Dual Principal Component Pursuit

Abstract

We consider the problem of outlier rejection in single subspace learning. Classical approaches work directly with a low-dimensional representation of the subspace. Our approach works with a dual representation of the subspace and hence aims to find its orthogonal complement. We pose this problem as an l1-minimization problem on the sphere and show that, under certain conditions on the distribution of the data, any global minimizer of this non-convex problem gives a vector orthogonal to the subspace. Moreover, we show that such a vector can still be found by relaxing the non-convex problem with a sequence of linear programs. Experiments on synthetic and real data show that the proposed approach, which we call Dual Principal Component Pursuit (DPCP), outperforms state-of-the art methods, especially in the case of high-dimensional subspaces.