Retrieving landmark and non-landmark images from community photo collections

Avrithis, Y; Kalantidis, Y; Tolias, G; Spyrou, E

dc.contributor.author	Avrithis, Y	en
dc.contributor.author	Kalantidis, Y	en
dc.contributor.author	Tolias, G	en
dc.contributor.author	Spyrou, E	en
dc.date.accessioned	2014-03-01T02:46:58Z
dc.date.available	2014-03-01T02:46:58Z
dc.date.issued	2010	en
dc.identifier.uri	https://dspace.lib.ntua.gr/xmlui/handle/123456789/32958
dc.subject	image clustering	en
dc.subject	image retrieval	en
dc.subject	sub-linear indexing, geotagging	en
dc.subject.other	Data sets	en
dc.subject.other	Image clustering	en
dc.subject.other	Image datasets	en
dc.subject.other	Photo collections	en
dc.subject.other	Reference image	en
dc.subject.other	Spatial matching	en
dc.subject.other	State of the art	en
dc.subject.other	sub-linear indexing, geotagging	en
dc.subject.other	Visual content	en
dc.subject.other	Visual feature	en
dc.subject.other	Wikipedia	en
dc.subject.other	Data processing	en
dc.subject.other	Indexing (of information)	en
dc.subject.other	Vector quantization	en
dc.subject.other	Walls (structural partitions)	en
dc.subject.other	Image retrieval	en
dc.title	Retrieving landmark and non-landmark images from community photo collections	en
heal.type	conferenceItem	en
heal.identifier.primary	10.1145/1873951.1873973	en
heal.identifier.secondary	http://dx.doi.org/10.1145/1873951.1873973	en
heal.publicationDate	2010	en
heal.abstract	State of the art data mining and image retrieval in community photo collections typically focus on popular subsets, e.g. images containing landmarks or associated to Wikipedia articles. We propose an image clustering scheme that, seen as vector quantization compresses a large corpus of images by grouping visually consistent ones while providing a guaranteed distortion bound. This allows us, for instance, to represent the visual content of all thousands of images depicting the Parthenon in just a few dozens of scene maps and still be able to retrieve any single, isolated, non-landmark image like a house or graffiti on a wall. Starting from a geo-tagged dataset, we first group images geographically and then visually, where each visual cluster is assumed to depict different views of the the same scene. We align all views to one reference image and construct a 2D scene map by preserving details from all images while discarding repeating visual features. Our indexing, retrieval and spatial matching scheme then operates directly on scene maps. We evaluate the precision of the proposed method on a challenging one-million urban image dataset. © 2010 ACM.	en
heal.journalName	MM'10 - Proceedings of the ACM Multimedia 2010 International Conference	en
dc.identifier.doi	10.1145/1873951.1873973	en
dc.identifier.spage	153	en
dc.identifier.epage	162	en