Browse wiki

Jump to: navigation, search
World-scale mining of objects and events from community photo collections
Abstract In this paper, we describe an approach forIn this paper, we describe an approach for mining images of objects (such as touristic sights) from community photo collections in an unsupervised fashion. Our approach relies on retrieving geotagged photos from those web-sites using a grid of geospatial tiles. The downloaded photos are clustered into potentially interesting entities through a processing pipeline of several modalities, including visual, textual and spatial proximity. The resulting clusters are analyzed and are automatically classified into objects and events. Using mining techniques, we then find text labels for these clusters, which are used to again assign each cluster to a corresponding Wikipedia article in a fully unsupervised manner. A final verification step uses the contents (including images) from the selected Wikipedia article to verify the cluster-article assignment. We demonstrate this approach on several urban areas, densely covering an area of over 700 square kilometers and mining over 200,000 photos, making it probably the largest experiment of its kind to date.he largest experiment of its kind to date.
Added by wikilit team Added on initial load  +
Collected data time dimension Cross-sectional  +
Conclusion We have presented a fully unsupervised minWe have presented a fully unsupervised mining pipeline for community photo collections. The sole input is a grid of tiles on a world map. The output is a database of mined objects and events, many of them labeled with an automatically created and verified link to Wikipedia. The pipeline chains processing steps of several modalities in a highly effective way. The basis is a pairwise similarity calculation with local visual features and multi-view geometry for each tile. Hierarchical clustering was demonstrated to be a very effective method to extract clusters of the same entities in different contexts (indoor, outdoor, etc.). We observed that the clustering step on visual data is far more reliable than on text labels. A simple tree-based classifier on the metadata of photos was introduced to discriminate between object an event clusters. Itemset mining on the text of the clusters created with visual features was proposed to mine frequent word combinations per cluster. Those were used to search Wikipedia for potentially relevant articles. The relevance was verified by matching images from the Wikipedia articles back to the mined clusters. Both the clustering and linking to Wikipedia showed high precision. Finally, in a last step we demonstrated how the database can be used to auto-annotate unlabeled images without geotags. Besides the effective mining pipeline proposed in the paper, we also carried out one of the largest experiments with local visual features on data from community photo collections by processing over 200000 photos. The results of this large-scale experiment are very encouraging and open a wealth of novel research opportunitiesn a wealth of novel research opportunities
Data source Experiment responses  + , Websites  + , Wikipedia pages  +
Doi 10.1145/1386352.1386363 +
Google scholar url http://scholar.google.com/scholar?ie=UTF-8&q=%22World-scale%2Bmining%2Bof%2Bobjects%2Band%2Bevents%2Bfrom%2Bcommunity%2Bphoto%2Bcollections%22  +
Has author Till Quack + , Bastian Leibe + , Luc Van Gool +
Has domain Computer science +
Has topic Geographic information retrieval +
Pages 47-56  +
Peer reviewed Yes  +
Publication type Conference paper  +
Published in CIVR '08 Proceedings of the 2008 international conference on Content-based image and video retrieval +
Research design Experiment  +
Research questions In this paper, we describe an approach forIn this paper, we describe an approach for mining images of objects (such as touristic sights) from community photo collections in an unsupervised fashion. Our approach relies on retrieving geotagged photos from those web-sites using a grid of geospatial tiles.eb-sites using a grid of geospatial tiles.
Revid 11,111  +
Theories Undetermined
Theory type Design and action  +
Title World-scale mining of objects and events from community photo collections
Unit of analysis Article  +
Url http://dl.acm.org/citation.cfm?id=1386363  +
Wikipedia coverage Main topic  +
Wikipedia data extraction Live Wikipedia  +
Wikipedia language All languages  +
Wikipedia page type Article  +
Year 2008  +
Creation dateThis property is a special property in this wiki. 15 March 2012 20:36:46  +
Categories Geographic information retrieval  + , Computer science  + , Publications with missing comments  + , Publications  +
Modification dateThis property is a special property in this wiki. 30 January 2014 20:34:23  +
hide properties that link here 
  No properties link to this page.
 

 

Enter the name of the page to start browsing from.