Key words : learning on hierarchical data, image segmentation, edge detection
Hierarchies, as described in mathematical morphology, represent nested regions of interest and provide mechanisms to create concepts and coherent data organization. They facilitate high-level analysis and management of large amounts of data. Represented as hierarchical trees, they have formalisms intersecting with graph theory and applications that can be conveniently generalized.
This work aims to create a learning framework that can operate with hierarchical data and is agnostic to the input and the application. The idea is to study ways to transform the data to a regular representation required by most learning models while preserving the rich information in the hierarchical structure. It proposes to study and formalize the concepts as graphs, a common point for hierarchies and multimedia, and a topic of great interest for machine learning.
The methods in this study use edge-weighted image graphs and hierarchical trees as input, evaluating different proposals on the edge detection and segmentation tasks.
Furthermore, it provides a critical systematic review of proposals in the literature that integrates machine learning and hierarchies. It demonstrates that it is possible to create a learning framework dependent only on the hierarchical data that performs well in multiple tasks with different models.
Co-directeur de thèse : Silvio GUIMARÃES Professeur, PUC-Minas, Brésil
Yukiko KENMOCHI Chargée de recherche DR CNRS, Greyc, France
Laurent NAJMAN Professeur, Université Gustave Eiffel, France
Zenilton PATROCÍNIO Professeur, PUC-Minas, Brésil
Davide BACCIU Professeur associé, Université de Pise, Italie
Alexandre FALCÃO Professeur, Unicamp, Brésil
Ewa KIJAK Maître de Conférence, Université de Rennes, France
Simon MALINOWSKI Maître de Conférence, Université de Rennes, France