Segmentace hlavy v hloubkových obrazech s využitím grafových metod

Abstract

The purpose of this thesis is the design and implementation of application, that automatically detects and segments a human head from depth images using graph cuts. Resulting program allows the user to set various parameters of detection, including types of algorithms; variants of graph cuts - basic, Grab Cut and also Watershed which uses different approach to segmentation. Detection between following frames is optimised using the information from previous images to find the head. On tested frames, the accuracy of 80% was achieved using the optimal configuration.