Metric inverted - an efficient inverted indexing method for metric spaces

The increasing amount of digital audio-visual content accessible today calls for scalable solutions for content based search. The state of the art solutions reveal linear scalability in respect to the collection size due to the large number of distance computations needed for comparing low level audio-visual features. As a result, search in large audio-visual collections is limited to associated metadata only done by text retrieval methods that are proven to be scalable. Search in audio-visual content can be generalized to search in metric spaces by assuming some distance function on low-level features. In this paper we propose a framework for efficient indexing and retrieval of metric spaces by extending classical techniques taken from the textual IR methods such as lexicon, posting lists and boolean constraintsthus enable scalable search in any metric space and in particular in audiovisual content. We show the efficiency and effectiveness of our method by experiments on a large image collection.