Indexation dans les espaces métriques Index arborescent et parallélisation

The efficient indexing and searching of complex data is an increasing need in order to face the size and diversity of current databases. We introduce a tree-based indexing technique that partitions a metric space thanks to balls and hyper-planes. The performances of this index structure are experimentally evaluated on chosen datasets presenting different intrinsic difficulties. Also, we parallelise the k nearest neighbour search algorithm in order to further improve the performances.