List of clustered permutations in secondary memory for proximity searching

Abstract

Similarity search is a difficult problem and various indexing schemas have been defined to process similarity queries efficiently in many applications, including multimedia databases and other repositories handling complex objects. Metric indices support efficient similarity searches, but most of them are designed for main memory. Thus, they can handle only small datasets, suffering serious performance degradations when the objects reside on disk. Most reallife database applications require indices able to work on secondary memory. Among a plethora of indices, the List of Clustered Permutations (LCP) has shown to be competitive in main memory.We introduce a secondary-memory variant of the LCP, which maintains the low number of distance evaluations when comparing the permutations themselves, and also needs a low number of I/O operations at construction and searching.