Democritus University of Thrace (DUTH)
Ανάπτυξη αποδοτικών αλγορίθμων για τον υπολογισμό συναρτήσεων ροπών και εφαρμογές στην επεξεργασία εικόνας και αναγνώριση προτύπων
Abstract
dc:descriptionThe subject of the current thesis is the study of moment functions and their application in the field of image analysis. The great performance of moments regarding the description of image content makes them essential with a variety of applications like pattern recognition, image coding, image classification etc. The thesis is based on three research directions: 1) the acceleration of image moments computation, 2) the development of new moment invariants and 3) the development of a new unified methodology for the computation of discrete and accurate continuous moments of color images.Regarding the first research direction, a methodology which is based on a new representation of grayscale images is proposed. This methodology proved to be very effective and independent of the used moment family. The reasons for its performance are both the ability of the method to be combined with any algorithm for accelerating the basis function computations and the fact that it reduces the calculations which take place in the moments computation. The performance of the proposed method depends on the spatial distribution of image intensities. However, this does not constitute a drawback, since, in the worst case, the moments computation time is the same with the traditional method, while, depending on the image content, the computation time can be significantly reduced.The second research direction emerged from the study of the aforementioned methodology. Its great advantage of decomposing the traditional moments into a set of components, lead us to propose a method in order to define new moment invariants. The method is characterized by improved performance and it is independent of the used moment family. Furthermore, it is also detected a drawback regarding the computation of the dual Hahn moment invariants, the study of which leads to their generalization through a new theorem. The proposed dual Hahn moment invariants are more effective than the traditional ones.During the course of this study, it was observed that most of the works of the field are associated with grayscale images. This fact resulted in the latest work which constituted the third research direction. A proper categorization of existing moments as well as the study of their basis functions leads us to propose a general unified methodology for calculating discrete and accurate continuous moments and moment invariants of color images. This method can be applied to the majority of the known moment families as well as to any family, whose basis function satisfies certain conditions. The proposed method demonstrates high discriminative power and great strength to noise conditions.
Degree
thesis:*- Grantor dc:publisher
- Democritus University of Thrace (DUTH)
- Year dc:date
- 2013
Author and committee
dc:creator, dc:contributor.*- Authors dc:creator
-
- Καρακάσης, Ευάγγελος
- Karakasis, Evangelos
Subjects
dc:subject × 20- Ροπές εικόνας
- Συναρτήσεις ροπών
- Ανάλυση εικόνας
- Αναγνώριση προτύπων
- Ταξινόμηση
- Ταίριασμα προτύπων
- Διακριτά; συνεχή; ορθογώνια πολυώνυμα
- Image moments
- Moment functions
- Image analysis
- Pattern recognition
- Classification
- Template matching
- Discrete; continuous; orthogonal polynomials
- Tchebichef; krawtchouk; dual hahn; racah
- Legendre; zernike; pseudo zernike; orthogonal fourier-mellin
- Φυσικές Επιστήμες
- Επιστήμη Ηλεκτρονικών Υπολογιστών και Πληροφορική
- Natural Sciences
- Computer and Information Sciences
Rights
- Language dc:language
- gre
Identifiers
dc:identifier.*- Identifier
- 10.12681/eadd/31558