@phdthesis{oai:tsukuba.repo.nii.ac.jp:00008676,
 author = {An, Jiyuan and 安, 際元},
 month = {},
 note = {High-dimensional data, such as documents, digital images, and audio clips, can be considered as spatial objects. The distances in a feature space between two objects measure their dissimilarities, and, the spatial indexing/access method R-tree [23] and its family on the space can be applied to the problem of the approximate retrieval. However, how to define the distance function is an important problem in high dimensional datasets. Though Euclidean distance (L2) is commonly used, in some cases the metric other than L2 is more appropriate for describing the feature of the data. However, except for L2 distance function, spatial index method R-tree and its family are not applicable, because they are based on Euclidean space. In this dissertaion, we propose a way to map L1 metric to a Euclidean space, then R-tree is applied to the Euclidean space.  ･･･, 2002, Includes bibliographical references},
 school = {筑波大学, University of Tsukuba},
 title = {Index method based on dimensional reduction},
 year = {2003}
}