中文摘要：

针对单一的人脸特征在识别中的局限性,将基于深度图像的全局特征和基于测地线的局部特征进行融合,以提高识别率.将三维人脸点云转换为深度图像后进行预处理,然后使用主成分分析法（PCA）找到一个低维的特征脸空间,依照最近邻法则将其与库集样本进行匹配,所得结果即为全局特征;将测试样本与模板人脸进行匹配,得到35个特征点,这些特征点间的测地线距离所组成的矩阵即为局部特征.使用加权求和法对这2种特征进行融合,并根据最近邻匹配法则,在FRGC人脸数据库上进行测试.实验结果表明,该方法可以很好地结合全局特征和局部特征的互补信息,识别效果优于各单一特征的分类性能,并且具有较好的表情鲁棒性.

英文摘要：

In order to overcome the limitation of sole facial feature in recognition,a novel approach is presented to improve the accuracy by fusing the global feature based on the range image and the local facial feature based on the geodesic curve.The global feature is extracted using the following steps.First,cloud data is converted to range image.After preprocessing and normalized steps are applied,the principal component analysis（PCA） is used to find a low dimensional feature face space and achieve match score in accordance with the nearest neighbor rule.The local feature is a geodesic distance matrix among 35 points on test face,which is ordered by template after registration.These two kinds of features are fused into a combined one over their respective matching score weighted by a set of coefficients,which are determined by the maximum identification rate of training sets.The method is evaluated on FRGC（face recognition grand challenge） database based on the nearest neighbor（NN） classifier.The experimental results show that the complementary information can be combined in this modified scheme more sufficiently and the fused feature outperforms single module in face recognition.Furthermore,it is also robust to expression.