The conventional 2D thinning algorithm can be explained as follows: for a given pixel location whose value is 1, its value will be changed to 0 if the neighboring pixels have values like one of the patterns shown in equation (4). This operation continues until no pixels change. It will leave thin lines a pixel in width. The structure elements used in 2D thinning are given as follows:

(4) |

First, the 2D thinning algorithm is applied once
for each (*x*,*y*), (*y*,*z*), and (*x*,*z*)
plane separately, and the pixel is considered to be removed
only if the removal happens in at least two of the three planes.
The thinning continues until no pixels change.
Finally, this results in 3D surfaces with a width of one pixel,
representing the locations of discontinuities.
This process can be summarized in a pseudo code as follows:

old_data(x,y,z) = binary_data(x,y,z) condition = 1 do while (condition .eq. 1) { temp1(x,y,z) = 2D thinning along every (x,y) plane on old_data(x,y,z) temp2(x,y,z) = 2D thinning along every (y,z) plane on old_data(x,y,z) temp3(x,y,z) = 2D thinning along every (x,z) plane on old_data(x,y,z) new_data = [(temp1.and.temp2).or.(temp2.and.temp3).or.(temp1.and.temp3)] if(new_data .eq. old_data) { condition = 0 } else { old_data(x,y,z) = new_data(x,y,z) } }The above 3D thinning algorithm is applied to the three differently thresholded images shown in Figure 9, Figure 10, and Figure 11. The results are shown at Figure 12, Figure 13, and Figure 14, respectively. These figures show that as the thresholding value increases, more detailed shapes of the discontinuities can be found without affecting the major structure of the discontinuity surfaces. This reveals the robustness of the method in choosing the threshold values. By overlapping the discontinuity map onto the original coherency cube, we can see that the locations of discontinuities are accurately positioned, as expected (Figure 15).

Figure 12

Figure 13

Figure 14

Figure 15

I describe an automatic algorithm for discontinuity extraction from a 3D seismic image cube. The proposed algorithm consists of three steps. The first step is the coherency computation which results in a coherency cube that gives event semblance at each point. Then next step is to represent the potential discontinuity locations in binary image form. This is accomplished by thresholding the histogram-equalized coherency cube. The final step is finding location of the discontinuity by thinning the binary image obtained in the previous step. For thinning in the 3D sense, a 2D thinning algorithm is consecutively applied to produce an arbitrary shaped 3D surface.

Testing the algorithm on a real seismic image demonstrated that it can successfully find discontinuity surfaces. The extracted discontinuity surfaces could be interpreted as fault, unconformity, or buried channel. However, in order to make the proposed algorithm a more useful tool that produces geologically meaningful surfaces, further research needs to be followed. Two areas of research are: (1) The development of topological relationships between surfaces to determine whether they are connected on not. and (2) The inclusion of smoothness criteria, as geology may often dictate that some discontinuity surfaces are expected to be smooth.

10/14/2003