摘要
通过一张二维照片计算其三维姿态是三维人脸识别和颅像重合中的难点问题。该文主要研究了绕轴的平面旋转、垂直深度旋转和侧深度旋转的姿态计算问题。通过两外眼点间的关系计算平面旋转角度;通过眉间点、鼻下点与颏下点的距离关系计算垂直深度旋转角度;通过鼻下点与两外眼点的角度关系计算侧深度旋转角度。试验证明提出的方法误差小,能够比较准确地计算出三维姿态。
It is a nut of face recognition and photographic superimposition to compute the 3D pose through only one photo. This paper presents the ideas for lane-rotation, vertical-rotation and side-rotation. Lane-rotation is got by the two outer-eye points, vertical-rotation is obtained by middle-eyebrow points, lower-nose point and chin point, side-rotation is gained by lower-nose and two outer-eye points. Experimental results show that presented methods perform well. Recognition ratio can be raised effectively.
出处
《计算机工程》
EI
CAS
CSCD
北大核心
2006年第6期232-233,共2页
Computer Engineering
基金
国家自然科学基金资助项目(60271032)
关键词
人脸识别
颅像重合
三维姿态
Face recognition
Photographic superimposition
3D pose
