Where was that photo taken? Deriving geographical information from image collections based on temporal exposure attributes

This paper demonstrates a novel strategy forinferring approximate geographical information from theexposure information and temporal patterns of outdoorimages in image collections. Image exposure is reliant onlight and most photographs are therefore taken in daylightwhich again depends on the position of the sun. Clearly, thesun results in different lighting conditions at differentgeographical location and at different times of the day, andhence the observed intensity patterns can be used to deducethe approximate location of the photographer at the timethe photographs were taken. Images taken inside or at nightare temporally connected to the daylight images and thegeographical information can therefore be transferred torelated ‘‘dark’’ photographs. The strategy is efficient as itonly considers meta information and not image contents.Large databases can therefore be indexed efficiently.Experimental results demonstrate that the current approachyields a longitudinal error of 15.7 and a latitudinal error of30.5 for authentic image collections comprising a mixtureof outdoor and indoor images. The strategy determined thecorrect hemisphere in all the tests. Although not as accurateas GPS receiver, the geographical information is sufficientlydetailed to be useful. Applications includeimproved image retrieval, image browsing and automaticimage tagging. The strategy does not require a GPSreceiver and can be applied to the existing digital imagecollections