Aggregate similarity queries in relevance feedback methods for content-based image retrieval

Content-based image retrieval techniques rely on automatic features extracted from images to process similarity queries. Usually low-level features are extracted, and when they are used to compare images stored in a database to a reference image (through single center selection queries), they often lack the ability to convey to the users what they understand as similarity. To deal with the gap between what the user expects and what the system can automatically provide, relevance feedback techniques have been employed. In this paper we present a generalization of the single center similarity queries over data in metric spaces, taking into account both range and k-nearest neighbors. Allowing a query to include multiple query centers, it straightforwardly attends the relevance feedback requirements. Thus, we analyze how well our new approach contribute to relevance feedback methods for content-based image retrieval.