Random consensus robust PCA (Q1688993)

The robust principal component analysis (RPCA) represents a problem of principal component analysis where every column in the data matrix \(M\) may have grossly corrupted entries. In this manuscript the authors introduce an algorithm for RPCA, the so-called random consensus algorithm for RPCA (R2PCA). The aim of this algorithm is to identify canonical projections which are used to recover subspace spanned by the columns of a low-rank matrix \(L\) of the matrix \(M\), where \(M=L+S\) and \(S\) is sparse. The algorithm is presented in its most basic setting, and a possible generalization to noisy settings is considered. At the end of the paper some extensive experiments are presented and the introduced model is compared with state-of-the-art methods.