论文部分内容阅读
基于核策略的核Fisher鉴别分析(KFD)算法已成为非线性特征抽取的最有效方法之一。但是先前的基于核Fisher鉴别分析算法的特征抽取过程都是基于2值分类问题而言的。如何从重叠(离群)样本中抽取有效的分类特征没有得到有效的解决。本文在结合模糊集理论的基础上,利用模糊隶属度函数的概念,在特征提取过程中融入了样本的分布信息,提出了一种新的核Fisher鉴别分析方法———模糊核鉴别分析算法。在ORL人脸数据库上的实验结果验证了该算法的有效性。
Nuclear Fisher’s Kernel-based discriminant analysis (KFD) algorithm has become one of the most effective methods for nonlinear feature extraction. However, the previous feature extraction based on the kernel Fisher discriminant analysis algorithm is based on the binary classification problem. How to extract effective classification features from overlapping (outlier) samples has not been effectively solved. Based on the fuzzy set theory and the concept of fuzzy membership function, this paper integrates the sample distribution information in the feature extraction process and proposes a new kernel Fisher discriminant analysis method - fuzzy kernel discriminant analysis algorithm. The experimental results on the ORL face database verify the effectiveness of the algorithm.