Special Topics in Computer Science:

Random Fields and Active Contours in Image Segmentation

There are two main approaches to image segmentation: Contour-based methods aim at detecting the boundary between regions, while region-based methods try to identify the regions directly. In this course we will address modern techniques of both approaches.
1) Active contours are energy-minimizing splines used to locate region boundaries.
2) Markov Random Fields are often used to capture contextual constraints in a probabilistic framework, allowing for efficient region-based segmentation of complex images.

Lecturer

Dr. Zoltan Kato
University of Szeged
Institute of Informatics
Department of Image Processing and Computer Graphics
kato@inf.u-szeged.hu

Dates

Contents

Part I: Variational methods

1. Mumford-Shah energy functional
2. Classical Active contour (Snake)
3. Region-based Active contours

Part II: Markov Random Fields

1. Segmentation model
2. Energy minimization
3. Parameter estimation via the EM algorithm

Exam

Students will have to do a project and send it to the lecturer. The marks for this course will be based on the project.

Literature

Some brief lecture notes are available for part I:

• Mumford-Shah energy functional & weak membrane model
• Snakes

For part II the following PhD thesis covers most of the material presented:

• Zoltan Kato: Multi-scale Markovian Modelisation in Computer Vision with Applications to SPOT Image Segmentation.