Emotion Recognition based on the multiple physiological signals

Emotion recognition became a key research issue in human-computer field. The inherent limitations are too many features to recognize emotions. However, the fusion of features may improve the performance, which was not investigated widely. This paper presents a study for emotion recognition by fusion features from multiple physiological signals. Four kinds of physiological signals were considered: electrocardiogram (ECG), electromyogram (EMG), respiratory changes (RSP) and skin conductivity (SC). For each signal, four kinds of features were extracted: time domain features, time frequency features, nonlinear features and intrinsic mode function (IMEs) features using the EEMD decomposition method. A C4.5 decision tree classifier was employed to select features that contribute the most to the classification performance. Then, the classifier is used to recognize emotions using the features set which fused the selected features. Finally, this study also tested various feature sets to recognize emotions. Experimental results showed that IMF based features outperform other features and a small feature set consisting of the selected features can detect emotions much more precisely than using a larger feature set.