Integrative Machine Learning augmentation

Abstract

In this article, an integrative approach for augmenting the segmentation capabilities of the off-line trained Machine Learning (ML) classifier is presented. The proposed approach augments the ML performance in the graph cut setup. The integration of the prediction capabilities of the classifiers and neighborhood relationship of the pixels result in increase of segmentation performance. The experimental setup includes an evaluation of the Bayesian Network, Multilayer Perceptron, Random Forest and the Histogram approach of Jones and Rehg [1]. The evaluation results based on the color based detection dataset reveal that the proposed integrative approach improves the detection performance compared to using the off-line classifiers alone.