Seismic data interpolation via low-rank matrix factorization in the hierarchical semi-separable representation

Recent developments in matrix rank optimization have allowed for new computational approaches in the field of seismic data interpolation. One of the main requirements of exploiting rank-minimization approaches is that the target data set should exhibit a low-rank structure. Seismic frequency slices exhibit low-rank structure at the low-frequencies, but not at the high frequencies. This behavior is due to the increase in oscillations as we move from low to high-frequency slices, even though the energy remains focused around the diagonal. Therefore, interpolation via rank minimization in the high-frequency range requires extended formulations that incorporate low-rank structure. We propose an approach for seismic data interpolation which incorporates the Hierarchical Semi-Separable Structure (HSS) inside rank-regularized least-squares formulations for the missing-trace interpolation problem. The proposed approach is suitable for large scale problems, since it avoids SVD computations and uses a low-rank factorized formulation instead. We illustrate the advantages of the new HSS approach by interpolating a seismic line from the Gulf of Suez and compare the reconstruction with conventional rank minimization.