Filtered By: Scitech

Machine learning to identify cities from postcards, travel pictures

Computers may soon be able to identify cities by just looking at random photos, if a new machine learning program by US and French succeeds.
The machine learning program looks for details and characteristics that are unique to a city, tech site CNET reported.
'Frequent and unique elements'  
Researchers at Carnegie Mellon University and INRIA/Ecole Normale Superieure in Paris fed 40,000 Google Street View images of Paris, London, New York, and Barcelona, as well as eight other cities to the system, for it to find "frequent and unique elements."
"Our data mining technique was able to go through millions of image patches automatically -- something that no human would be patient enough to do," it quoted Alexei Efros, CMU associate professor of robotics and computer science, as saying.
Efros said that in the long run, they aim to "automatically build a digital visual atlas of not only architectural but also natural geo-informative features for the entire planet."
Dubbed "What Makes Paris Look Like Paris?," the research is being presented at Siggraph 2012 in Los Angeles this week.
The study also touches on the emerging field of visual data mining, which CNET said is more complex than looking for patterns in text or numbers.
US cities lack 'distinct' style
In the case of Paris, the machine learning program found the street signs, balconies, and lampposts to be "distinct," the CNET article said.
On the other hand, it had difficulty identifying US cities.
CNET said the researchers blamed this on the lack of distinct styles in American cities, which are considered a "melting pot of styles and influences."
CNET said the system could aid film look designers.
It quoted CMU as saying art directors for the 2007 Pixar movie "Ratatouille" spent a week running around Paris taking photos to capture the look and feel of Paris in their computer model.
Their system used 150 processors to automatically produce its own set of iconic Parisian images. — TJD, GMA News