In real life,the same object can be frequently observed from different view-points,resulting in a large amount of multi-view data.Although different views of same object could facilitate better data representations,the huge difference between the data from different perspectives poses a great challenge to the classification task between cross-view data.Therefore,the cross-view classification problem has great research value.In recent years,Low-rank Multi-view Subspace Learning(LMvSL)based method can effectively solve the cross-view classification problem,which learns a common projection by low-rank constraint on subspace.The matrix projects multi-view data into a common subspace,thereby obtaining perspective-independent features to solve the cross-view classification problem.Despite the existing LMvSL based methods have achieved great success,they are incapable of well handling view discrepancy and discriminancy simultaneously.To circumvent this drawback,in this paper,the Structured Low-rank Matrix Recovery(SLMR)is proposed,and the algorithm is solved by the alternating direction method of multipliers.By restoring the block diagonal low-rank matrix,SLMR obtains the ideal subspace representation coefficient,which effectively eliminates view discrepancy and improves the inter-view discriminancy.Next,Experimental results on four common used databases demonstrate the superiority of SLMR.Furthermore,when dealing with noisy multi-view data,existing LMvSL based methods assume that noise is subject to pre-assumed distribution(e.g.Gaussian distribution,Laplacian distribution).However,these models are not practical since complicated noise in practice may violate those assumptions and the distribution is generally unknown in advance.To alleviate such limitation,modal regression is elegantly incorporated into the framework of SLMR(term it MR-SLMR),and the model is solved by semi-quadratic theory.Different from previous LMvSL based methods,MR-SLMR can handle any zero-mode noise variable that contains a wide range of noise,such as Gaussian noise,random noise and outliers.Finally,the four common noises are experimentally analyzed on a public database and the results verify the robustness of modal regression model. |