Aristotle University Of Thessaloniki (AUTH)
Μη-λεκτική επικοινωνία ανθρώπου-μηχανής: αναγνώριση εκφράσεων του προσώπου
Abstract
dc:descriptionFacial expression recognition is the research topic of this PhD thesis. The aim of this thesis was firstly to create an appropriate facial expression database for the development and the statistical evaluation of facial expression recognition systems, and secondly to develop facial expression recognition methods. In the context of this thesis, the MUG facial expression database was created. The database consists of sequences of posed expressions (six basic expressions) and sequences of induced expressions. The MUG database is the first database where the same subjects perform the expressions deliberately and spontaneously. Another advantage is that it can be used in subject-independent and in subject-dependent experiments, since it includes more than one sequence per expression for each subject. Regarding the development of facial expression recognition methods, this thesis presents two methods. Both methods are based on the same features, the expression vectors. The structure of the feature space is examined. The two approaches that recognize the six basic expressions, were developed based on this structure. The first one assumes that expression vectors that correspond to the same expression live in a manifold and that this manifold can be parameterized by a small number of low-dimensional subspaces. This method is simple and fast, since it uses low-dimensional feature vectors, linear approximations and limited number of low-dimensional subspaces. The method is tested on the MUG database and on the commonly used Cohn-Kanade database, achieving greater accuracy than other recently published methods. The second proposed method assumes that all expression vectors live in a manifold. It parameterizes this manifold using a non-orthogonal sparse representation basis, that is calculated by Independent Component Analysis. The efficiency of this method is tested using four classifiers on both databases (MUG, Cohn-Kanade) in three experimental scenarios. The results indicate that the use of expression vectors in an ICA-based method improves performance compared to the use of image texture. This improvement makes the proposed method outperform or compete other state-of-the-art methods.
Degree
thesis:*- Grantor dc:publisher
- Aristotle University Of Thessaloniki (AUTH)
- Year dc:date
- 2010
Author and committee
dc:creator, dc:contributor.*- Authors dc:creator
-
- Αϋφαντή, Νίκη
- Aifanti, Niki
Subjects
dc:subject × 10- Αναγνώριση εκφράσεων προσώπου
- Σημεία οροθεσίας του προσώπου
- Ανάλυση ανεξάρτητων συνιστωσών
- Facial expressions recognition
- Facial landmark points
- Independent component analysis
- Επιστήμες Μηχανικού και Τεχνολογία
- Επιστήμη Ηλεκτρολόγου Μηχανικού, Ηλεκτρονικού Μηχανικού, Μηχανικού Η/Υ
- Engineering and Technology
- Electrical Engineering, Electronic Engineering, Information Engineering
Rights
- Language dc:language
- gre
Identifiers
dc:identifier.*- Identifier
- 10.12681/eadd/23712