Aristotle University Of Thessaloniki (AUTH)
Αναγνώριση συναισθημάτων από ηλεκτροεγκεφαλογράφημα μέσω προηγμένων μεθόδων επεξεργασίας σήματος
Abstract
dc:descriptionThe aim of the present thesis is to develop advanced signal processing techniques for Electroencephalogram (EEG)-based emotion recognition. The proposed methodologies lie in two major lines: (i) the extraction of valuable information, related with emotions, that would lead to the enhancement of the emotion recognition task and (i) the evaluation of emotion elicitation procedures, as far as their ability to induct emotions is concerned. Moreover, the aforementioned methodologies are based on the basic principles that govern the emotional expression in EEG signals, as are proved from the current findings in neuroscience, and intend to incorporate those principles in the relative emotion recognition systems. In the first part of the thesis, the recognition of the six basic emotions is examined, i.e., the emotions of happiness, surprise, anger, fear, disgust, and sadness. For the elicitation of the aforementioned discrete emotions, the theory behind the Mirror Neuron System (MNS) is adopted. According to MNS, the EEG activity that is expressed when a subject feels an emotion, is the same or akin when a subject is actually watches another person expressing this emotion through face or body gesture, via a mechanism that is responsible for decoding other people's intentions. A new technique for the classification of the six basic emotions is proposed, where a feature set drawn from Higher Order Crossings (HOC) analysis is employed. Subsequently, an attempt to efficiently isolate the emotion-related EEG characteristics, a novel algorithmic tool is proposed, namely Hybrid Adaptive Filtering (HAF). High classification rates are accomplished, reaching 85%. In the second part of this thesis, the recognition of generalized affective states is studied. These affective states are not described as discrete emotions but rather as states characterized by two fundamental dimensions, arousal and valence. The subjective criteria that govern the emotional evaluation of the pictures emotional content, revealed the need of a measure that would evaluate the emotion elicitation trial in terms of its ability to induct emotional experience. Towards that, an index is proposed, namely Asymmetry Index (AsI). Furthermore, in order to localize the emotional experience in time and better isolate the frequency components that incorporate emotional information, four different implementations of the AsI measure are proposed. The resulted methodologies accomplish, as the HAF algorithm does, a kind of “emotional” filtering that boosts the emotional content of the filtered signals and leads to better classification results. Finally, for the first time in the EEG-based emotion recognition area, the detection of emotional transitions is studied. The conduct of a specifically designed experiment, that aims to elicit emotional transitions, is described. Moreover the methodology to detect these transitions is presented. The method proposed, is based on a variation of the AsI index, i.e., the transitional AsI (TAsI). The effectiveness of TAsI is tested for various subjects' mood scenarios and its dependence from mood is revealed. The whole approach shows promising results and lays the foundations for a more realistic implementation of EEG-based emotion recognition systems.
Degree
thesis:*- Grantor dc:publisher
- Aristotle University Of Thessaloniki (AUTH)
- Year dc:date
- 2011
Author and committee
dc:creator, dc:contributor.*- Authors dc:creator
-
- Πετραντωνάκης, Παναγιώτης
- Petrantonakis, Panagiotis
Subjects
dc:subject × 12- Επεξεργασία σήματος
- Συναισθηματική υπολογιστική
- Ηλεκτροεγκεφαλογράφημα
- Αναγνώριση συναισθημάτων
- Signal processing
- Affective computing
- Electroencephalogram (EEG)
- Emotion recognition
- 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/28049