An interpreter desires a detailed image of the subsurface's features of interest, such as river channels or faults. Simultaneously, the interpreter may want to remove other features, such as a plane layer contribution.
A PE filter removes the predictable parts of an input image and creates a white output. Plane layer contributions are highly redundant and predictable and therefore likely to be removed by a PE filter. Unfortunately, even faults and river channels are predictable since they are continuous surfaces that locally are approximated by planes. I attempted to implement a PE filter scheme that is able to remove the plane layer volume, but does not remove planar discontinuities such as faults and river channel flanks.
In a more theoretical accompanying report Schwab (1997), I describe a set of three 2-D filters that are designed to remove plane layer volumes without removing planar discontinuities such as fault reflections. Additionally, the article suggests how to unite the individual three images of the three filters to a single final image volume.
The filter results in earlier experiments Schwab et al. (1996) were disappointing since, despite their design, the filters removed all planar features that I attempted to isolate. In seismic sections, faults, flanks of river channels or salt domes are rarely pure planar features. Their limited extension out of a plane are sufficient for the PE filters to predict and remove the target events. Consequently, I decided to restrict the filter even more.
I know three methods to restrict a filter's ability to remove predictable input components:
In my earlier coherency paper 1996, I experimented with filter gaps and filter shape (second approach). Since these approaches failed to restrict the filter sufficiently, I attempt in this paper to pre-whiten the filter input (third approach).