In most practical applications of image retrieval, high-dimensional featurevectors are required, but current multi-dimensional indexing structures losetheir efficiency with growth of dimensions. Our goal is to propose a divisivehierarchical clustering-based multi-dimensional indexing structure which isefficient in high-dimensional feature spaces. A projection pursuit method hasbeen used for finding a component of the data, which data's projections onto itmaximizes the approximation of negentropy for preparing essential informationin order to partitioning of the data space. Various tests and experimentalresults on high-dimensional datasets indicate the performance of proposedmethod in comparison with others.
展开▼