Emotion Recognition Using Time-Frequency Distribution and GLCM Features from EEG Signals

Deep learning techniques are commonly used for emotion recognition from Electroencephalography (EEG) signals. However, some disadvantages of employing these classifiers are the high memory requirements and the low number of available EEG samples in datasets. This work proposes a novel approach for increasing the number of extracted features based on the Gray Level Co-occurrence Matrices (GLCMs) technique using reassigned spectrogram images. EEG signals are transformed using spectral analysis to construct the reassigned spectrogram images. Different feature sets are employed to train multiple classification models based on the leave-one-out method. K-Nearest Neighbor technique achieves the highest accuracy results, 77.40% and 77.30% for valence and arousal primitive emotion classification. Comparative results show that the proposed approach is competitive to those existing in the state-of-the-art.