next up previous [pdf]

Next: Transformation of input data Up: Approach Previous: Approach

Data input and manipulation

After modifying the original algorithm to work with seismic data rather than integer RGB values, seismic images may be segmented according to the same rules used by Felzenszwalb and Huttenlocher (2004) to segment RGB images. The results can be seen in Figure 3(a) for the synthetic seismic image, and Figure 3(b) for the field data image. In these figures, each segment or region is assigned a random color, and the segments are overlain on the seismic image itself for reference. The results from the (relatively) unaltered algorithm are promising, but require improvement. In Figure 3(a), the salt body is fairly well-resolved. However, it appears that the smaller canyon along the left of the salt body has been improperly included, along with another portion adjacent to the salt body at the far right of the image. The segmentation of the field data example in Figure 3(b) is especially poor; however, it is interesting to note that portions of the salt boundary itself have been recognized as segments. This may be related to the nature of the input data, an issue treated in the next section.


next up previous [pdf]

Next: Transformation of input data Up: Approach Previous: Approach

2010-05-19