A modified optical flow algorithm and its application in facial expression recognition

Optical flow algorithm is an important approach to estimate motion from image sequence. Using it can effectively analyze the dynamic information of facial expressions and extract the characteristic flow which can reflect the facial expression changes. In this paper, based on the Lucas-Kanade optical flow algorithm, the Hession matrix is introduced to filter the unreliable constraint points in the local neighborhood and improve the accuracy of the optical flow field in facial expression changes. Experimental results show the proposed method offers a better performance for the Lucas-Kanade optical flow algorithm.