LCTL solves the following optimization problem,
where X is the training data matrix, Q is the ideal representation matrix, A is a matrix that transform the learned coefficients to Q, T is the learned transform matrix and Z is the learned coefficient matrix.
The difference between [1] and our work is that, [1] actually introduces the classification error into the original transform learning.
The feature of Scene 15 dataset can be downloaded from https://pan.baidu.com/s/13fLCEhZONv_CtroZnacT1w or http://users.umiacs.umd.edu/~zhuolin/projectlcksvd.html
[1] Maggu J, Aggarwal H K, Majumdar A. Label-Consistent Transform Learning for Hyperspectral Image Classification[J]. IEEE Geoscience and Remote Sensing Letters, 2019.
[2] He-Feng Yin, Xiao-Jun Wu. Label consistent transform learning for pattern classification. Journal of Algorithms & Computational Technology, 2019, 13: 1748302619881395.