Mean Shift segmentace obrazu pomocí technologie NVIDIA CUDA

Abstract

Man looking at the picture is naturally able to recognize its individual parts and objects displayed. The computer is not capable of doing so, because of it its is appropriate to implement a method that makes it possible. Image segmentation is a group of methods leading to the distribution of digital image into segments built on different principles. One of these methods is the Mean Shift image segmentation, which is one of the clustering methods and its computationally very demanding. The aim of this thesis is to get familiar with this method of segmentation and its implementation, as in the classic C / C + + language, as using massive parallelism NVIDIA CUDA technology. Part of this work is to compare the performance of different implementations and compare their implementations performance.