Geographical Coordinates Estimation by Tag/Image Infor- mation

Location coordinates detection according to the visual/text information is widely used in this age, like geo-visual, geotagging and LGTA. We propose a two-level LDA scheme applied on Geo-tagging that can demonstrate relatively high accuracy compared with the tradition Standard Language Model (LM), Search Similarity (SS) and LM/SS hybrid. Our method is evaluated on a dataset from a task of MediaEval2013. LDA shows much more improvement with the comparison with same distance threshold, especially when the distance is short. It means estimation with LDA is much more precise not only within a long range, but also in a small area. Meanwhile, we present a Geovisual ranking method that can estimate geographic information using only the visual content of those images. Geovisual ranking combines both ranking and classification. It can classify images that visual similarity to the geo-location and generated similarity with other query images. Both geo-visual ranking and geo-tagging methods demonstrate outstanding performance with high prediction accuracy result on geographic location.