摘要
提出一种基于多尺度哈尔小波变换的三维人脸识别方法。首先将三维人脸模型的上半张人脸区域经过平面参数化和线性插值方法映射至几何图像,然后利用多尺度哈尔小波变换把几何图像分解为不同尺度下包含不同频率、不同方向人脸信息的频域分量,根据实验确定垂直低频分量具有表情不变性并将其线性组合作为人脸特征。在FRGC v2.0数据库中进行的三维人脸识别实验与三维人脸认证实验分别得到了98.81%的Rank_1识别率以及97.2%的正确认证率(错误接受率为0.1%)。
A 3D face recognition method based on multi-scale haar transform was proposed. The upper half face region of a 3D face model was mapped into a geometry image by mesh parameterization and linear interpolation. The geometry image was then decomposed by a multi-scale haar transform into frequency components of different scales, frequencies and orientations. The linear combination of vertical components of low frequencies was proved to be expression-invariant and was used as the facial feature. Experiments carried out in the FRGC v2.0 database achieved 98.81% Rank-1 recognition rate and 97.2% verification rate with False Accept Rate of 0.1%.
出处
《系统仿真学报》
CAS
CSCD
北大核心
2013年第7期1442-1445,共4页
Journal of System Simulation
基金
国家自然科学基金(51175081)
江苏省自然科学基金(BK2010058)
关键词
三维人脸识别
多尺度哈尔小波变换
频域分量
表情变化
3D face recognition
multi-scale haar transform
frequency components
facial expressions
