Visual Wikipedia: Landmark Detection for Cities

When people visit a certain city there are some particular places that strike to their tourist appeal and these places summarize the city, in some sense, for them. We want to automatically summarize the tourist appeal for cities by singling out these visually striking landmarks along with a brief verbal description for them. Earlier applications have looked at finding visual distinction between cities, but we are aiming at identifying distinctions within the local domain. We divide the city in blocks on the basis of its latitude and longitude, and try to find out those blocks which have more unusual or interesting visual information as compared to the rest. By intuition, all the interesting places must be exclusive to their own blocks hence must get highlighted upon comparison with the rest of the dataset. Using our approach we were able to find a few landmarks for a portion of Paris and Pittsburgh which are shown in the form of heat maps. We also try to find out the similarities or patterns between these two cities and see if these unusual places follow a certain theme.