Adaptive curvelet-domain primary-multiple separation |
We presented a method that improves estimates for the primaries for situations where multi-term SRME is unviable. Our alternative augments Bayesian primary-multiple separation with a data-adaptive step, during which the amplitudes of the predicted multiples are matched to the multiples in the data. This match is achieved in the curvelet domain, which allows for a position, scale and dip-dependent amplitude correction through a diagonal scaling of the transform coefficients. Overfitting, i.e., distortion of the primaries, during the matching is avoided by promoting smoothness amongst neighboring coefficients in the scaling vector. Application of our method to synthetic and real-data sets shows a clear improvement in multiple suppression and primary preservation, which can be attributed to the curvelet-domain amplitude correction by scaling. Since our correction is based on a relatively mild smoothness assumption, stating that the amplitude errors can not vary too rapidly as a function of position, scale and angle, we envisage applications in other areas, such as the suppression of internal multiples, where angle-dependent reflection and transmission errors play a role.
Adaptive curvelet-domain primary-multiple separation |