Sparse approximation to discriminant projection learning and application to image classification

作者：

Highlights：

• Developing a new estimation method of projection matrix to avoid the matrix singularity problem.

• Joint using of F-norm and L2,1−norm to construct a feature selection framework, which is effective to select informative features.

• Proposing a supervised sparse discriminant projection learning algorithm, which preforms subspace learning and feature selection simultaneously.

• Proposing an effective optimization algorithm to solve the derived objective function, which can be theoretically proved for the convergence.

摘要

•Developing a new estimation method of projection matrix to avoid the matrix singularity problem.•Joint using of F-norm and L2,1−norm to construct a feature selection framework, which is effective to select informative features.•Proposing a supervised sparse discriminant projection learning algorithm, which preforms subspace learning and feature selection simultaneously.•Proposing an effective optimization algorithm to solve the derived objective function, which can be theoretically proved for the convergence.

论文关键词：Image classification,Feature selection,Subspace learning,Discriminant analysis,Dimensionality reduction

论文评审过程：Received 1 December 2018, Accepted 10 July 2019, Available online 12 July 2019, Version of Record 17 July 2019.

论文官网地址：https://doi.org/10.1016/j.patcog.2019.106963