Supervised Pattern Classification Using Optimum-Path Forest

We present a graph-based framework for pattern recognition, called Optimum-Path Forest (OPF), and describe one of its classifiers developed for the supervised learning case. This classifier does not require parameters and can handle some overlapping among multiple classes with arbitrary shapes. The method reduces the pattern recognition problem into the computation of an optimum-path forest in a graph derived from the data samples. The advantages in efficiency and/or accuracy over Artificial Neural Networks using Multilayer Perceptrons, Support Vector Machines, Bayesian Classifiers and the k-Nearest Neighbors algorithm have been demonstrated in several applications. We present some of these results in this paper.