As we have seen, increasing the number of directional derivatives will increase the resolution and flexibility of the inversion results. However, increasing the number of drections will also slow the inversion, because each direction has a model residual the size of the model. Instead of using many directions, it is possible to use steering filters to pre-define the local direction of maximum variance and then use that information to align the directions of the regularization to be parallel and perpendicular to it. This way, we might only need two directions in the regularization.
![[pdf]](icons/pdf.png){kind=icon}