中文摘要：

针对人脸识别中特征提取的小样本问题，对原始的非监督判别映射（UDP）算法进行了改进，提出一种基于Log．Gabor和正交非监督判别映射（OrthogonalUDP）的人脸识别算法——LGouDP算法。此算法首先采用Log．Gabor小波对图像进行滤波来提取高阶统计信息，然后提出最大化非局部散度和局部散度的权值差和加入基向量正交约束的目标函数，对该目标函数的求解有效地避免了小样本问题，而且正交的基向量使得算法更利于保留人脸非线性子流形空间与距离有关的结构信息和重构样本。在ORL和PIE库上的人脸识别实验证明了提出算法的有效性。

英文摘要：

In view of the small sample size （SSS） problem of feature extraction in face recognition, the paper improves the version of unsupervised discriminant projection （UDP） algorithm for face recognition, and proposes the LGOUDP, a Log-Gabor based orthogonal UDP algorithm. The proposed algorithm gets the high-order statistical information by calculating the I.og-Gabor wavelet representation of face images, and then gives a new objective function that maximizes the weighted difference between the non-local scatter and the local scatter and puts in the orthogonal constraint on the basis vectors. The SSS problem algorithm can be effectively avoided by solving the new objective function. The orthogonal basis vectors help to preserve the information of nonlinear sub-manifold space which is related to distance and reconstruction data. The results of the experiments on the face databases of ORL and PIE demonstrate the effectiveness of the proposed algorithm.