Valence Detection Using EEG Signals

Abstract

Emotions exert an incredibly powerful force on human behaviour. Strong emotions can cause one to take actions which one might not normally perform or avoid situations that one generally enjoys. Emotion can be seen in person’s face in voice or in gesture such as reaction to stimuli. By emotions human can communicate, express his view, and share his feelings. Most of the people feel surprise, fear, happiness, sadness, disgust and anger many times, these are called basic emotions. Emotions though a psychological phenomenon, can be easily quantified in a three dimensional plane along valence, arousal and dominance axis.EEG signals play a significant role in knowing feelings or emotions of a person. The EEG signals can be acquired using various EEG acquiring devices with electrodes placed using different standards. This work describes the technique used for acquiring EEG signals from 3 participants with electrodes placed as per 10-20 system by using the images provided by International Affective Picture System (IAPS). The sampling frequency is chosen to be 500 Hz. About 80 images with suitable mean Valence and mean arousal values have been used. Some of the images are repeated to elicit the desired emotion in a subject. Various techniques to process upon the raw EEG signals for extracting the features for emotion recognition has as well been explained in this work. It also presents a review on the data acquisition techniques used for acquiring EEG signals for emotion recognition. In an endeavour to classify emotions into two classes namely Low Valence High Arousal (Negative Valence) and High Valence High Arousal (Positive Valence). Event Related Potential features are determined from the processed EEG signals. The event related potential features selected for emotion classification are P100, N100, P200, N200, P300 and N300 collected from three electrodes namely Cz, f3 and p4. The Support Vector Machine classifier has been used to classify the emotions into two classes along the valence axis. An accuracy of 95% is achieved on p4 electrode, 93.75% accuracy on f3 electrode and 92.5% accuracy is achieved on Cz electrode when a polynomial Order of 14 is used for the classifier. In Order to reduce the dataset obtained for classification, average Event Related Potential (ERP) features are also taken into account. The average ERP features in this case are vi collected from 3 subjects but at five electrodes namely Cz, fp1, fp2, p3 and p4. Not only we could reduce the data set but also classified the emotions into two classes by using the SVM classifier with a low polynomial Order and that with improved accuracy. An accuracy of 100% has been achieved for both Cz electrode and a combination of p3-p4electrodes while 91.67% accuracy has been achieved on combination of fp1-fp2 using 5 th Order SVM. Using averaged ERP, the accuracy of emotion classification along valence axis improves and Order of SVM decrease as compared to unaveraged ERP.