Joint Cross-view Heterogeneous Discriminative Subspace Learning via Low-rank Representation

The cross-view data are very unforced to capture due to the face that different viewpoints or data collected from different sensors are already very common in recent years. However, cross-view data from different views present a significant difference, that is, cross-view data from different categories but in the same view have a higher similarity than the same category but within different views. To solve this problem, we have developed a dual low-rank representation framework to unbind these interleaved structures in a learning space. In addition, we consider that each cross-view sample of the same category is from isomorphic and heterogeneous information of two interlaced structures. Hence, we propose a powerful joint cross-view heterogeneous subspace feature learning model. In addition, the subspace learned by our algorithm contains more useful information and is more adaptable to cross-view data.