Automatic fault interpretation with optimal surface voting

Numerous types of fault attributes have been proposed to detect faults by measuring reflection continuities or discontinuities. However, these attributes can be sensitive to other seismic discontinuities, such as noise and stratigraphic features. In addition, fault features within a fault attribute image often cannot be continuously tracked. We have developed an optimal surface-voting method to enhance a fault attribute image so that the noisy features (unrelated to faults) are suppressed whereas the fault features become cleaner and more continuous. In this method, we first automatically pick seed points from the input attribute image and use these seeds as control points to compute optimal surface patches that pass through the control points and follow globally maximum fault attribute values. Then, we consider all the computed surfaces as voters and define voting scores for each voter by using fault attribute values that are smoothed along the surface voter. We further collect voting scores of all the voters to compute a voting score map as a new fault attribute image, in which fault features (with high scores) are much cleaner, sharper, and more continuous than those in the input attribute image. With the optimal surface voters, we can also accurately estimate fault orientations (strikes and dips) by computing weighted averages of the surface voter orientations. From a voting score map with clean and continuous fault features, fault surfaces can be extracted by tacking the fault features along the estimated fault orientations. The computational cost of the method depends on the number of seed points, not the size of the seismic volume, which makes the method highly efficient. With an four-core computer, our parallel implementation can process more than 1000 seeds in 1 s to compute the corresponding optimal voting surfaces and a final voting score map.