Seleção dinâmica de classificadores para o reconhecimento de gênero musicais

• Ano de defesa: 2016
• Autor(a) principal: Fiorin Júnior, Luciano
• Tipo de documento: Dissertação
• Tipo de acesso: Acesso aberto
• Idioma: por
• Instituição de defesa: Universidade Estadual de Maringá; Brasil; Departamento de Informática; Programa de Pós-Graduação em Ciência da Computação; UEM; Maringá, PR; Centro de Tecnologia
• Palavras-chave em Português: Reconhecimento de padrões; Seleção dinâmica de classificadores; Combinação classificadores; Reconhecimento de gêneros musicais - Padrões (Informática) - Sistema de reconhecimento; Recuperação de informação por conteúdo; Ciências Exatas e da Terra; Ciência da Computação
• Link de acesso: http://repositorio.uem.br:8080/jspui/handle/1/2555

Classification can be defined as a routine in which a class is assigned to a pattern described by a set of attribute. This routine is one of the tasks present in the pattern recognition field. The observation process of the attributes from a specific element and the representation of its values with a data structure, often a vector, denotes their vector of features, thereafter applied to a classifier. Commom knowledge in a set of classifiers, at most cases, reaches a better decision than a single one. The classifiers selection is defined as a process for handling an initial set of classifiers to obtain a better small subset, which can achieve better results, closer to the goal. In music genre recognition, the main goal by using classifiers selection is to improve the final recognition rates. This work discusses a bunch of strategies for dynamic classifiers selection in music genre recognition tasks to obtain classifiers subsets who are able to correctly classify unknown patterns using combining strategies for individual estimatives. KNORA method variations, are also investigated were generated dynamically to support a classification cycle. In these cases, the neighborhood members, based on similar samples from a well known set. Finally, investigate the complementary methods with the combination of individual outputs and verify the improvement in recognition rates.