This project on emotion recognition is studying the physiological changes that occurred in an actor as she intentionally induced eight different emotional states over a period of several days. These changes were measured with bio-sensors including electromyogram (EMG) placed on the jaw, blood volume pressure (BVP), skin conductivity (GSR), and respiration. This study includes extracting and analyzing useful features from the physiological signals of each state, with the intention of deveoping algorithms that can discriminate between emotional states. The different states studied in this experiment were: 'No Emotion', 'Anger', 'Hate', 'Grief', 'Platonic Love', 'Romantic Love', 'Joy', and 'Reverence', recorded over 3-minute periods each.
Sample biosignal readings from two emotional states.
Following the diagram of affect pattern recognition (below), several features are extracted from each signal, including the mean and variance. Each emotion each day is therefore characterized by a set of values-features ranging from 24 to 40, depending on which features we decide to include in the analysis. So each emotion each day becomes a point in a multidimensional space. The classification can take place in this space, on an arbitrary subspace of it, or in one of reduced dimensionality, produced by the Fisher Projection algorithm. In any case, gaussian probability distributions are then fitted to the data. Further data points are classified according to their posterior probabilities. The first classifiers gave a classification result of 88% success when discriminating among 3 emotions (pure chance would be 33.3%), and of 51% when discriminating among 8 emotions (pure chance 12.5%). New, improved classifiers reach an 80% success rate when discriminating among all 8 emotions. Furthermore, an online classifier has been built using the old method, which gives a success rate only 8% less than its old offline counterpart (i.e. 43%). We expect this percentage to sharply increase if the new methods are used instead.
Affect pattern recognition diagram.