Abstract:Linear discriminant analysis (LDA) has been a useful tool in pattern recognition and data analysis research and practice. While linearity of class boundaries cannot always be expected, nonlinear projections through pre-trained deep neural networks have served to map complex data onto feature spaces in which linear discrimination has served well. The solution to binary LDA is obtained by eigenvalue analysis of within-class and between-class scatter matrices. It is well known that the multiclass LDA is solved by an extension to the binary LDA, a generalised eigenvalue problem, from which the largest subspace that can be extracted is of dimension one lower than the number of classes in the given problem. In this paper, we show that, apart from the first of the discriminant directions, the generalised eigenanalysis solution to multiclass LDA does neither yield orthogonal discriminant directions nor maximise discrimination of projected data along them. Surprisingly, to the best of our knowledge, this has not been noted in decades of literature on LDA. To overcome this drawback, we present a derivation with a strict theoretical support for sequentially obtaining discriminant directions that are orthogonal to previously computed ones and maximise in each step the Fisher criterion. We show distributions of projections along these axes and demonstrate that discrimination of data projected onto these discriminant directions has optimal separation, which is much higher than those from the generalised eigenvectors of the multiclass LDA. Using a wide range of benchmark tasks, we present a comprehensive empirical demonstration that on a number of pattern recognition and classification problems, the optimal discriminant subspaces obtained by the proposed method, referred to as GO-LDA (Generalised Optimal LDA), can offer superior accuracy.

General Averaged Divergence Analysis

Geometric mean for subspace selection.

Harmonic Mean for Subspace Selection

Global subclass discriminant analysis

Data-Dependent Kernel Discriminant Analysis for Feature Extraction and Classification

GO-LDA: Generalised Optimal Linear Discriminant Analysis

Subspace Clustering by Directly Solving Discriminative K-means

Bregman Divergence-Based Regularization for Transfer Subspace Learning

Local Feature Discriminant Projection

Measuring Generalized Divergence for Multiple Distributions with Application to Deep Clustering

Reconstructable generalized maximum scatter difference discriminant analysis

Making Fisher Discriminant Analysis Scalable.

Learning Log-Determinant Divergences for Positive Definite Matrices

Linear discriminant analysis based on genetic algorithm

3E-LDA

A Generalized Form of Fisher Linear Discriminant Function

Robust sparse manifold discriminant analysis

Linear Discriminant Analysis Using Rotational Invariant L1 Norm

A generalization of regularized dual averaging and its dynamics

Generalized two-dimensional linear discriminant analysis with regularization

Generalized Arithmetic and Geometric Mean Divergence Measure and their Statistical Aspects