Aristotle University Of Thessaloniki (AUTH)
Αναγνώριση προσώπων και εκφράσεων από τρισδιάστατες εικόνες
Abstract
dc:descriptionThis doctoral thesis has focused on face and facial expression recognition using two-dimensional and three-dimensional face images. After a short presentation of the relative bibliography, we present a new face recognition algorithm based on the level-curves (or iso-curves) of the facial depth function. The advantage of this approach is that surface matching is reduced to planar curve matching. The latter is better studied and leads to improved recognition performance in comparison with other techniques that treat face as a rigid surface. However, the previous technique cannot handle face deformation due to expressions that leads to lower recognition rate. Therefore, a new technique for expression-invariant recognition is presented. This technique is based on the assumption that face is deformed isometrically and uses a geodesic polar parameterization of the face to construct expression invariant attribute images. Our technique was compared to other techniques that use geodesic distances between facial points and found to improve performance and handle theoretical and practical problems at the same time. In the following, we describe a new expression recognition technique. First we detect several anatomical points such as the eyes, nose, lips, et.c. by fitting a deformable model to the target face and then we classify the expression of the face based on spacial relations between anatomical point by means of rule-based classifiers. The rules of the latter are found using swarm optimization techniques, namely ant colony optimization and particle swarm optimization. These methods rely on the collective behavior of a number of agents to optimize an objective function, which in our case defines classification rules. Despite their simplicity, they were shown to result in high recognition rates. Finally, we present a new method for joint face and facial expression recognition. Instead of trying to devise suitable feature vectors invariant to unwanted sources of variation, we propose encoding the contribution of each face appearance component (identity and expression) into independent parameters allowing separate control and thus joint face and expression recognition. Identity and expression are decoupled using two bilinear models, a symmetric and an asymmetric one. Experimental results on public face databases showed that the symmetric model favors face recognition while the asymmetric model favors expression recognition.
Degree
thesis:*- Grantor dc:publisher
- Aristotle University Of Thessaloniki (AUTH)
- Year dc:date
- 2009
Author and committee
dc:creator, dc:contributor.*- Authors dc:creator
-
- Mpiperis, Iordanis
- Μπιπέρης, Ιορδάνης
Subjects
dc:subject × 21- 3Δ αναγνώριση προσώπων και εκφράσεων
- Ισοσταθμικές καμπύλες
- Σημειακές υπογραφές
- Γεωδαισιακές πολικές συντεταγμένες
- Βελτιστοποίηση αποικίας μερμηγκιών
- Βελτιστοποίηση σμήνους σωματιδίων
- Συμμετρικό-ασύμμτερο διγραμμικό μοντέλο
- Ελαστικά παραμορφώσιμο μοντέλο προσώπου
- 3D face recognition
- 3D facial expression recognition
- Iso-curves
- Point signatures
- Geodesic polar coordinates
- Ant colony optimization
- Particle swarm optimization
- Symmetric-asymmetric bilinear model
- Elastically deformable face model
- 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/20530