The doctoral dissertations of the former Helsinki University of Technology (TKK) and Aalto University Schools of Technology (CHEM, ELEC, ENG, SCI) published in electronic format are available in the electronic publications archive of Aalto University - Aaltodoc.
|
|
|
Studies in Probabilistic Methods for Scene Analysis
Timo Kostiainen
Dissertation for the degree of Doctor of Science in Technology to be presented with due permission of the Department of Electrical and Communications Engineering for public examination and debate in Auditorium E at Helsinki University of Technology (Espoo, Finland) on the 27th of October, 2006, at 12 noon.
Dissertation in PDF format (ISBN 951-22-8412-X) [2042 KB]
Dissertation is also available in print (ISBN 951-22-8411-1)
Abstract
In this thesis, probabilistic methods are applied to a number of problems in computer vision. The goal is to provide means for a vision based system that is able to analyze and recognize scenes and objects in camera images and to use that information for autonomous navigation and machine learning. New methods are developed for different functions that are needed in such a system, including segmentation of images, model-based recognition of objects, robot navigation and model complexity control.
The approach is based on generative probability models, and Bayesian statistical inference is used to match these models with image data. Stochastic sampling methods are applied to obtain numerical results.
The self-organizing map is a neural network algorithm that has many applications in computer vision. In this thesis, the algorithm is analyzed in a probabilistic framework. A probability density model is derived and new model selection techniques are proposed, which enable complexity control for the self-organizing map.
The analysis of images is discussed from the point of view of segmentation and object recognition. Segmentation aims at dividing the image into parts of different appearance, while object recognition is meant to identify objects that fulfill given criteria. These are different goals, but they complement each other. When the recognition of all objects in an image is not possible, segmentation can provide an explanation to the rest of the image. For object recognition, different two and three dimensional object models are considered and Bayesian matching techniques are applied to them. Efficient techniques for image segmentation are proposed and results are presented.
Keywords: scene analysis, MCMC methods, image segmentation, object matching, self-organizing maps
This publication is copyrighted. You may download, display and print it for Your own personal use. Commercial use is prohibited.
© 2006 Helsinki University of Technology
Last update 2011-05-26