Segmentação de imagens similares por casamento de descritores de textura

• Ano de defesa: 2019
• Autor(a) principal: Souza, Juliana Gonçalves de
• Tipo de documento: Dissertação
• Tipo de acesso: Acesso aberto
• Idioma: por
• Instituição de defesa: Universidade Estadual de Maringá; Brasil; Programa de Pós-Graduação em Ciência da Computação; UEM; Maringa; Centro de Ciências de Tecnologia
• Palavras-chave em Português: Processamento gráfico; Segmentação de imagem; Ciências Exatas e da Terra; Ciência da Computação
• Link de acesso: http://repositorio.uem.br:8080/jspui/handle/1/5359

With the popularization and improvement of image and video capture devices, it is becoming increasingly common the usage of this method for data recording, both for leisure and for professional purposes, for education, safety and health. These formats require specific processes for manipulation and processing. One such process is the image matching, which seeks to find common elements between a reference image and one or more objective images. In this work the JUG method for segmentation of similar images is proposed, consisting of three main phases: initial segmentation of the images, extraction of characteristics of the regions and matching between the regions of the reference image and the objective image. The first phase is performed through hierarchical segmentation with minimum area parameter, implemented through the watershed technique with markers. The second phase uses the LBP algorithm to extract the texture feature. The matching between regions is performed by comparing feature vectors using a similarity measure. For a more detailed analysis of the operation and effectiveness of the method, three empirical studies composed of nine experiments were performed with good results. The JUG method was then used to elaborate two methods of segmenting objects into sequences of images. The first semi-automatic segmentation method was used to analyze the JUG method in the context of this problem and presented good results in simple sequences. The second method has interactive targeting and has been validated with the characteristics of this type of segmentation