Análise de textura utilizando descritores fractais estimados pela interferência mútua de grupos de píxeis coloridos

• Ano de defesa: 2019
• Autor(a) principal: Marasca, Andre Luiz
• Tipo de documento: Dissertação
• Tipo de acesso: Acesso aberto
• Idioma: por
• Instituição de defesa: Universidade Tecnológica Federal do Paraná; Pato Branco; Brasil; Programa de Pós-Graduação em Engenharia Elétrica; UTFPR
• Palavras-chave em Português: Automação; Fractais - Programas de computador; Microeletrônica; Automation; Fractals - Computer programs; Microelectronics; CNPQ::ENGENHARIAS::ENGENHARIA ELETRICA; Engenharia Elétrica
• Link de acesso: http://repositorio.utfpr.edu.br/jspui/handle/1/4431

Texture analysis is a technique that has been widely used in systems that use automatic image pattern recognition as an alternative to human cognitive ability. Applications of this technique directly benefit areas such as control theory, manufacturing, biomedical, etc. The main idea is to extract numerical characteristics of images, to classify them. The main problem involving texture analysis is its accuracy. The fact that there are no 100% accurate classification appro- aches so far implies that the applications using the technique impose an operational margin of error and, when this margin exceeds the tolerable limit for a given application, its use is unfeasible. Recently, an approach to texture analysis based on fractal theory has been proposed in the literature, which potentiates the indices of correct classification. In this context, the present work presents a contribution to the texture analysis area by fractal theory proposing a method of feature extraction of texture images based on the idea of labels. The proposal consists of converting a two-dimensional image into clouds of labeled points, which are then enlarged revealing characteristics of the texture image. The texture descriptors are obtained from the relation between the volume of the dilated points, their labels and the radii of dilation. The validation of the proposed method was performed by estimating its accuracy on benchmark image datasets and comparing it with the accuracy of texture analysis methods in the literature. The results presented show that the proposed approach provides a more robust and accurate texture descriptor than some of the results in the state of the art, thus expanding the practical aspect of the technique.