Vybrané hybridní metody regionální segmentace s využitím evolučních strategií

Abstract

The main topic of this thesis is the segmentation of medical images using hybrid methods optimized by evolutionary algorithms. The aim of image segmentation is to divide the image into a predetermined number of segmentation classes (regions) and identify objects of interest. The theoretical part describes conventional segmentation methods and evolutionary algorithms. Evolutionary algorithms are optimization methods that find application in solving complex problems where the exact outcome cannot be predetermined. In the context of image segmentation, evolutionary algorithms are used to optimize segmentation method algorithms (e.g., determining threshold values in the histogram). In the practical part, a comparative analysis of conventional segmentation methods (K-means and Otsu) versus hybrid methods optimized using evolutionary algorithms (PSO, DPSO, and GA) is described. As part of the robustness and efficiency analysis, segmentation techniques are tested on medical images artificially degraded by additive noise. The segmentation outputs of the methods used are evaluated using evaluation parameters. The thesis also includes an assessment of the computational complexity of segmentation methods. In conclusion, the results of the comparative analysis are objectively summarized.