A Multimodal Database for Affect Recognition and Implicit Tagging

Emtions repersentation

In order to represent emtions or feelings accurately, The author represents emtions in an 3-dimensional space, which is the 3D valence-arousal(激励)-dominance(支配) or pleasure-arousal-dominance space.
The valence scale ranges from unpleasant to pleasant.
The arousal scale ranges from passive to active or excited.
The dominance scale ranges from submissive (or “without control”) to dominant (or “in control, empowered”).

Video Selection

Emotion Experiment Paradigm

After watching a video, five questions which were asked during selfreporting were:
1. emotional label/tag,
2. arousal,
3. valence,
4. dominance,
5. predictability
The emtional labels included neutral, anxiety, amusemment, sadness, joy, disgust, anger, surprise, and fear. Questions 2 to 5 were on a nine point scale, which makes participants easy to choose.

Normalize the Features

To reduce the between participant differences, it is necessary to normalize the features. Each feature was separately normalized by mapping to the range [0,1] on each participant’s signals. In this normalization the minimum value for any given feature is subtracted from the same feature of a participant and the results were divided by the difference between the maximum and minimum values.

Cross Validation

At each step of cross validation, the samples of one participant were taken out as test set and the classifier was trained on the samples from the rest of the participants. This process was repeated for all participants’ data.