Segmentace obrazu metodou spektrálního shlukování a difuzního spektrálního shlukování

Abstract

In recent years, spectral clustering has established itself as an robust segmentation algorithm. Using spectral clustering for, particularly real, image segmentation opens a wide scope to optimize and modify this algorithm further. This thesis introduces the theoretical background of spectral clustering algorithm focusing on its different modifications including diffuse spectral clustering. Experimental part of this thesis focuses on the implementation of spectral diffuse clustering using the Mean-shift algorithm and based on its outputs, using both real and synthetic inputs, it provides a sober perspective of possibilities of using spectral clustering for image segmentation concerning various use cases.