Abstract:Linear discriminant analysis (LDA) and its variants are popular supervised dimension reduction methods, which have been widely used to handle high-dimensional data. Since most of previous LDA methods are developed based on the trace ratio (TR) criterion, they usually obtain the low-dimensional data feature with weak discriminative ability, due to the projections with small variance. The ratio sum (RS) criterion was developed to alleviate this drawback. However, the conventional ratio sum is formulated to maximize the arithmetic mean of items which suffers from the domination of the largest objectives and might deteriorate the recognition accuracy in practical applications. In this paper, a novel ratio sum minimization based linear discriminant analysis (RSM-LDA) method is proposed in this paper. Specifically, a new ratio sum minimization (RSM) criterion is developed, which is based on the properties of the harmonic mean and effectively avoids the dominance problem to discover more discriminative features of the data. However, obtaining a closed solution for the RSM-LDA problem is challenging. For this purpose, three optimization methods are used to solve the optimization problem of RSM-LDA, and the inherent relationships between the methods are discussed. Experimental results demonstrate that the proposed RSM-LDA method has better performance in classification tasks on several datasets, when compared with some comparison methods based dimensionality reduction methods. In addition, combined with other experimental results, the effectiveness of RSM-LDA was verified.

Linear Discriminant Analysis Using Rotational Invariant L1 Norm

Robust Linear Discriminant Analysis Using Ratio Minimization of L1,2-Norms

Fisher Discriminant Analysis with L1-Norm

L1-Norm Distance Linear Discriminant Analysis Based on an Effective Iterative Algorithm.

Polynomial Linear Discriminant Analysis

L1-norm Based Two-dimensional Linear Discriminant Analysis

Sparse L1-norm-based linear discriminant analysis

A New Formulation of Linear Discriminant Analysis for Robust Dimensionality Reduction

Linear Discriminant Regularized Regression

Learning Robust Discriminant Subspace Based on Joint L₂,ₚ- and L₂,ₛ-Norm Distance Metrics.

Robust L1-norm two-dimensional linear discriminant analysis.

Robust and Efficient Linear Discriminant Analysis With <italic>L</italic><sub>2,1</sub>-Norm for Feature Selection

Capped Norm Linear Discriminant Analysis and Its Applications

Robust Two-Dimensional Capped L2, 1-Norm Linear Discriminant Analysis with Regularization and Its Applications on Image Recognition.

A Novel Linear Discriminant Analysis Based on Alternate Ratio Sum Minimization

Linear discriminant analysis with spectral regularization

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

Binary linear discriminant analysis based on infinity norm

Robust Adaptive Linear Discriminant Analysis With Bidirectional Reconstruction Constraint

Neighborhood linear discriminant analysis

Towards Robust Discriminative Projections Learning via Non-Greedy <i>l</i><sub>2,1</sub>-Norm MinMax