Supporting multi-example image queries in image databases

Query by example (or image) is commonly supported in image databases. However, it is also desirable for a content-based image retrieval system to support multi-example queries. A multi-example query refers to a query which contains one or more example images. Although multi-example query problem can be solved by dividing it to several single-example queries, how to merge these single-example queries' results to find the top n relevant images is a non-trivial problem. The Netview system, which is a geographic data retrieval system developed at the State University of New York at Buffalo, provides users with flexible choices to adjust query parameters such as importance of each example image and result combination strategies to generate satisfiable retrieval output. In this paper, we formally define the multi-example image query problem. We then present a variety of result combination strategies, including linear distance combination and non-linear result combination. We finally discuss the experimental performance of these strategies.