A Novel Low-rank and Sparse Decomposition Algorithm Based on Laplacian Distribution

Abstract

The principal component pursuit (PCP) method has an excellent performance in foreground/background separation, but this method is also acknowledged to have some drawbacks: 1) the poor robustness; 2) the choice of balancing parameters is a tricky matter. To address these problems, we propose a new low-rank and sparse decomposition model based on the nuclear norm and Laplacian scale mixture. This model uses the Laplacian scale mixture to approximate the sparse term to improve the robustness of PCP and reduce the difficulty of adjusting parameters. Experimental results show that our approach is more effective than the PCP algorithm.