Affective Computing Model Using Source-temporal Domain

This paper proposes a new Electroencephalographic (EEG) emotion recognition system (EEG-ER) that captures human emotion dynamics. EEG signals are collected from ten healthy subjects, aged 5-6 years. Four basic emotions namely; happy, sad, neutral and fear were induced from the participants using affective photographs of varying arousal from the Radbound faces database (RaFD). The affective space model proposed by was used for classifying the acquired signals into a 2-dimensional structure of valence and arousal. Feature extraction method utilized was Time Difference of Arrival (TDOA) approach for reconstructing the relative source domain of brain activity. Regularized Least Square (RLS) and Multi-Layer Perception (MLP) neural network was used for classification process. The results were compared with wavelet coefficients (WC) method and showed high accuracy around 96% for user independent classification and approximately100% for user dependent classification. Overall the results reflect significant stability of accuracy rate among subjects using the proposed method.