基于几何特征与深度数据的三维人脸识别

3D Face Recognition Based on Geometric Features and Range Images

提出一种基于三维点云数据多特征融合的人脸识别方法。利用深度信息提取人脸中分轮廓线和鼻尖横切轮廓线;采用曲率分析的方法定位出人脸关键点,针对鼻子等人脸刚性区域,选取并计算了4类(包括曲率、距离、体积和角度)共13维的特征向量作为三维几何特征。深度图特征采用结合LBP与Fisherface的方法进行提取与识别。在3DFACE-XMU和ZJU-3DFED数据库上比较了该方法与PCA、LBP等单一方法的识别性能,识别效果有比较明显的提升。

This paper proposes a face recognition method based on fusing features from 3D face point cloud.The central vertical profile and the nasal tip transverse profile are extracted by the depth information.Calculate the curvature value of the points on the profiles and locate the feature points.For the rigid region of face such as nose,the algorithm calculates four types of geometric features,13 dimensional feature vectors in all,including curvature,distance,volume and angle.Combines Local Binary Pattern(LBP) with Fisherface method to extract the depth features.The experimental results on 3DFACE-XMU and ZJU-3DFED show that the proposed method is more effective in face recognition with compare to the single module method such as PCA and LBP.