Náhodná procházka a její využití v segmentaci obrazu

Abstract

This thesis deals with the problem of Random Walk and it's application to the image segmentation. High quality and fast image segmentation is nowadays one of the most important problem in the image processing. The Random Walk and the Random Walk with Restart are two possible approaches that will be thoroughly studied in this thesis. The Random Walk with restart is also used for obtaining Google-Pagerank therefore finding a fast and efficient low memory solution for a big graphs is currently a crucial problem. One of the main goals of this thesis was to implement parallel solutions on CUDA for these problems and compare them with the CPU versions. The chapter Výsledky shows that we achieved considerable improvements in computational speed. We also made some improvements to obtain better segmentation for the RW algorithm by using the Generative Model from RWR and probability threshold T.