Multiresolution framework for emotion sensing in physiological signals

This paper propose a new framework for emotion recognition and classification using a Continuous Wavelet Transform (CWT) for features extraction from physiological signal. Data from the emotional corpus recorded at Augsburg university were used in our study. In the first phase four wavelet families were chosen to analyze EMG RESP SC and ECG signals in order to extract emotional features in multi level of wavelet coefficients. The most relevant features vectors were combined to create a multimodal representation for each class of emotion. The proposed system was performed using an SVM classifier for the training and the test of the generated models, a classification accuracy of 95% was obtained and it clearly prove the performance of our framework in the estimation and characterization of emotional pattern.