A topological constrained model-based approach to correspondence problems for branched structures

Type of content
Conference Contributions - Published
Thesis discipline
Degree name
Publisher
University of Canterbury. Computer Science and Software Engineering
Journal Title
Journal ISSN
Volume Title
Language
Date
2014
Authors
Floriello, D.
Botterill, T.
Green, R.
Abstract

Finding correspondences between images representing branched or tree structures is a challenging task. Branched structures are often encountered in many disciplines: anatomy, medicine, agronomy and engineering are only few examples of such fields. The difficulties come from the self-similarity of the structure and frequent occlusions and overlappings, which increase the number of ambiguous matchings. The aim of our research is to correspond images showing vine structure for an application in robotics. In this paper we propose a modelbased approach to the problem, together with a constraint on adjacent branches, which guarantees a topological correctness of the reconstructed structure. Our work shows that the method we tested can find almost all of the correct correspondences and outperforms a maximum likelihood algorithm in terms of precision and recall

Description
Citation
Floriello, D., Botterill, T., Green, R. (2014) A topological constrained model-based approach to correspondence problems for branched structures. Hamilton, New Zealand: 29th International Conference on Image and Vision Computing New Zealand (IVCNZ), 18-21 Nov 2014.
Keywords
Network problems, Trees, Computer vision
Ngā upoko tukutuku/Māori subject headings
ANZSRC fields of research
Field of Research::08 - Information and Computing Sciences::0801 - Artificial Intelligence and Image Processing::080104 - Computer Vision
Rights