摘要
由于非参数子空间分析的非参数方法(KNSA)的运算法则还是有一定的限制性:首先,该方法中的类内散步矩阵Sw还是和LDA一样,如此对识别结果就会有很大的影响.其次,该方法在计算类间散步矢量时,未考虑到不同的KNN点会产生不同的类间散射矩阵.本文提出了一种非参数特征分析的非线性(核)鉴别分析方法(KNFA),并在ORL和XM2VTS人脸库上验证了该方法在识别性能上优于KDA和KNSA方法.
The Kernel Non-Parameter subspace analysis still has some limitations. Firstly, the walking matrix in the class is sinilarte the LDA and Sw, so it will affect the results of the identification.Secondly, this method can generate different class scatter matrix among different KNN points in the computation of the walk matrix of the kind.In order to solve these problems. A Kernel Non-Parameter Feature Analysis（KNFA） method is proposed.Experimantal results on ORL and XM2 VTS face datafaces show the performance of Kernel Non-Parameter Feature Analysis is obviously better than the Kernel Non-Parameter subspace analysis and Kernel Discriminant Analysis.
出处
《广东技术师范学院学报》
2016年第2期24-28,共5页
Journal of Guangdong Polytechnic Normal University
