We test the above-described adaptive separation algorithm by examining
synthetic and real data. The main purpose of these tests is to study
the improvement by curvelet-domain matching compared to optimized
results for single-iteration SRME. This case is relevant for
situations where the data quality does not permit iterative SRME or
where the cost of multiple iterations of SRME is a concern. In either
situation, the predicted multiples will contain amplitude errors,
which may give rise to residual multiple energy and dimmed
primaries. We show that the proposed scaling by curvelet-domain
matched filtering improves the estimation for the primaries as long
the curvelet-to-curvelet variations for this scaling are sufficiently
controlled by the smoothness constraint. Relaxation of this constraint
may leads to overfitting and hence to inadvertent removal of primary
energy.
![[pdf]](icons/pdf.png){kind=icon}