Discriminative-CCA Promoted By EEG signals For Physiological-based Emotion Recognition

Recently, multi-modal signals such as physiological and EEG signals are increasingly utilized because of their consistency and complementarity in emotional representation. Compared with physiological signal, however, EEG signal is more difficult to obtain and more expensive so that it still cannot be utilized effectively and sufficiently. In this paper, we propose a method to enhance the performance of physiological-based emotion recognition system, where EEG signal is considered as privileged information and only available during training. Discriminative canonical correlation analysis (DCCA) is used to capture the consistency and complementarity from between physiological and EEG features, and then, a new emotional-relevant discriminative space can be constructed. During training period, physiological and EEG features are projected into an emotional discriminative space by using DCCA with the help of EEG signals. And then, machine learning techniques are utilized to build emotion recognizer for projected affective samples. During testing period, only peripheral signals are used for emotion recognition. The experimental results on two databases demonstrate that our proposed method achieves better recognition performance than existing methods.