情報処理学会 第75回全国大会講演要旨

In the animation of facial expression with point clouds from Range Images, looking for corresponding points in different PCs is a critical problem for local face motion while global face motion can be decided by ICP algorithm. In local face motion, to find the relationship between partial movements of two different faces, we propose a method of feature distraction based on implicit geometry information of facial point clouds. Assuming that the point in PC is fitted by parabolic surface locally, E, F and G for First Fundamental Form of surface and L, M and N for Second Fundamental Form of surface can be calculated. From the figures of this implicit geometry information, the face may be divided into different parts in the case that feature lines and points can be distracted from PCs, PCs can be segmented into different parts which will enhance the precision and speed of searching process of corresponding points in different PCs by comparison with the method of considering geometry distances between points merely.