The regularization operator (**A**) should be designed based on
the expected model covariance Tarantola (1986). Since we are particularly
concerned with the effects of poor illumination, we need to design **A**
to compensate for these effects. In this case, since our downward
continuation operator or CAM operator
will create a model that is a cube of depth, CRPX, CRPY (for the CAM
operator), and *p*_{hx},
we can expect illumination problems to appear as gaps in events in the
CRP plane(s) and along the *p*_{hx} axis. Prucha et al. (2001)
demonstrated the use of steering filters Clapp et al. (1997) as a
regularization operator to compensate for sudden changes in amplitude
along events with an expected dip. Accomplishing this in the CRP planes
requires some interpretation of the result of migration, but along the
*p*_{hx} we expect the events to be flat and horizontal as long as the
correct velocities have been used for imaging. In this paper, to keep
**A** simple, we will just be applying the steering filters horizontally
along the *p*_{hx} axis. This regularization scheme will minimize changes
in amplitude along the *p*_{hx} axis, penalizing large amplitude changes
the most.

5/23/2004