Enhanced iterative projection for subclass discriminant analysis under EM-alike framework

作者：

Highlights：

• We focus on the classification of the large sample size problems.

• We propose a new EM-alike iterative projection approach (EMIPA) for subclass division.

• Our proposed approach outperforms the traditional MDA and SDA.

• Our approach operates subclass division and eigenvector seeking class by class.

• Our proposed approach takes only a bit more time than MDA and SDA.

摘要

Highlights•We focus on the classification of the large sample size problems.•We propose a new EM-alike iterative projection approach (EMIPA) for subclass division.•Our proposed approach outperforms the traditional MDA and SDA.•Our approach operates subclass division and eigenvector seeking class by class.•Our proposed approach takes only a bit more time than MDA and SDA.

论文关键词：Linear discriminant analysis,Larger sample size problems,Mixture discriminant analysis,Subclass discriminant analysis,EM-alike framework,Iterative steps,Generalized eigenvalue problem,K-means clustering

论文评审过程：Available online 6 August 2013.

论文官网地址：https://doi.org/10.1016/j.patcog.2013.07.001