University of Patras
Αναγνώριση προτύπων σε εικόνες και ακολουθίες πολυμέσων με δομές πολλαπλότητας
Abstract
dc:descriptionImages and multimedia data are intrinsically characterized by high dimensionality. At the same time, applications that utilize such data are often based on the estimation of a smaller number of macroscopic parameters that constitute the schematic information of the data. From a geometric perspective, the above remark translates to the fact that data characterized by the same set of macroscopic parameters lie on multidimensional and non-linear geometrical structures, within the primary data space, called manifolds. During the past decade, the field of pattern recognition has witnessed a systematic effort to develop effective methods for manifold learning, especially for applications on image and multimedia data. At the same time, recent advances in other fields of computer science have offered new tools and perspectives into manifold learning, as well as in addressing of non-linear relations between primary data and schematic content in pattern recognition applications in general.The subject of this thesis is the utilization of cotemporary manifold learning techniques and modern tools from disciplines such as graph theory and redundant representations, in pattern recognition and analysis of image and multimedia data. Exploitation of the intrinsic geometric structure of the data is studied both at the level of local manifold, where the geometrical properties of local features from individual signals are analyzed, and the level of global manifold where more general manifold structures characterizing signals of specific nature are considered. A wide range of applications were studied, spanning from biometrics (face recognition, signature recognition, action recognition) to biomedical imaging, where methods for classification of images from fluorescence microscopy are proposed. For each application a complete system is presented, properly utilizing the information of manifold structures, while several new descriptors and data encoding schemes were developed, where necessary. Furthermore, novel manifold-to-manifold distance measures are proposed for the comparison of local manifold structures in pattern recognition and analysis applications.
Degree
thesis:*- Grantor dc:publisher
- University of Patras
- Year dc:date
- 2014
Author and committee
dc:creator, dc:contributor.*- Authors dc:creator
-
- Theodorakopoulos, Ilias
- Θεοδωρακόπουλος, Ηλίας
Subjects
dc:subject × 28- Αναγνώριση προτύπων
- Δομές πολλαπλότητας
- Εκμάθηση πολλαπλότητας
- Πολυμέσα
- Επεξεργασία εικόνας
- Επεξεργασία σήματος
- Θεωρεία Γράφων
- Πλεονάζουσες αναπαραστάσεις
- Αραιή αναπαράσταση
- Αναπαράσταση ανομοιοτήτων
- Pattern recognition
- Manifold structures
- Manifold learning
- Multimedia
- Image processing
- Signal processing
- Graph theory
- Redundant representations
- Sparse representation
- Dissimilarity representation
- Φυσικές Επιστήμες
- Επιστήμη Ηλεκτρονικών Υπολογιστών και Πληροφορική
- Επιστήμες Μηχανικού και Τεχνολογία
- Επιστήμη Ηλεκτρολόγου Μηχανικού, Ηλεκτρονικού Μηχανικού, Μηχανικού Η/Υ
- Natural Sciences
- Computer and Information Sciences
- Engineering and Technology
- Electrical Engineering, Electronic Engineering, Information Engineering
Rights
dc:rights- Statement dc:rights
-
- BY_NC_ND
- Language dc:language
- gre
Identifiers
dc:identifier.*- Identifier
- 10.12681/eadd/35319