抄録
Lip motion features such as lip width and lip length provide important information to identify individuals or commands. The interface using lip motion features is practical because operation is possible without touching devices. However, changes regarding sentiments using lip motion features have never been detected yet. Therefore, we reduce the influence of the unevenness in the lip motion features and then detect the sentiments changes. The proposed method has four steps. First, we extracted the lip motion features on the basis of both positions and shapes of lip in each frame of facial images. Second, features were created from the lip length of six utterances, which were made on the basis of these features' relative values obtained from the primary and object frame. Third, we select the data with little unevenness out of six utterances. Finally, the days when sentiments changes occurred were detected by fuzzy c-means. Our experimental results for three persons suggest that the proposed method is able to detect sentiments changes at about 73 percent.
