Filtrace MR obrazu na základě metody Nonlocal Means

Abstract

This diploma thesis is focused on filtration of MR image data by means of Non-Local Means filter. The aim of this thesis was to design an appliacation for objective evaluation and filtration of MR images. In such application MR images can be downloaded, damaged by particular noises with selected values of noise parameters, and subsequently filtered by using the Non-Local Means filter. Part of the developed application makes it possible to carry out simultaneous examination of iage data in individual stages of processing as well as switching beteween undagamaged, damaged by noise, and filtered images. The application also provides information on evaluation parameters determining the quality of filtration. This thesis also deals with testing the Non-Local Means filter on five selected sets of MR images. These sets of data were purposefully degraded by means of deterministic noises used in this thesis by means of value vector of noise parameters. This way it is possible to see and evaluate the dynamic effect of a particular noise in the context of specific MR images. This diploma thesis presents evaluation and comparison analysis of filtration for only two selected noises, including evaluation graphs and charts for one chosen data set. The last part is focused on the evaluation of multilevel OTSU segmentation between an undamaged and damaged image, and undamaged and a filtered image of two chosen areas of a human body for two chosen noises.