Application of fractional Fourier transform in feature extraction from ELECTROCARDIOGRAM and GALVANIC SKIN RESPONSE for emotion recognition

Abstract

Emotion recognition from physiological signals plays an essential role in human–computer interaction and affective computing. This paper aims to study the effectiveness of Fractional Fourier Transform (FrFT) as a novel feature extraction method in improving the accuracy of emotion recognition from physiological signals. Emotion detection is performed in two dimensions, of arousal and valence, using Electrocardiogram (ECG) and galvanic skin response (GSR) signals recorded on the ASCERTAIN database. Features extracted in the FrFT, time, and frequency domains are classified using two binary SVMs. The results suggest the usefulness of the phase information of the FrFT coefficients in arousal and valence detection and above-chance emotion recognition is achieved with both ECG and GSR signals. Comparison of the time domain features, frequency domain features, and their combination shows that FrFT are more distinct in emotion detection. The best recognition accuracy in both valence and arousal is achieved from the phase information of the FrFT coefficients using the ECG signal, which is equal to 78.32% and 76.83%, respectively.