부산대학교 도서관

Active Appearance Model has been successfully used in many applications. However, a single AAM is prone to fail in fitting and synthesis when it trained with various poses. In this paper, we propose to build a model covering only a narrow range of angles for each pose. We also provide an inter-model synthesis by training the relationships between the models. We tested the proposed method on the Yale B face database. Experimental results show that the proposed method outperforms the original AAM with respect to stable fitting and the synthesis results.