Skip to Main Content
3D face models provide more robust shape information of facial features than intensity or color in 2D images. However, many current facial feature extraction methods on 3D face models still depend on human instruction. This paper proposes an automatic human face feature extraction method adaptive to 3D face models in various poses and various scales based on analysis of surface curvature and a priori knowledge of human face structure. Moreover, during processing of segmented regions on 3D model, a novel region processing approach, called "combine and split", is proposed to significantly reduce undependable candidate regions for facial organs from hundreds to around ten. Experimental results demonstrate that proposed method can effectively extract eye, nose, mouth and ear regions from various 3D face models.
Date of Conference: 13-16 Dec. 2005