Segmentace obrazu s využitím teorie grafů

Abstract

The goal of this thesis is to focus on the problems of image´s segmentation to areas. To read up an efficient graph-based method, which is described in the article by Daniel P. Huttenlocher and Pedro F. Felzenszwalb and then to implement this method and to make out its properties experimentally. In the first part of my thesis where the most widely used techniques are described, I focused on the general problems of the image´s segmentation. Afterwards I properly perused and described an efficient graph-based image segmentation algorithm designed by Daniel P. Huttenlocher and Pedro F. Felzenszwalb. This algorithm is based on the graph representation of image and its principle is the forest creation from minimal spanning tree based on Kruskal’s algorithm. This method I implemented and experimentally tested. My experiments´ results are very affirmative, when right settings are used, algorithm generates very precise segmentation and it can be used even in the real time.