, but would you be able to tell the difference between Paris and, say, Milan? Can you tell which one is which by looking at the buildings below?

Photo: Flickr/FredArt

Photo: Flickr/a-m-a-n-d-a

At quick glance it's not easy to tell one from the other. (In case you're keeping track, the top picture is Paris and the bottom picture is Milan.)

But new software developed by researchers at Carnegie Mellon University and INRIA in Paris can recognize the small details in architecture that gives a city its unique look.

Using the substantial amount of geotagged images that are available through Google Street View the researchers built software that can automatically find the visual elements of cities -- for example windows, balconies, and street signs -- that give them their unique character.

The algorithm the researchers created -- which could be useful to architects, urban historians, or even film makers -- looks for patterns that are both recurring in the city and that do not appear in other places. The goal is to point out what makes Paris look and feel like Paris, because it's more than just the Eiffel Tower.

The software might work well in European cities by pointing to building characteristics, but as the researchers note in their report, it might also say something about the urban design of a city. "On the whole, the algorithm had more trouble with American cities: it was able to discover only a few geo-informative elements, and some of them turned out to be different brands of cars, road tunnels, etc. This might be explained by the relative lack of stylistic coherence and uniqueness in American cities (with its melting pot of styles and influences), as well as the supreme reign of the automobile on American streets."

But it's not just cities that this algorithm could be useful for, the researchers point out: "The proposed algorithm is not limited to geographic data, and might potentially be useful for discovering stylistic elements in other weakly supervised settings, e.g. 'What makes an Apple product?'"

(h/t New Scientist)

This post was originally published on Smartplanet.com