Segmentação de imagens por texturas através de um método de agrupamento de dados

Segmentation is one of the most complex process in image processing and the focus on research in this area has been increasing. Segmentation, consists of dividing an image into distinct regions from pixel properties, as gray level and texture. This paper presents a new method of images segmentation...

Descripción completa

Detalles Bibliográficos
Autor: Carvalho, Taiane Viana de
Tipo de recurso: tesis de maestría
Estado:Versión publicada
Fecha de publicación:2018
País:Brasil
Institución:Universidade Federal do Rio Grande do Norte (UFRN)
Repositorio:Repositório Institucional da UFRN
Idioma:portugués
OAI Identifier:oai:repositorio.ufrn.br:123456789/25963
Acceso en línea:https://repositorio.ufrn.br/jspui/handle/123456789/25963
Access Level:acceso abierto
Palabra clave:Processamento de imagens
Segmentação de imagens
Texturas
Mistura de gaussianas
Clustering
CNPQ::ENGENHARIAS::ENGENHARIA ELETRICA
Descripción
Sumario:Segmentation is one of the most complex process in image processing and the focus on research in this area has been increasing. Segmentation, consists of dividing an image into distinct regions from pixel properties, as gray level and texture. This paper presents a new method of images segmentation based on extraction of textures features in images. The method consists of forming a dataset based on the pixel values in the image and then group the data into classes that represent the textures that make up the image. Pixel grouping occurs through the clustering application technique. This technique group the points in the data set in auxiliary centers using vector quantization. The distance between the auxiliary centers is estimated and all centers that have a distance less than a threshold are linked together. Conventional distances used in clustering basically grow with the square of the distance in means. An alternative used in this paper is a new method based on gaussian mixture concept, witch estimates the separation between a data set modeled by Gaussians with model of one single Gaussian. The purpose of this approach is to perform the separation of regions with different textures in an image using an efficient algorithm.