Segmentação interativa de imagens utilizando competição e cooperação entre partículas

• Ano de defesa: 2015
• Autor(a) principal: Silva, Bárbara Ribeiro da [UNESP]
• Tipo de documento: Dissertação
• Tipo de acesso: Acesso aberto
• Idioma: por
• Instituição de defesa: Universidade Estadual Paulista (Unesp)
• Palavras-chave em Português: Computação - Matematica; Processamento de imagens - Tecnicas digitais; Segmentação de imagens; Algoritmos de computador; Aprendizado do computador; Computer science Mathematics
• Link de acesso: http://hdl.handle.net/11449/138499; http://www.athena.biblioteca.unesp.br/exlibris/bd/cathedra/08-04-2016/000863304.pdf

To study and classify images it is necessary to understand the subdivisions they present. For this purpose there are image segmentation techniques, one of the computer vision tasks, which consists in the distribution of a digital image into multiple segments, regions or objects, in order to facilitate analysis and understanding. Image segmentation methods are often used for locating objects in images and borders. For this task, machine learning algorithms are often employed; scientific discipline that deals with the design and development of algorithms that automatically improve with experience, imitating the human learning behavior. The machine learning process can be used in different ways, however the category that best adapted to the project proposal, which represents a blend of supervised learning and unsupervised learning, called semi-supervised learning. Such category of algorithms works with both labelled and unlabelled data in their training phase, allowing the use of individual characteristics and or per set. Given the necessity of combining semi-supervised learning with image segmentation, some means of using information delivered from image to feed the semi-supervised learning machine were studied. As a result of this study, the semi-supervised learning algorithm known as competition and cooperation among particles was extended to be applied in the interactive image segmentation task. In the input image, some pixels are classified by the user and serve as a starting pixels to the algorithm. These pre-labeled points are used to create particles in the proposed model. Also, some parameters and algorithm structures were changed to optimize the classification and means of expressing the results in order to make it more suitable for the image segmentation task