【摘要】 目的探索基于流形学习的人脸识别方法,将流形学习中的局部切空间排列算法(LT-SA)应用于人脸识别.方法利用样本点领域的切空间表示局部的几何性质,将局部切空间排列起来构造流形的全局坐标;用高斯核近似映射关系;在降维空间中用线性判别分析技术(LDA)提取特征;使用最近邻分类器进行分类识别;在Yale和CMU PIE人脸数据库上进行仿真实验.结果实验表明在Yale数据库上LTSA+LDA算法比已有LLE+LDA方法、LLTSA方法平均识别率分别高7.22%、19.11%;在CMU PIE数据库上分别高3.71%、29.56%.结论笔者提出的LTSA+LDA算法能较为有效地将局部切空间排列算法应用于人脸识别,显著提高了识别率.更多还原

【Abstract】 In order to explore face recognition methods based on manifold study,local tangent space alignment was proposed.This method used the tangent space in the neighborhood of a data point to represent the local geometry,and aligned those tangent spaces in the low-dimensional space to line up the global coordination of manifold and used Gaussian kernel to approximate the relationship.In the reduced space linear discriminating analysis was used to extract final feature.And the nearest neighbor classifier was used for classification.The results show that the average recognition rate of LTSA+LDA in Yale database is 7.22% and 19.11% higher than that of LLE +LDA and LLTSA respectively,and 3.71% and 29.56% higher in CMU PIE.As a result,local tangent space alignment algorithm can be applied to face recognition by using LTSA +CDA and improve recognition efficiency greatly.更多还原