| It is very important that achieve and recognize emotion signals which as a basis for an emotional communication between computer and human.This paper focuses on study of achieving and recognizing these two points. Access to emotional signals mainly through making external stimulate emotional of subjects. Affective recognition include limb emotion recognition, facial emotion recognition, speech recognition, physiological signal recognition etc. One of the crucial area of emotion recognition is physiological signal recognition, physiological signals are not susceptible to be controlled by subjective conscious, so that it can more realisticly and objectively response to person’s emotional state and changes. Physiological signals will be employed into emotional recognition then emotional recognition model will be established, through this model, computer will have the abilities of recognizing and generating emotional features as human. Finally the man-machine communication will be come true.Previous research found that it is feasible to adopt electromyography(EMG) signals, electrocardiogram(ECG) signal, pulse(PVP) signal, respiratory signal(RSP) and skin conductance(GSR) signal in the physiological signals. EMG signal reflects the states of nerve and muscle, and shows the mood changes in some respects. There is great significance to apply EMG signals to affect recognition. This study is on the affect recognition based on electromyography(EMG) signal for joy、grief、surprise、angry、fear and disgust.The following work is discussed in this paper:(1) Video clips are used as materials to evoke six emotions (surprise, joy, disgust, grief, fear and anger) in the experiment. EMG signals are collected by the multi-channel physiological recorder MP150. In order to extract valid data, at the end of each movie clip, the subject is asked to fill in a questionnaire which reflects the feelings of watching the video.(2) Then wavelet transform method was used for de-noised effective EMG signals, after noise reduction, statistical features in time-domain were obtained, Daubechies5 wavelet with orthogonality and compact support was adopted as basic function to do 5-layer decomposition of EMG signal after noise, then 21 statistical features of each layer’s detail coefficients were extracted. So we can obtain 126 original features in total.(3) However, not all features make contributes to emotion recognition, so it is necessary to find affective features from them, namely feature selection. Feature selection in emotion recognition is a combinatorial optimization problem thus a NP problem. So an effective intelligent optimization algorithm is advisable to find a satisfying solution to the problem. The phase algorithm is adopted in this paper. At first,the original features with high correlation were deleted to reduce dimensionality of original feature set by correlation analysis. And then,an improved Tabu Search algorithm was proposed to achieve affective feature selection in the feature space with reduced dimension. Then improved Tabu Search algorithm is adopted to find feature subsets which best represent the affective responses of EMG signal, and six emotions(joy, surprise, disgust, grief, anger and fear) are recognized by means of Fisher classifier.(4) The new data is adopted to verify the effectiveness of the feature subset, calculate the recognition rate verification, verify the promotion of affective computing model.Ultimately, the experimental results show that the stimulation result shows that obtaining feature fusion related to emotions from EMG signals is totally doable. While selecting features, after the original features with high correlation were deleted to reduce dimensionality of original feature set by correlation analysis, we used improved Tabu Search algorithm and fisher classifer, therefore, feature-dimension is reduced, good recognition effect is obtained, and the problem of TS algorithm’s dependency on the initial solution and field structure is also solved. |
PDF Full Text Request