Abstract:Direct linear discriminant analysis (DLDA) is claimed that it can take advantage of all the information, both within and outside of the within-class scatter matrix's null space. In order to analyze this claim's flaw in theory, the discriminative characteristics of DLDA for face recognition was studied. Since the optimal solution of DLDA is inside the range space of the between-class scatter matrix, a theoretical analysis was unfold via the following three aspects: the relationship between the within-class and between-class scatter matrix's range space, the relationship between the within-class scatter matrix's null space and the between-class scatter matrix's range space, the characteristics of DLDA under keeping all the discriminative vectors. The results show that: in undersampled cases DLDA nearly can not make use of the information inside the null space of the within-class scatter matrix, thus some discriminative information may be lost; DLDA is degenerated as PCA of the between-class scatter matrix with all nonzero principal components if it keeps the complete discriminative vectors found. The comparative results on the face database, ORL and YALE, indicate that DLDA is inferior to the other extensions of linear discriminant analysis in terms of recognition accuracy. Which is consistent with the theoretical analysis.

The theoretical analysis of null projection directions in linear discriminant analysis

Theoretical Analysis of Direct LDA in Small Sample Size Problem

Discriminative analysis of direct linear discriminant analysis for face recognition

Computational and Theoretical Analysis of Supervised Dimensionality Reduction

A novel optimal discriminant principle in high dimensional spaces

Constrained Linear Discrimination Analysis and Face Recognition

A NEW ALGORITHM FOR COMPUTING OPTIMAL DISCRIMINANT VECTORS IN CONDITION OF HIGH-DIMENSIONAL DATA

Neighborhood linear discriminant analysis

A Note on Two-Dimensional Linear Discriminant Analysis.

L2,1-Norm Regularized Robust and Sparse Linear Discriminant Analysis via an Alternating Direction Method of Multipliers

Quadratic Discriminant Analysis by Projection

Nonparametric Locality Reforming Projection for Linear Discriminative Dimensionality Reduction

Minimum-distance discriminant projection and its application to face recognition

Null Space-Based LDA with Weighted Dual Personal Subspaces for Face Recognition.

One improvement to two-dimensional locality preserving projection method for use with face recognition

Locality-based Linear Discriminant Projection

Reverse nearest neighbors Bhattacharyya bound linear discriminant analysis for multimodal classification

Two-dimensional Nearest Neighbor Discriminant Analysis

Discriminant Projective Non-Negative Matrix Factorization

Spectral Clustering Based Null Space Linear Discriminant Analysis (Snlda)

Complete discriminant analysis