Ph.D. Thesis - Modeling Affective State using Learning Vector Quantization
Emotions and stress are vital for our day to day functioning. The busy life we live, can however, cause emotional unbalance. Smart devices can help us remain balanced by providing early warnings, for example, in case of high levels of stress. A key enabler for such devices is to be able to measure emotions and stress. This thesis studies how a certain type of self-learning computer system can be used in this so-called affective domain and explores emotion recognition based upon three different methods of measuring: the body (physiology), the face (facial expressions) and the brain (cognitive processes). By doing so, this thesis combines theory and application at the boundary of computing science and psychology, also termed affective computing.
The research performed shows that computers, based upon these self-learning systems, can detect emotions from photos of facial expressions and can detect stress from cardiac signals, both with high accuracy. The methods used, also provide models based upon which new knowledge can be gained. As an example, the mouth and eyes were found most vital for recognizing facial expressions and it was found that emotion recognition from physiology can be improved by adding measurements of specific heart frequencies.
Some typesetting errors were found after printing the thesis. These have been corrected in the version that is available for download above.
Click here for a list of errata.
Errata:
- Corrected title of 7.4.1. to "Mapping to a dimensional model", (page vii and 136)
- Corrected quotation marks on page 5, 7, 64, 65, 66, 67, 84, 86, 90, 107 and 127
- Corrected Equation (2.14) on page 17 to "d_{\lambda,T}^\mu=\sum_{i=1}^N \lambda^\mu_i(\xivec_i^\mu-\wvec_{T,i}^\mu)^2"
- Corrected Equation (2.17) on page 18 to "\Lambda^\mu=(\Omega^\mu )\transp \Omega^\mu"
- Corrected reference to Chapter 2 on page 24, 25 and 26
- Improved positioning of Equation (3.18) on page 31
- Corrected sentence on page 103 to "The second step of mapping appraisals to ... did not involve more sophisticated classification techniques ..."
- Corrected sentence on page 109 to "Note that, by following this procedure, we fully isolated ..."
- Removed extra ")" on page 111 and 112
- Corrected reference to Equation (6.1) on page 112
- Adjusted spacing on page 117 and 134
- Corrected sentence on page 121 to "In all settings there were at least 227 data samples ..."
- Corrected sentence on page 131 to "When \tau is chosen to represent ... will represent the
very handful of really stressful events ..."
- Corrected sentence on page 133 to "The paced breathing exercise is ..."
- Corrected reference to Figure 7.5 on page 138