Hierarchical clustering and outlier detection for effective image data organization

This paper proposes an approach of hierarchical image data organization for effective image retrieval. Our approach is basically based on the Cross-Association (CA) that was originally devised for uncovering hidden communities in data without requiring any parameters. We first modify the CA to be appropriate for the clustering context, and propose a hierarchical clustering algorithm based on the modified version of CA. Then, we propose a novel algorithm for outlier detection that is well matched to the CA framework. We perform extensive experiments to show the effectiveness of our clustering algorithm and also our outlier detection algorithm. We also demonstrate the results obtained by applying our algorithms to real-world data.