Automatic discovery of popular landmarks
In one embodiment the present invention is a method for populating and updating a database of images of landmarks including geo-clustering geo-tagged images according to geographic proximity to generate one or more geo-clusters, and visual-clustering the one or more geo-clusters according to image similarity to generate one or more visual clusters. In another embodiment, the present invention is a system for identifying landmarks from digital images, including the following components: a database of geo-tagged images; a landmark database; a geo-clustering module; and a visual clustering module. In other embodiments the present invention may be a method of enhancing user queries to retrieve images of landmarks, or a method of automatically tagging a new digital image with text labels.
1. A method for populating and updating a database of images of landmarks comprising:
(a) geo-clustering geo-tagged images according to geographic proximity and irrespective of tags other than tags having geo-location codes or unique user identifiers associated with the geo-tagged images to generate one or more geo-clusters, wherein the geographic proximity is determined based upon geo-location codes of the respective geo-tagged images; and
(b) visual-clustering respective ones of the generated one or more geo-clusters according to image similarity to generate one or more visual clusters for each of the generated one or more geo-clusters, wherein the generated one or more geo-clusters are input to the visual clustering without being selected using tags other than tags having geo-location codes or unique user identifiers,
wherein said visual-clustering includes selecting visual clusters based on a region graph, wherein each node of the region graph corresponds to a respective image from the one or more geo-clusters and a link between two nodes of the region graph represents at least one matching feature between the respective images corresponding to the two nodes, and wherein respective connected subtrees of the region graph are selected as visual clusters.