Image segmentation techniques in the HPC environment and their applications.

Abstract

An image decomposition into regions of interests, which are commonly called segments, is an integral part of modern techniques for classifying and processing a piece of information which is decoded from discrete image functions. Typically, the segments are interesting objects in an image-foreground scene, and they are apart from an insignificant background.

The goal of this diploma thesis is to study at least two different techniques for image segmentation and their massively parallel implementation for HPC platforms, and test them on real-world examples. One of the techniques will be based on a spectral clustering method.

The implemented techniques will be imported into PERMON Toolbox specific modules. PERMON Toolbox is developed by IT4Innovations National Supercomputing Center in Ostrava in cooperation with the Department of Applied Mathematics at VŠB-TUO.