基于注释人脸模型的三维面部数据局部匹配

传统人脸识别在非正常光照和姿势的情况下鲁棒性较低。针对非正常角度下人脸数据的局部缺失导致识别率降低的问题,提出了一种统一的解决局部匹配问题的方法。通过引入基于曲率和径向距离的3D标志指示器,进行3D人脸标志检测,采用注释人脸模型(annotation facial model,AFM)进行初始配准,扩展可变模型框架即利用在拟合过程中人脸对称性的优点,使得局部匹配问题可以将缺失数据完整化,从而真正地实现了姿态的不变性。最后,在通用的3D人脸数据库FRGCv2及来自圣母大学的人耳数据库的集合F、G上进行实验,验证了所提方法的有效性。

Partial Matching Based on Annotation Facial Model for Three-Dimensional Facial Data

Traditional face recognition has low robustness under the unnormal illustration and pose condition, for the issue that partial losing of face data causes low recognition accuracy in the unnormal situation , a union approach to addressing partial matching problem is proposed in this paper, it uses 3D label indicator based on Curvature and the radial distance to detect 3D face label. Adoption of Annotation Facial Model (AFM) on initial registration and expanding the variable model framework namely using the advantages of facial symmetry in fitting process makes data losing can be completed by partial matching so that the invariance of gesture can be implemented truly. Finally, the efficiency of proposed method is verified on popular 3D face database FRGC v2 and sets F and G of the human ear database from University of Notre Dame.