Segmentação de texturas em imagens digitais utilizando wavelets redundantes

Carregando...
Imagem de Miniatura

Data

2015-09-25

Autores

Dobler, Joyce Aline Duarte [UNESP]

Título da Revista

ISSN da Revista

Título de Volume

Editor

Universidade Estadual Paulista (Unesp)

Resumo

The texture analysis plays an important role in many applications, such as analysis of medical imaging, industrial inspection, remote sensing, etc. The main applications that use the analysis of textures is the segmentation. In particular, the study related to the textures segmentation spans for more than five decades, however, it continues to be topic of enormous interest. Because of the variety of contexts and situations in which the term texture is used, there isn't formal definition of the concept. Intuitively, texture is related to tonal variations (gray level) in images, and can be that considered local discontinuities that are generally associated with edges. The edge detection is generally seen as a process that reduces the amount of data that represents a sign, however, the wavelet transform is a powerful tool that uses scale, directionality and the local energy distribution of the edges, and provides a complete and stable representation of a signal, without loss of information. In this study, we studied and adapted the texture segmentation method developed by Pagamisse using the dyadic wavelet transform with four directions, including starlet wavelet transform in order to improve their performance. The method agregates and combines the responses of the redundant wavelets transform (invariant by translation) in scales and orientations, providing the best segmentation of textures. The efficiency of the method constructed was proven by applying to mosaics of textures and comparing the results obtained with the results of other texture segmentation methods found in the literature
A análise de texturas desempenha um papel importante em diversas aplicações, tais como análise de imagens médicas, inspeção industrial, sensoriamento remoto, etc. Dentre as principais aplicações que utilizam a análise de texturas, está a segmentação. Em particular, o estudo relacionado à segmentação de texturas já se estende por mais de cinco décadas, entretanto, continua sendo tópico de enorme interesse. Devido à variedade de contextos e situações em que o termo textura é utilizado, não se tem uma definição formal do seu conceito. Intuitivamente, textura está relacionada às variações tonais (níveis de cinza) nas imagens que podem ser consideradas descontinuidades locais e, em geral, estão associadas às bordas. A detecção de bordas é geralmente vista como um processo que reduz a quantidade de dados que representa um sinal, porém, a transformada wavelet é uma poderosa ferramenta que utiliza escala, direcionalidade e a distribuição local da energia das bordas e fornece uma representação completa e estável de um sinal, sem perda de informações. Neste trabalho, estudou-se e adaptou-se o método de segmentação de texturas desenvolvido por Pagamisse que utiliza a transformada wavelet diádica com quatro direções, incluindo a transformada wavelet starlet no intuito de melhorar o seu desempenho. O método agrega e combina as respostas das transformadas wavelets redundantes (invariantes por translação) em escalas e orientações, fornecendo a melhor segmentação de texturas. A eficiência do método construído foi comprovada aplicando-o em mosaicos de texturas e comparando os resultados obtidos com os resultados de outros métodos de segmentação de texturas encontrados na literatura

Descrição

Palavras-chave

Computação - Matematica, Wavelets (Matematica), Imagens de sensoriamento remoto, Cores, Computer science - Mathematics

Como citar

DOBLER, Joyce Aline Duarte. Segmentação de texturas em imagens digitais utilizando wavelets redundantes. 2015. 114 p. Dissertação (mestrado) - Universidade Estadual Paulista Júlio de Mesquita Filho, Faculdade de Ciências e Tecnologia, 2015.