Global and local structure projection learning for image feature extraction.

The feature extraction method based on projection learning has been widely studied and achieved good results in practical application. However, most current methods often have the following limitations: (1) No simultaneous attention to global and local structural information between samples; (2) When dealing with complex noise in the data, the feature extraction performance may be affected. In view of this, a global and local structure projection learning (GALSPL) for image feature extraction is proposed in this article. First, we use the nuclear norm to reveal the global structure information from the samples, while we can explore the low-rank structure of the feature spatial data. In addition, we design a local structure constraint to extract the local feature information of the data to ensure the sparsity of the low-rank matrix. Specifically, we construct an additional matrix for data reconstruction, so that the projection matrix can more effectively learn the main features of the data. Finally, we conduct a great many comparative experiments on different datasets and verify that GALSPL has better feature extraction performance. [ABSTRACT FROM AUTHOR]