In this paper, we present a new method for indexing a large amounts of feature vectors in high dimensional space. We introduce a partitioning method based on lattice vector quantization that divides the feature vectors progressively into smaller partitions using a finer scaling factor. The resulting hierarchical structure is then represented as a tree-structured lattices and the efficiency of the similarity queries is significantly improved by utilizing firstly the hierarchy and secondly the good algebraic and geometry properties of the lattice. Moreover, the dimensionality reduction that we perform on the feature vectors translating from one upper level to a lower level of the tree reduces the complexity of measuring similarity between feature vectors and enhances the performance on nearest neighbor queries especially for high dimensions. We include the performance test results that verify the advantage of the proposed indexing structure and show that the tree-structured lattices outperforms one of the best standard indexing structure: the SR-tree.
展开▼
