科研之友
微信
新浪微博
Facebook
分享链接
摘要
Multi-view subspace learning has been aroused much concern recently. Although there exist a few multi-view subspace learning methods taking both the discrimination information and the correlation information into consideration, they always ignore the use of the inter-view discriminant information. In view of this, we propose an approach called multi-view local discrimination and canonical correlation analysis (MLDC(2)A) for image classification. MLDC(2)A aims to learn a common multi-view subspace from multi-view data, by making use of not only the discriminant information from both intra-view and inter-view but also the correlation information between paired view data. Furthermore, in the learned subspace, the local geometric structure of multi-view data is preserved. We conduct experiments on MNIST, COIL-20, Multi-PIE, Caltech-101, and COCO datasets and the results indicate the effectiveness of the proposed approach.
