复杂光照下的人脸姿态估计

Complex Illumination Face Pose Estimation

人脸姿态估计是人机交互和人脸识别领域的一项关键研究内容,为了提高复杂光照下姿态估计的正确率,提出一种基于层次支持向量机(SVM)的估计方法.首先通过读取三维人脸模型,添加多种光照,并进行不同角度的投影变换生成训练样本图像;然后对训练样本图像进行大小归一化,伽马矫正,高斯差分滤波等预处理并提取方向梯度直方图(HOG)特征;最后采用层次支持向量机训练人脸姿态分类器,并利用该分类器对一幅具有复杂光照的输入图像估计人脸姿态.在Color FERET数据库上的实验表明,该方法能较好的估计复杂光照下的人脸姿态.

Face pose estimation is a key research topic in the fields of human-computer interaction and face recognition. In order to improve the accuracy,this paper proposes a method of complex illumination face pose estimation based on hierarchical support vector machine（ SVM）. Firstly,the training samples are generated by reading the 3D face model,adding complex illumination and projection transformation in different angles. Secondly,this method uses gamma correction and difference of Gaussian filtering to preprocess pictures and extracts the histogram of oriented gradient（ HOG） of training samples. Finally,training hierarchical support vector machines to estimate face pose. Experiment on Color FERET database shows that the method has a better face pose estimation in complicated illumination conditions.