Facial expression recognition based on dimension model of emotion with autonomously extracted sparse representations

Abstract

This paper presents a facial expression recognition system based on dimension model of internal states with autonomously extracted sparse representations. Sparse representations of facial expressions are extracted to the three steps. In the first step, Gabor wavelet representation can extract edges of face components. In the second step, sparse features of facial expressions are extracted using fuzzy C-means(FCM) clustering algorithm on neutral faces, and in the third step, are extracted using the Dynamic Linking Model(DLM) on expression images. Finally, we show the recognition of facial expressions based on the dimension model of internal states using a multi-layer perceptron. With dimension model we have improved the limitation of expression recognition based on basic emotions, and have extracted features automatically with a new approach using FCM algorithm and the dynamic linking model. © Springer-Verlag Berlin Heidelberg 2004.