An ensemble approach to improve microaneurysm candidate extraction

Abstract

In this paper, we present a novel approach to microaneurysm candidate extraction. To strengthen the accuracy of individual algorithms, we propose an ensemble of state-of-the-art candidate extractors. We apply a simulated annealing based method to select an optimal combination of such algorithms for a particular dataset. We also present a novel classification technique, which is based on a parallel ensemble of kernel density estimators. The experimental results show improvement in the positive likelihood rate compared to the individual candidate extractors. © 2012 Springer-Verlag.