Exploiting class-wise coding coefficients: Learning a discriminative dictionary for pattern classification

Jianqiang Song, Xuemei Xie, Guangming Shi, Weisheng Dong

Abstract: Highlights•A novel DDL method, named CW-DDL, is proposed to learn a discriminative dictionary for classification by exploiting class-wise coding coefficients.•A label-aware constraint is first presented to make the coefficient matrix has class-wise approximate sparse structure, then it is integrated with the graph regularization.•The above two terms are reinforced each other in the learning process, resulting in a very robust and discriminative dictionary.•To further promote the discrimination ability of coding coefficients, a support vector based classifier is employed on them.•The obtained coding coefficients have two class-wise characteristics: the class-wise sparseness and class-wise separation.