Image Clustering and Retrieval Combining fixed/adaptive-binned Histograms and various Distance Functions
In the context of content-based image retrieval, we comparetwo types of histograms, fixed and adaptive, both frequentlyused for modeling the image features. We demonstrate thata choice of a histogram type, combined with the choice of adistance function, can have a huge impact onto theclustering structure of the dataset. Such a hierarchicalclustering structure visualization of database objects helpsoften the user to find similar objects and discover unknownpatterns. In our experiments we use real data sets with largenumber of semantic categories, and evaluate both thereachability plots and the clustering accuracy, to show theeffects of appropriate choice of fixed and/or adaptivebinning in combination with various distance functions.Results show that significant clusters, along with theirrepresentatives, can be automatically extracted, which is abasis for visual data mining but even more important fornon-visual data mining.
|Authors:||Jovic M., Seidl T., Stejic Z., Assent I.|
|Published in:||2004 IEEE Conference on Cybernetics and Intelligent Systems (CIS 2004)|
|Type:||Conference papers (peer reviewed)|
|Research topic:||Exploration of Multimedia Databases|