Similarity search - The metric space approach

Similarity searching has become a fundamental computational task in a variety of application areas, including multimedia information retrieval, data mining, pattern recognition, machine learning, computer vision, biomedical databases, data compression and statistical data analysis. In such environments, an exact match has little meaning, and proximity/distance (similarity/dissimilarity) concepts are typically much more fruitful for searching. In this tutorial, we review the state of the art in developing similarity search mechanisms that accept the metric space paradigm. We explain the high extensibility of the metric space approach and demonstrate its capability with examples of distance functions. After a survey of specialized partitioning and pruning concepts, we introduce the main indexing representatives and provide performance comparison. The efforts to further speed up retrieval are demonstrated by a class of approximated techniques and the very recent proposals of scalable and distributed structures based on the P2P communication paradigm.