Abstract: Emotion recognition is the key to solving the problem of the absence of emotional communication in intelligent tutoring systems．According to the problem of effective extraction of facial expression spatio-temporal features from videos for emotion recognition,a recognition method based on unsupervised feature extraction using stacked convolutional independent subspace analysis (ISA) model was proposed to recognize three emotions including puzzlement,delight and boredom that most often appear in learning．This method first detects face in video and normalizes it,then adopts stacked convolutional ISA model to learn (without supervision) facial expression spatio-temporal features from video blocks,finally uses linear SVM classifier to recognize different emotions．Experimental results indicate that this method can extract spatio-temporal expression features more effectively than the use of hand-designed features,as well as recognition rate is better,and it meets the requirement of real-time.

Key words: Emotion recognition,Unsupervised learning,Independent subspace analysis,Spatio-temporal feature,Facial expression