Carnegie Mellon researchers have devised a photo analysis engine, IMG2GPS, that can make a guess as to the location of a photo based on similar photos found on Internet photo sharing web sites such as Flickr. The image analysis technique doesn’t dwell on identifying specific objects in the photo, but looks for similarities in lines and colors. Since many images on Flickr now have GPS tagged geocodes, the analysis engine can suggest that if your photo looks like these, and these are all in Utah, then perhaps your photo is of something in Utah as well.
PhysOrg.Com notes that even when the image analysis program can’t find a good location match for a photo, it often finds similar-looking locales. For example, it might not be able to name the beach in a given photo, but it is likely to note that it is similar to several other photos of beaches. Or given a photo of a narrow street in Barcelona, it suggests photos of streets in other Mediterranean towns but not American alleyways.
This is a pretty cool achievement, a step in the direction of imbuing computers with the appearance of some degree of common knowledge. My high school chemistry teacher coined the term “a brillant grasp of the obvious.”
human: “Where was this photo taken?”
computer: “Looks like a beach.”