一种新的正交保局投影人脸识别方法

A new Alternative formulation of orthogonal LPP with Application to Face Recognition

针对人脸识别中判别特征的提取问题,提出了一种新的人脸识别算法—Schur正交保局投影(Schur-OLPP)。该方法在保局投影(LPP)的基础上引入Schur分解,求取最佳正交投影矩阵,充分提取样本的判别特征。本文采用最小近邻分类器估算识别率。在Yale人脸库以及AR人脸库的测试结果表明,在姿态、光照、表情、时间变化的情况下,Schur-OLPP都具有较好的识别率。

Feature extraction is an important area of face recognition. A new face image feature extraction and recognition method-Schur Orthogonal Locality Preserving Projections （Schur-OLPP） is proposed in this paper. Schur-OLPP introduces Schur decomposition in Locality Preserving Projections （LPP） to get the orthogonal vectors and extracts discriminant features. The proposed method was tested and evaluated using the Yale face database and AR face database. Nearest neighborhood （NN） algorithm was used to construct classifiers. The experimental results show that Schur-OLPP has good performance when pose, illumination condition, face expression and time change.