Advancement Through Research
The quality and adequacy of input data play a vital role in the performance of Machine Learning (ML) techniques due to their strong dependency on input data. This dependency creates the limitations for each classifier on the overall classification, segmentation, and detection process. In this article, we propose a graph-cut based integrated approach for classifiers, especially Deep Learning (DL) to augment the two-class image classification and segmentation problem. The segmentation capabilities of an offline trained DL approach is augmented through integration that is employed for pixel-based object detection in images using graph and weights for pixels. The integration process involves the off-line trained DL model, histogram, pixel weights, the neighborhood weights, and the graph-cut, thus boosting the performance. The evaluation results of the DL approach, Bayesian Network (BayesNet), Multilayer Perceptron (MLP), Random Forest and the Histogram approach of Jones and Rehg (JR) on three datasets illustrate that the proposed integrative approach efficiently boosts the detection performance compared to the non-integrated classifiers.
Augmentation, Skin Detection, Classifier, Segmentation, Graph Cuts, Segmentation Boosting
