Abstract: A nose tip detection method based on the corner detection of face silhouette is proposed, in order to realize the rapid and accurate location of the nose tip when the face has an expression and attitude changes. Firstly, the cylinder head model is used to correct the face attitude roughly. Then, the rotation projection method is applied to extract the profile signatures, and the candidate nose tip regions are located according to the corner of face silhouette on B-spline scale space. Finally, the position of the nose point is accurately positioned according to the shape feature and the protruding characteristic of the nose point. The results of the experiment on CAISA 3D and BOSPHORUS demonstrate that the proposed method has great robustness to expression and attitude, and the positioning accuracy is superior to the methods based on the prior information and the statistical template.

Key words: feature extraction, face recognition, corner detection, curve fitting

摘要： 为实现人脸表情及姿态变化下，鼻尖点的快速准确定位，提出一种基于人脸侧影线角点检测的鼻尖点定位方法。首先利用柱状人头模型，进行人脸姿态粗矫正；然后通过旋转投影法提取人脸的侧影轮廓线，并基于B样条尺度空间检测侧影线角点，根据角点位置定位鼻尖点候选区域；最后根据鼻尖点的形状特征及凸出特性准确定位鼻尖点位置。在CASIA 3D和BOSPHORUS三维人脸数据库的实验结果表明，该方法对表情和姿态鲁棒性较好，且定位精度优于基于先验信息和基于统计模板的方法。

关键词: 特征提取, 人脸识别, 角点检测, 曲线拟合