Segmentace obrazu s využitím spektrálního shlukování

Abstract

The goal of this bachelor work is to verify the possibilities of image segmentation by making use of the Laplacian matrix of the graph (grid) that is created by image, which is followed by subsequent clustering. In the bachelor work, do the following: 1) Get acquainted with the technique of spectral clustering (e.g. from the paper "A tutorial on Spectral Clustering" by Ulrike von Luxburg). 2) Verify the usefulness of spectral clustering in the area of image segmentation. Pay attention to the selection of efficient method of computing eigenvalues and eigenvectors (e.g. the ARPACK library can be used). Also, pay attention to the selection of the clustering method. 3) Improve the method from the previous paragraph into the method of diffusion spectral clustering (Gaura, Sojka: Diffusion Spectral Clustering for Image Segmentation). 4) Both the methods should be tested. The results should be summarised.

The software should be created in C/C++.