Estudo de técnicas para separação de objetos agrupados em imagens digitais 2D

• Ano de defesa: 2016
• Autor(a) principal: Preti, Maria Cristina Delgado
• Tipo de documento: Dissertação
• Tipo de acesso: Acesso aberto
• Idioma: por
• Instituição de defesa: Universidade Federal de Uberlândia; Brasil; Programa de Pós-graduação em Ciência da Computação
• Palavras-chave em Português: Computação; Processamento de imagens - Técnicas digitais; Segmentação de imagens; Objetos agrupados; Divisão de agrupamentos; Watershed; Análise de concavidade; Image segmentation; Overlapping objects; Clump splitting; Concavity analysis; CNPQ::CIENCIAS EXATAS E DA TERRA::CIENCIA DA COMPUTACAO
• Link de acesso: https://repositorio.ufu.br/handle/123456789/21239; http://doi.org/10.14393/ufu.di.2016.511

Image segmentation is an important operation in several computer vision and image processing applications, since it represents the first step and most difficult in an image analysis task. One of the factors that make segmentation a challenge is the existence of clustered objects. To deal with this problem, some works focus on the development of computational methods that aim to accurately identify markers in the image, while others are concerned with the development of techniques aimed at selection of concave points on the contourn of clustered objects, as well as in identifying pairs of corresponding points, which are essential to define the subsequent division lines. In this context, this work aims to discuss and compare three important proposals in the literature dealing with the above scenario. In general, it sought to comparatively evaluate the performance of computational methods proposed by the works under study. And when necessary, inconsistent gaps were identified in order to enable improvement and quality. It is evident that the three proposals examined are strongly dependent on user-entered parameters to occur successfully in the final segmentation. Thus, there is a need for a priori knowledge of the image, causing the performance drop, especially when it has a set of images or objects having various shapes, and/or extremely clustered and/or with many concavities. Thus, the three proposals evaluated have vulnerabilities to target clustered objects, either by a gap in the proposed algorithm, or the need to have prior knowledge of the image to be segmented, requiring the insertion of manually parameters. So this makes it difficult to select an optimal method in a real practical situation.