Resource description and selection for range query processing in general metric spaces
Daniel Blank, Andreas Henrich · 2013
Abstract: Similarity search in general metric spaces is akey aspect in many application fields. Metric space indexing provides aflexible indexing paradigm and is solely based on the use of adistance metric. No assumption is made about the representation of the database objects. Nowadays, ever-increasing data volumes require large-scale distributed retrieval architectures. Here, local and global indexing schemes are distinguished. In the local indexing approach, every resource administers aset of documents and indexes them locally. Resource descriptions providing the basis for resource selection can be disseminated to avoid all resources being contacted when answering aquery. Onthe other hand, global indexing schemes are based on asingle indexwhich is distributed so that every resource is responsible for acertain part of the index. For local indexing, only few exact approaches have been proposed which support general metric space indexing. In this paper, weintroduce RS4MI—an exact resource selection approach for general metric space indexing. We compare RS4MI with approaches presented in literature based on apeer-to-peer scenario when searching for similar images by image content. RS4MI can outperform two exact general metric space resource selection schemes in case of range queries. Fewer resources are contacted by RS4MI with—at the same time—more space efficient resource descriptions. 1