A multi-objective decomposition estimation of distribution algorithm for multilevel thresholding image segmentation

Melanoma is a skin cancer that can cause death in humans. Using tools such as digital image processing helps professionals diagnose it with greater reliability. Image segmentation by multilevel thresholding is used to extract homogeneous regions of an image using information from pixel intensities. This work presents a methodology based on multi-objective optimization for medical image segmentation. The between-class variance and the minimum cross-entropy are used as a criterion for the segmentation. The optimization is done using an Estimation of Distribution Algorithm (EDA) and compared with a Decomposition-based Multi-objective Evolutionary Algorithm EDA (MOEA/D-EDA). The thresholds obtained with the multi-objective algorithm are selected by approximating the Pareto Fronts that balance the two objectives. The results compared to a single optimization approach are better visually.