2-D and 3-D segmentation

Threshhold passing of wavelet components leads to a surprisingly effective event discrimination algorithm. Figure 4 displays the results of a global threshhold, about 50%. This segmentation works for dipping events as well as axis-parallel events.

 

Figure 4: Top is reconstruction of the 2-D seismic signal when 50% strongest amplitudes passed (left) or 10% strongest erased (right). Below are a central California 3-D dataset with 50% strongest amplitudes passed. Left and right are opaque and transparent views of the same data volume. (Generated with Sun Microsystem's SunVision software.)

It is hypothesized that non-global threshhold discriminants will do even better. One approach is to choose threshholds based on wavelength. Another approach is to track extrema through wavelengths, similar to the discriminant of Witkin (1983).