Metoda medoidshift a její využití pro filtraci a segmentaci obrazu

Abstract

Image segmentation is one of the most important processes in digital image processing. There are known many methods of automatic image segmentation requiring more or less user-defined parameters. One of later method is so-called medoidshift. This method doesn’t require user to arbitrarily specify number of segments prior to computation. It also doesn’t require specifying iteration stopping criterion. From this point of view can be medoidshift classified as a non-parametric method. Likewise in similar method, meanshift, the only one required parameter is so-called window width. The goal of this work was to implement medoidshift algorithm, explore its behavior in image segmentation process and compare its performance to the meanshift. Properties of the medoidshift method were explored on real and artificial images including performance on images including white Gaussian noise.