Graph-Based Sparse Matrix Regression for 2D Feature Selection

Abstract

It is common to perform feature selection for pattern recognition and image processing. However, most of conventional methods often convert the image matrix into a vector for feature selection, which fails to consider the spatial location of image. To address this issue, we propose a graph-based sparse matrix regression for feature selection on matrix. We incorporate a graph regularization term into the objective function of the sparse matrix regression model. The role of this graph structure is to make the matrix samples sharing the same labels keep close together in the transformed space. Extensive experimental results can demenstrate the effectiveness of our proposed method.