Organização de Dados Multimídia para Busca Conceitual Baseada em Ontologia

Statistical Learning Approaches for Discriminant Features Selection

Luciano B. de PaulaRodolfo S. VillaçaMaurício F. Magalhães

Gilson A. GiraldiPaulo S. RodriguesEdson C. KitaniJoão R. satoCarlos E. Thomaz

The current available large volume of multimedia data, semantically annotated, from several different sources, generates new issues about storing and retrieval of those data. In this scenario, it is common to use simple ontologies to classify data, creating a relationship between them, that relationship may be measured by a similarity index. The contribution of this paper is to propose a way to organize multimedia data by its conceptual classification, using LSH (Locality Sensitive Hashing) functions, facilitating the conceptual search in P2P networks. Supervised statistical learning covers important modelslike Support Vector Machines (SVM) and Linear DiscriminantAnalysis (LDA). In this paper we describe theidea of using the discriminant weights given by SVM andLDA separating hyperplanes to select the most discriminantfeatures to separate sample groups. Our method,called here as Discriminant Feature Analysis (DFA), isnot restricted to any particular probability density functionand the number of meaningful discriminant featuresis not limited to the number of groups. To evaluate thediscriminant features selected, two case studies have beeninvestigated using face images and breast lesion data sets.In both case studies, our experimental results show thatthe DFA approach provides an intuitive interpretation ofthe differences between the groups, highlighting and reconstructingthe most important statistical changes betweenthe sample groups analyzed.

Caso o link acima esteja inválido, faça uma busca pelo texto completo na Web: Buscar na Web

Biblioteca Digital Brasileira de Computação - Contato:
     Mantida por: