Estimating proximity of metric ball regions for multimedia data indexing

The problem of defining and computing proximity of regions constraining objects from generic metric spaces is investigated. Approximate, computationally fast, approach is developed for pairs of metric ball regions, which covers the needs of current systems for processing data through distances. The validity and precision of proposed solution is verified by extensive simulation on three substantially different data files. The precision of obtained results is very satisfactory. Besides other possibilities, the proximity measure can be applied to improve the performance of metric trees, developed for multimedia similarity search indexing. Specific system areas concern splitting and merging of regions, pruning regions during similarity retrieval, ranking regions for best case matching, and declustering regions to achieve parallelism.