Multi-view subspace clustering for learning joint representation via low-rank sparse representation

Multi-view data are generally collected from distinct sources or domains characterized by consistent and specific properties. However, most existing multi-view subspace clustering approaches solely encode the self-representation structure through consistent representation or a set of specific representations, leaving the knowledge of the individual view unexploited and resulting in bad performance in self-representation structure. To address this issue, we propose a novel subspace clustering strategy in which the self-representation structure is contemplated through consistent and specific representations. Specifically, we apply the low-rank sparse representation scenario to uncover the global shared representation structure among all the views and deploy the nearest neighboring method to preserve the geometrical structure according to the consistent and specific representation. The L-1-norm and frobenius norm are applied to the consistent and specific representation to promote a sparser solution and guarantee a grouping effect. Besides, a novel objective function is figured out, which goes under the optimization process through the alternating direction technique to evaluate the optimal solution. Finally, experiments conducted on several benchmark datasets show the effectiveness of the proposed method over several state-of-the-art algorithms.