In this paper, a new technique based on motion imformation of dynamic images is proposed for expression recognition. The intensity differences of corresponding pixels are calculated from a peak and an initial frame of image sequences, and then expressions are classified based on statistical features extracted from the intensity difference images. With the new technique, most expression imformation is reserved while the influence of illumination and complexion is eliminated. The contrast trials of person-independent expression recognition are performed on the database of CED-WYU (1.0) and COHN-KANADE, and higher recognition rates are obtained when the new method is used.
展开▼