Watershed com marcadores propagados por casamento de padrões de pontos via correspondência de grafos

• Ano de defesa: 2013
• Autor(a) principal: Ortoncelli, André Roberto
• Tipo de documento: Dissertação
• Tipo de acesso: Acesso aberto
• Idioma: por
• Instituição de defesa: Universidade Estadual de Maringá; Brasil; Departamento de Informática; Programa de Pós-Graduação em Ciência da Computação; UEM; Maringá, PR; Centro de Tecnologia
• Palavras-chave em Português: Watershed com marcadores propagados; Segmentação morfológica de imagens; Casamento entre padrões de pontos via correspondência entre grafos; Segmentação hierárquica; Segmentação interativa de objetos em sequências de imagens; Brasil.; Watershed from propagated markers; Morphological image segmentation; Graph Matching; Hierarchical segmentation; Interactive segmentation of objects in image sequences; Brazil.; Ciências Exatas e da Terra; Ciência da Computação
• Link de acesso: http://repositorio.uem.br:8080/jspui/handle/1/2493

Watershed from propagated markers is a generic method to interactive segmentation of objects in image sequences, through the combination of classical watershed from markers technique to motion estimation. This dissertation introduces two variations of the watershed from propagated markers, supported by graph matching methods. Basically, both variations consists in: (i) the markers are propagated by a graph matching method that computes, by minimizing a cost function, the matching between each edge of a graph (that represents markers created around the segmentation mask of the current frame) with one edge of another graph that denotes the hierarchical segmentation of the next frame of the sequence. Each edge resulting from this matching is used as a pair of markers applied to the segmentation of the next frame; and (ii) first, a pre-segmentation mask is computed for the next frame of the sequence, this mask is computed by a graph matching method for image segmentation that computes the matching between two graphs: one which represents the model of the object to be segmented and other which represents the result of the hierarchical image segmentation. Final markers are created around the pre-segmentation mask. Both methods were submitted to the application of a benchmark that quantitatively assesses assisted object segmentation methods in image sequences. Such benchmark was applied to as a comparison tool among the results of the two proposals and other segmentation methods, such as other watershed from propagated markers variations and semi-automatic segmentation methods found in literature as well. Experimental results shows that both proposals in this dissertation are promising ones since they are robust and provides a substantial gain in the segmentation error reduction.