公共矢量的最小类内方差SVM与噪音人脸分类

Minimum class variance support vector machines based on common vectors for noisy face classification

提出基于公共矢量的最小类内方差支持向量机(CV-MCVSVM),用于提高噪音人脸图像分类问题中的抗噪性能。它继承了最小类内方差支持向量机(MCVSVMs)的优点,引入了由公共矢量(CVs)构成的散度矩阵Scom,由于CVs包含了样本中的共同信息,因此CV-MCVSVM在定义中将每个样本减去了CVs的均值,保留了更多的分类信息,进一步提高了抗噪能力。给出了CV-MCVSVM的推导过程。经实验验证,在含有噪音人脸图像的分类问题中,CV-MCVSVM获得了比MCVSVMs和总间隔v-支持向量机(TM-v-SVM)更好的分类性能。

In this paper,the Minimum Class Variance Support Vector Machines Based on Common Vectors（CV-MCVSVM） as the improved version of Minimum Class Variance Support Vector Machines（MCVSVM） is presented for noisy face recognition,which inherites the advantages of MCVSVM.S com which constituted by the Common Vectors（CVs） is introduced.CVs contain the common information in the samples,so CV-MCVSVM utilizes the mean value of CVs to retain more information on the classification and improve the performance of noisy face classification.The experimental results about noisy face classification demonstrate that the proposed CV-MCVSVM has better classification performance than both MCVSVMs and TM-v-SVM.