基于稀疏保持拉普拉斯判别分析的特征提取算法

Feature Extraction with Sparsity Preserving Laplacian Discriminant Analysis

摘要:

针对稀疏保持投影算法在特征提取过程中无监督和l1范数优化计算量较大的问题，提出一种基于稀疏保持拉普拉斯判别分析的快速特征提取算法.首先通过逐类主元分析（PCA）构造级联字典,并基于该字典通过最小二乘法快速学习稀疏保持结构;其次利用学习到的稀疏表示结构正则化拉普拉斯判别项达到既考虑判别效率又保持稀疏表示结构的目的;所提算法最终转化为一个求解广义特征值问题.在公共人脸数据库(Yale，ORL和扩展Yale B)的测试结果验证了该方法的可行性和有效性.

Abstract:

Aiming at the unsupervised and timeconsuming l1 norm optimization problems of the existing sparsity preserving projection, a novel fast feature extraction algorithm named sparsity preserving laplacian discriminant analysis (SPLDA) is proposed. SPLDA first creates a concatenated dictionary via classwise principal component analysis(PCA) decompositions and learns the sparse representation structure of each sample under the dictionary using the least square method. Then SPLDA considers both the sparse representation structure and the discriminative efficiency by regularizing the Laplacian discriminant function from the learned sparse representation structure. Finally, the proposed method is transformed into a generalized eigenvalue problem. Extensive experiments on several popular face databases (Yale, Olivetti Research Laboratory(ORL) and Extended Yale B) are provided to validate the feasibility and effectiveness of the proposed algorithm.