Optimum-Path Forest based on k-connectivity: Theory and applications

Imagem de Miniatura
Papa, Joao Paulo [UNESP]
Nachif Fernandes, Silas Evandro
Falcao, Alexandre Xavier
Título da Revista
ISSN da Revista
Título de Volume
Elsevier B.V.
Graph-based pattern recognition techniques have been in the spotlight for many years, since there is a constant need for faster and more effective approaches. Among them, the Optimum-Path Forest (OPF) framework has gained considerable attention in the last years, mainly due to the promising results obtained by OPF-based classifiers, which range from unsupervised, semi-supervised and supervised learning. In this paper, we consider a deeper theoretical explanation concerning the supervised OPF classifier with k-neighborhood (OPFk), as well as we proposed two different training and classification algorithms that allow OPFk to work faster. The experimental validation against standard OPF and Support Vector Machines also validates the robustness of OPFk in real and synthetic datasets. (C) 2016 Elsevier B.V. All rights reserved.
Pattern classification, Optimum-Path Forest, Supervised learning
Como citar
Pattern Recognition Letters. Amsterdam: Elsevier Science Bv, v. 87, p. 117-126, 2017.