PROJECT TITLE :

MPED A Multi-Modal Physiological Emotion Database for Discrete Emothion Recognition

ABSTRACT:

In this study, we construct and build a multi-modal physiological emotion database to investigate human emotions. This database captures four modal physiological signals: EEG, galvanic skin reaction, breathing, and EKG (ECG). We particularly collect an emotion elicitation material database picked from more than 1500 video clips to lessen the influence of culture dependent elicitation materials and provoke desirable human feelings. We carefully select 28 videos as standardized elicitation samples, which are examined using psychological procedures, based on a significant degree of stringent man-made labeling. When participants saw these standardized video clips that described six different emotions and neutral emotion, their physiological signals were concurrently recorded. Distinct Feature Extraction methods and classifiers (support vector machine and k-NearestNeighbor) were applied with three different classification protocols to recognize the physiological responses of different emotions, which gave the baseline results. Simultaneously, we introduce a new attention-long short-term memory (A-LSTM) system that improves the efficiency of useful sequences in extracting more discriminative features. Correlations between the EEG signals and the ratings of the participants are also explored. The database has been made open to the public in order to encourage other researchers to test their own emotion estimate methods against it.