Noise attenuation via robust low rank matrix factorization to singular Spectrum Analysis

Summary The Singular Spectrum Analysis (SSA) method is an efficient tool for seismic data white noise attenuation. When data contains outlier noise this subspace based technique faces problems due to its L2 norm optimization that is very sensitive to outliers. In this paper, we solve the SSA problem with a robust low rank matrix factorization algorithm that also uses the L2 norm but with constraints on both the rank of the matrix and the number of outliers. The basic idea of this method is similar to the recently developed nuclear norm minimization that constrains outliers, but our formulation and implementation are simpler. In addition, any a priori information regarding outliers can be easily incorporated to make the performance of this method more effective. We applied this method to synthetic data and real data to show the effect of this algorithm.