Mean shift na CUDA

Abstract

Mean Shift is a flexible segmentation algorithm, by which we can for example determine individual parts of a picture that belong together. Man is able to recognize these by himself, while for a computer we must first implement this recognition method, for example by using this algorithm. The time demands for computation of this algorithm are however very big because of its complexity, which negatively affects performance and thus its usefulness. This bachelor work aims to analyze the existing algorithm for processor, suggest a solution using parallel processing by CUDA, implement it and evaluate the accomplished goals. This work consists of design and analysis of such algorithm together with evaluation of its results.