基于Candide-3模型的姿态表情人脸识别研究  

Study of pose and expression variant face recognition based on Candide-3 model

作　　者：杜杏菁[1,2] 白廷柱[1] 何玉青[1] 

机构地区：[1]北京理工大学光电学院光电成像技术与系统教育部重点实验室,北京100081 [2]华北科技学院计算机系,河北三河065201

出　　处：《计算机工程与设计》2012年第3期1017-1021,共5页Computer Engineering and Design

摘　　要：针对姿态表情严重影响人脸识别准确率的问题,基于Candide-3模型的简化,提出了形状表情关键点拟合的人脸几何结构重建和基于三角网格模型的纹理映射的方法,该方法确定关键特征点,根据人脸的几何结构信息确定姿态角,提取Candide-3模型形状表情对应点,调整模型参数,进行几何结构重建;对几何结构中每个三角网格模型进行纹理影射,得到逼真的特定人脸模型。实验结果表明,该方法提高了人脸重建速度,达到减弱姿态表情对人脸识别影响的目的。For pose and expressions seriously affecting to the accuracy of face recognition, based on the Candide-3 model, a method of pose and expression key points fitting to reconstruct face geometry and texture mapping based on Triangular mesh model are proposed. First the eyes, nose, mouth feature points are determined according to the face geometry information, then the pose angle are determined, facial expressions corresponding points are extracted with Candide-3 model , the model parame- ters are adjusted to make structural reconstruction, then texture mapping of each triangle mesh in the geometry are completed to obtain specific 3D face reconstruction model. Experiment shows that the method improve the speed of 3D face reconstruction and the effect of pose and expression variant to face recognition is reduced.

关 键 词：姿态角确定 特征点提取 人脸重建 Candide-3模型 人脸表情 

分 类 号：TP391.4[自动化与计算机技术—计算机应用技术]