基于随机子空间-正交局部保持投影的支持向量机

摘要

In order to deal with the classification problem for high-dimensional and small-sized data, a kind of support vector machine based on random subspace and orthogonal locality preserving projection was proposed. The random subspace method was used to select a feature subset from the original feature space randomly for several times. Based on the selected feature subset, several base support vector machine (SVM) classifiers were generated.The orthogonal locality preserving projection method was adopted to carry out feature extraction on the samples of each base classifiers, which can, effectively, realize dimensionality reduction. We applied the processed samples to train each base classifiers. The results of the base SVM classifiers were integrated to obtain the final classification result,using a bayesian sum rule.Results on two publicly available face databases show the feasibility and validity of our proposed method.%针对高维数、小样本数据分类问题,提出一种基于随机子空间-正交局部保持投影的支持向量机.利用随机子空间方法对原始高维样本的特征空间进行多次随机采样,生成多个具有不同特征子集的基支持向量机(SVM)分类器；利用正交局部保持投影对各基SVM分类器的样本进行特征提取,实现维数约简；然后,利用降维后的样本对各基SVM分类器进行训练；采用贝叶斯求和准则将各基SVM的分类结果进行融合以得到最终的分类结果.典型人脸数据库识别结果验证了本方法的可行性和有效性.