Publication Details
Shlukování založené na Voronoiově dláždění pro klasifikaci a vyhledávání ve videu
Burgetová Ivana, Ing., Ph.D. (DIFS)
Clustering, Classification, Video Search, Local Features
Althoughthere are many clustering techniques, it is not possible to use themfor all purposes. The initiative problem was to create as manyclusters as possible (eg. thousands) for the local image featuresdescription in huge amount of video for TRECVid 2008 evaluation.These large dimensional vectors cover the space almost continuouslyand commonly used clustering methods are unable to create enoughclasses or to finish in serious time.
Therefore,we have invented a new method based on Voronoi tessellation thatneeds no more than two passes through the data. It is based ondiscovery of clusters in higher density locations. Because of largedataset, it is possible to create higher amount of candidate clustersand select appropriate number of classes (large but not huge) and therest data assign to these classes. The method has been implemented asa set of SQL functions and queries and tested on a huge problem andlarge amount of classes. Performed experiments have proven that it issignificantly faster than common techniques.
@inproceedings{BUT30195,
author="Petr {Chmelař} and Ivana {Burgetová}",
title="Shlukování založené na Voronoiově dláždění pro klasifikaci a vyhledávání ve videu",
booktitle="ZNALOSTI 2008, Proceedings of the 8th annual conference",
year="2009",
pages="71--82",
publisher="Vydavateľstvo STU",
address="Brno",
isbn="978-80-227-3015-0"
}