Aristotle University Of Thessaloniki (AUTH)
Τεχνικές ψηφιακής επεξεργασίας εικόνας με εφαρμογή στην αναγνώριση προσώπου και συναισθηματικής κατάστασης
Abstract
dc:descriptionThe main field of interest of this thesis includes computational intelligence and image processing techniques in order to recognize facial expressions and achieve face verification. More specifically, the use of elastic graphs is studied thoroughly for facial expression recognition. Facial expression recognition can be greatly improved if the identity of the person posing for the facial expression is known. Moreover, a method that exploits the discriminant information in every phase of the elastic graph matching procedure, in order to achieve automatic face verification, is proposed. Every discriminant step is especially important for the face verification procedure. The presented experiments indicate that the proposed discriminant modifications greatly increase the verification performance. We also improve the performance of elastic graphs by training graph models that are person specific, using expandable graphs. The graphs that are trained have their nodes placed in discriminant facial landmarks and are dense in the facial areas that are considered to be important for face verification and sparse in the facial areas that are less important. Afterwards, inspired by the above mentioned discriminant graphs, a set of techniques for the discriminant decomposition of facial image sets is proposed, based on the Non-negative Matrix Factorization (NMF) algorithm. The discriminant NMF technique decomposes a set of facial images in basis images and appropriate weights, by introducing discriminant constraints at the cost function of the NMF decomposition. The discriminant NMF decompositions produce basis images that can be regarded as discriminant facial parts. Finally, a new family of decision functions is proposed, by modifying the Support Vector Machines functional using the discriminant Fisher criterion. The proposed Support Vector Machines are applied in facial image characterization problems, like the recognition of the neutral state.
Degree
thesis:*- Grantor dc:publisher
- Aristotle University Of Thessaloniki (AUTH)
- Year dc:date
- 2007
Author and committee
dc:creator, dc:contributor.*- Authors dc:creator
-
- Zafeiriou, Stefanos
- Ζαφειρίου, Στέφανος
Subjects
dc:subject × 16- Αναγνώριση προσώπων
- Ταυτοποίηση προσώπων
- Ανάλυση γραμμικής διακεκριμενοποίησης
- Μηχανές εδραίων διανυσμάτων
- Παραγοντοποίηση μη-αρνητικών πινάκων
- Ταίριασμα ελαστικών γραφών
- Face recognition
- Face verification
- Linear discriminant analysis
- Support vector machines
- Non-negative matrix factorization
- Elastic graph matching
- 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/15404