Aumento da Eficiência das Estruturas de Indexação Métricas com Uso de Conceitos da Lógica Nebulosa

The retrieval of complex data e.g. image, audio, video, genomic sequences etc. in Database Management Systems (DBMSs) has been in the focus of the academic community and of DBMSs developers. The most promising technique is the content-based search, in which a set of descriptors with high discrimination power is extracted from each data element, and a feature vector is obtained and used to index the data to fasten the search operations. The feature vectors are data in high dimensional spatial domains or metric domains, and its indexing requires the use of appropriate indexing structures. The most promising structures are the ones for metric domains, e.g. the M-tree, the Slim-tree, the OMNI-Family and the DF-tree. This paper shows that it is possible to enhance the performance of these structures by using Fuzzy Logic concepts, through an algorithm that uses more than one parameter to define how the most appropriate subtree must be chosen in insertion operations. Doing so, the construction of these structures is more “intelligent”, in other words, closer to human analysis and perception. Experiments with the proposed algorithm show that its performance is noticeably better than other existing structures found in literature at structure construction and performing of similarity queries.