• JoomlaWorks Simple Image Rotator
  • JoomlaWorks Simple Image Rotator
  • JoomlaWorks Simple Image Rotator
  • JoomlaWorks Simple Image Rotator
  • JoomlaWorks Simple Image Rotator
  • JoomlaWorks Simple Image Rotator
  • JoomlaWorks Simple Image Rotator
  • JoomlaWorks Simple Image Rotator
  • JoomlaWorks Simple Image Rotator
  • JoomlaWorks Simple Image Rotator
 
  Bookmark and Share
 
 
Master's Dissertation
DOI
https://doi.org/10.11606/D.45.2021.tde-21072021-152120
Document
Author
Full name
Caio Lopes Demario
E-mail
Institute/School/College
Knowledge Area
Date of Defense
Published
São Paulo, 2021
Supervisor
Committee
Miranda, Paulo Andre Vechiatto de (President)
Rittner, Letícia
Tsuzuki, Marcos de Sales Guerra
Title in Portuguese
Métodos híbridos via energias quadráticas em grafos direcionados para segmentação de imagens 
Keywords in Portuguese
Passeios aleatórios
Segmentação de imagens
Transformada imagem-floresta
Abstract in Portuguese
Neste trabalho são investigados métodos híbridos para segmentação de imagens, tomando como base o algoritmo dos Passeios Aleatórios (Random Walks - RW), porém adotando grafos direcionados, de modo a explorar a polaridade de borda dos objetos. Como resultado temos um novo método híbrido baseado em sementes, chamado OIFT Relaxada (Relaxed OIFT - ROIFT), que estende o método proposto por Malmberg et al. para grafos direcionados, devidamente incorporando a restrição de polaridade de borda. Os resultados de segmentação via ROIFT se encontram entre os obtidos pela Transformada Imagem-Floresta Orientada (OIFT) e pela extensão do RW para grafos direcionados, tal como proposto por Singaraju et al., com estes refletindo casos particulares extremos do novo método. A ROIFT é avaliada em imagens médicas de Ressonância Magnética e Tomografia Computadorizada, produzindo resultados de segmentação superiores e mais intuitivamente corretos que os obtidos por Singaraju et al. e OIFT, além de ser facilmente extensível para imagens multidimensionais. Também é proposto um método híbrido chamado Relaxed Deep Extreme Cut, que estende o método Deep Extreme Cut, de Maninis et al., atuando em algumas de suas deficiências. Esse método é avaliado em imagens bidimensionais do conjunto de dados do GrabCut, apresentando resultados superiores aos obtidos pelo método de Maninis et al.
Title in English
Hybrid methods via quadratic energies in directed graphs for image segmentation
Keywords in English
Image foresting transform
Image segmentation
Random walks
Abstract in English
In this work, we investigate hybrid methods for seeded image segmentation, based on the Random Walks algorithm (RW), but adopting directed graphs, in order to explore the boundary polarity of the ob- jects. As results, we have a new seeded hybrid method, named Relaxed OIFT, which extends a method by Malmberg et al. to directed graphs, to properly incorporate the boundary polarity constraint. Relaxed OIFT lies between the pure Oriented Image Foresting Transform (OIFT) at one end and the extension of Random Walks (RW) to directed graphs as proposed by Singaraju et al., being OIFT and RW particular cases of the proposed method. Relaxed OIFT is evaluated in MR and CT medical images, producing more intuitively correct segmentation results than both OIFT and RW, and being easy to be extended to multi-dimensional images. We also propose a hybrid method, named Relaxed Deep Extreme Cut, which extends the method Deep Extreme Cut by Maninis et al., acting on some of its deficiencies. This method is evaluated in bidimensional images of the GrabCut Dataset, producing results superior to those obtained by the method of Maninis et al.
 
WARNING - Viewing this document is conditioned on your acceptance of the following terms of use:
This document is only for private use for research and teaching activities. Reproduction for commercial use is forbidden. This rights cover the whole data about this document as well as its contents. Any uses or copies of this document in whole or in part must include the author's name.
Publishing Date
2021-09-03
 
WARNING: Learn what derived works are clicking here.
All rights of the thesis/dissertation are from the authors
CeTI-SC/STI
Digital Library of Theses and Dissertations of USP. Copyright © 2001-2024. All rights reserved.