Representação do espaço de características por meio de conjuntos difusos

In the recent years we have witnessed great interest in content-based image retrieval with emphasis in the development of visual feature extractors and similarity measures. In this paper we propose a novel approach to represent the visual feature space, taking into account the uncertainty presents i...

Descripción completa

Detalles Bibliográficos
Autor: Genari, Alan Carlos
Tipo de recurso: tesis de maestría
Estado:Versión publicada
Fecha de publicación:2010
País:Brasil
Institución:Universidade Federal de Uberlândia (UFU)
Repositorio:Repositório Institucional da UFU
Idioma:portugués
OAI Identifier:oai:repositorio.ufu.br:123456789/21234
Acceso en línea:https://repositorio.ufu.br/handle/123456789/21234
Access Level:acceso abierto
Palabra clave:Computação
Conjuntos difusos
Processamento de imagens
Sistemas de recuperação da informação
Representação por conjuntos difusos
Partições difusas
Recuperação de imagens por conteúdo
Fuzzy set representation
Fuzzy partitions
Content based image retrieval
Fuzzy sets
CNPQ::CIENCIAS EXATAS E DA TERRA::CIENCIA DA COMPUTACAO
Descripción
Sumario:In the recent years we have witnessed great interest in content-based image retrieval with emphasis in the development of visual feature extractors and similarity measures. In this paper we propose a novel approach to represent the visual feature space, taking into account the uncertainty presents in the extraction feature process. The idea is to re- present each dimension of the feature space by a fuzzy set, according to the fuzzy partition associated to this dimension. Because the fuzzy representation is strongly dependent of the fuzzy partition, we also propose a novel automatic unsupervised method to obtain the fuzzy partition for each dimension of the feature space based on Fuzzy C-Means clustering. We tested the fuzzy representation, constructed from di erent fuzzy partitions, using synthetic data sets and real data sets. The evaluation of the tests indicated that the fuzzy representation constructed from the proposed fuzzy partition provides excellent results. Finally, di erent similarity measures were applied to the proposed fuzzy representation, indicating that the results are not strongly sensitive to the choice of the similarity measure.