US: In a paper titled, ‘IM2GPS: estimating geographic information from a single image’, researchers James Hays and Alexei Efros proposed a simple algorithm for estimating a distribution over geographic locations from a single image using a purely data-driven scene matching approach. For this task, a dataset of over 6 million GPS-tagged images from the Internet was used.
Researchers from Carnegie Mellon University, US, created software that helps identify where in the world a photo was taken. The software matches a given photo against millions of geo-tagged photos available on Flickr. By finding similarly-composed shots on Flickr—such as those containing narrow streets or tall cathedrals—the software can figure out where an image was likely to have been taken.
Images annotated with GPS coordinates are geographically unambiguous and accurate. The researchers found that images with both GPS coordinates and geographic keywords greatly increased the likelihood of ﬁnding accurately geolocated and visually useful data.
Their research also highlighted that geolocation estimates can provide the basis for numerous other image understanding tasks such as population density estimation, land cover estimation or urban/rural classification.