Abstract
Lip-reading is to recognize what someone is saying from motion of lip. Since lip-reading is able to recognize one's words without voice, it is robust against noisy environment. To achieve accurate recognition results, accurate shape of lip contour should be detected. In this paper, a new algorithm to detect accurate contour of lip is proposed. Since, Sampled Active Contour Model(S-ACM) reacts only with edge information, the discontinuous edge of lip causes shape errors. The proposed algorithm generates accurate shape of lip, even though there is discontinuity in the edge, by combining color and brightness information with conventional S-ACM. In addition, the proposed algorithm is able to cope with misdetection of lip contour by using gradient information of points on the contour.
