Multi-class Fukunaga Koontz discriminant analysis for enhanced face recognition

作者：

Highlights：

• Solve small-sample-size problem in LDA, UDP, LPP using FKT formulation.

• Can work with high dimensional data without inverting any scatter matrices.

• Finds optimal projection direction vectors that are orthogonal.

• Finds exact solutions to the objective in the form of trace ratio.

• Improvement in unconstrained face recognition scenarios.

摘要

Highlights•Solve small-sample-size problem in LDA, UDP, LPP using FKT formulation.•Can work with high dimensional data without inverting any scatter matrices.•Finds optimal projection direction vectors that are orthogonal.•Finds exact solutions to the objective in the form of trace ratio.•Improvement in unconstrained face recognition scenarios.

论文关键词：Linear discriminant analysis,Unsupervised discriminant projection,Locality preserving projections,Fukunaga Koontz transform,Face recognition

论文评审过程：Received 9 September 2013, Revised 21 July 2015, Accepted 9 October 2015, Available online 17 October 2015, Version of Record 24 December 2015.

论文官网地址：https://doi.org/10.1016/j.patcog.2015.10.007