直接无监督正交局部保持特征提取算法

基于局部保持投影发展出的一系列特征提取算法,在应用于人脸识别等高维小样本问题时,均需先采用PCA算法对高维样本降维后才能应用,故此以无监督鉴别分析算法为理论基础,提出了一种直接无监督正交局部保持算法。该算法利用拉普拉斯矩阵的性质进行相应的矩阵分解,可直接从高维样本的原始空间中提取投影矩阵,因而无需先采用PCA降维处理,且解决了无监督鉴别分析算法的小样本问题。为了进一步提高算法的识别性能,给出了基于QR分解的正交投影矩阵的求解方法。人脸库和掌纹库上的实验结果表明了所提算法的有效性。

Direct Unsupervised Orthogonal Locality Preserving Method for Feature Extraction

A series of feature extraction algorithms based on locality preserving projection were proposed. Principal Component Analysis (PCA) algorithm must be firstly used for high-dimensional samples when these algorithms are applied in such as face recognition. Therefore，using unsupervised discriminant analysis algorithm as the theoretical basis，a direct unsupervised orthogonal locality preserving algorithm is proposed. Through the corresponding matrix decomposition according to the properties of the Laplace matrix, the projection matrix can be directly extracted from the original high-dimensional space without first using PCA algorithm processing and the proposed algorithm can solve the small sample size problem. To further improve the recognition performance, the orthogonal projection matrix obtained based on QR decomposition is given. Experimental results on face database and palmprint database demonstrate the effectiveness of the proposed method.