Enhanced interpreter-aided salt-boundary extraction using shape deformation

Introduction

Fig1
Figure 1. (a) A typical seismic image showing the salt body in the center. Some parts of the boundary are not well imaged due to the limited image quality. (b) The same image shown with the human-interpreted boundary.

Interpreting the salt body in seismic images is of great importance for accurate velocity model building. Due to the poor quality of many seismic images, simple image processing filters followed by local boundary tracking algorithms give very poor results or fail easily. For example, in Figure 1, some parts of the salt boundary are simply missing from the image. A human interpreter's input is essential in these troublesome regions (Halpert et al., 2011). In the near term, the development of a super-algorithm that can fully automate boundary extraction without sacrificing the quality of the results is highly unlikely.

However, as today's seismic imaging practices evolve to three dimensions, manual interpretation of every single slice in a 3-D image cube is increasingly unrealistic. How can we achieve a good trade-off between the amount of manpower required and the quality of the boundary extraction result? The idea of manually segmenting only a small number of ``key'' slices and then intelligently propagating these results to the entire volume becomes attractive. Here we use a landmark-based shape-deformation technique to propagate a manual segmentation result of a single slice onto its neighboring slices, thus yielding a much better segmentation result overall than what we can achieve by simply applying fully automatic methods. The design goal for such intelligent boundary propagation consists of two parts:

• First of all, we force the new boundary to honor the available boundary information that can be confidently extracted from the image.
• Secondly, the new boundary should preserve the shape information known from the manual segmentation input, such that the boundary will deform reasonably where we do not have well-defined boundaries.

Enhanced interpreter-aided salt-boundary extraction using shape deformation