Locality Preserving Fisher Discriminant Analysis with Clustering
Lishan Zou, Yuechao Wang, Zhenzhou Chen, Xiaorong Wu · 2012
Fisher discriminant analysis (FDA) is an important feature extraction method for many classifiers.However, it tends to give undesired results if samples in some classes form several separate clusters, i.e., multimodal.This paper proposed a new feature extraction method called locality preserving Fisher discriminant analysis with clustering (LPFDA) for multimodal data.First new classes are formed by clustering data according to labels, then the between-subclass scatter matrix and within-subclass scatter matrix are computed by new classes, finally the vectors are choose which will maximize the Fisher criterion function as the discriminant vector .When our method is applied to the recognition problems of digits and images, and the experimental results show the better performance than the original one.