Abstract: Accurate location of face feature points is one of the key steps in 3D face recognition. A fully automatic locating algorithm for 3D facial nose tip and inner eye corner in different views was presented, which relied on a statistical model combining Local Shape Map (LSM) and local shape index, and the statistical radius of the local shape map was adaptively determined by error analysis. The experimental results on the CASIA 3D face database demonstrate that this method performs more precise location than heuristic algorithm and curvature-based method.

Key words: feature point location, shape index, curvedness, Local Shape Map (LSM), Support Vector Machine (SVM)

摘要： 准确定位人脸特征控制点是三维人脸识别的关键技术之一。提出了一种新的三维人脸特征点自动定位方法,结合局部形状索引与基于局部形状图(LSM)的统计模型，通过误差分析自适应地确定局部形状图的统计半径,实现任意姿态下的三维人脸鼻尖和内眼角的自动精确定位。在CASIA 3D人脸数据库的比较实验结果表明，该方法比基于先验信息和基于曲率分析的定位方法都具有更高的定位精确度。

关键词: 特征点定位, 形状索引, 曲度, 局部形状图, 支持向量机