Využití spektrálního shlukování pro segmentaci obrazu

Abstract

Spectral clustering is a method which can separate input points into individual regions on the basis of eigenvalues and their corresponding eigenvectors of similarity matrix made for these input points. This method has been successfully applied on the segmentation of various input points outside the domain of image analysis. This diploma thesis aims to explore this method and experimentally verify its usability for the needs of digital image segmentation. The thesis also aims to describe the influence of setting the input parameters for this method on the results, because this influence is not described in too much detail in the available literature. In the framework of this thesis it will also be shown, which libraries could be used for effective computation of eigenvalues and their eigenvectors. Attention will also be paid to the analysis of implementation of such program using the method of spectral clustering as well as its results.