{\it R}$_{\mbox{1}}$-PCA: rotational invariant {\it L}$_{\mbox{1}}$-norm principal component analysis for robust subspace factorization