Hierarchical Methods of Image Segmentation

Abstract

Segmentation and filtration of various datasets is still very alive computation problem. Many approaches for such computations exist. They can provide different outputs, speed and many other properties. I am going to deal with the mean-shift method that can be used for filtering and segmenting various datasets. In my case, I am going to filter digital images consisting of pixels. The mean-shift methods achieve good filtration results but most of them suffer from a lower speed. Mean shift can also be used for object tracking where much faster versions of this method exist and are widely used in practice. In this thesis, I am going to talk about the image segmentation problem using various mean-shift approaches and I will present some speed and quality improvements. For this goal, I will use mainly various types of hierarchical approaches. Filtration abilities and the stability of the algorithms will be studied too. Speed issues and problem of a proper setting of parameters will be studied.