Content-Based Image Indexing

Abstract

We formulate the content-based image in-dexing problem as a multi-dimensional nearest-neighbor search problem, and de-velop/implement an optimistic vantage-point tree algorithm that can dynamically adapt the indexed search process to the character-istics of given queries. Based on our perfor-mance study, the system typically only needs to touch less than 20 % of the index entries for well-behaved queries, i.e., when the query images are relatively close to their nearest neighbors in the database. We also report in this paper the results of extensive perfor-mance experiments, which characterise the