Parallel Implementation of Image Segmentation for Tracking 3D Salt Boundaries

Normalized cuts image segmentation can be used to track salt boundaries when the salt boundary amplitude is greater than any other local reflections. We apply a modified version of this image segmentation method to partition seismic images along salt boundaries. In principle, our method is capable of tracking boundaries that are not continuous: boundaries where conventional horizon tracking algorithms may fail. Normalized cuts image segmentation creates a weight matrix of weights connecting each pixel in the image to other pixels in a local neighborhood. The weight is made weak where the amplitude between two pixels is high. The weight matrix for 3D seismic data cubes can be quite large thus requiring memory and computational-time saving measures to be implemented. By imposing bounds, we significantly reduce the size of the problem and, as a result, increase efficiency and robustness. Also, errors created by segmenting thin images can be rectified with novel boundary conditions described here. To reduce the extensive computation time of this algorithm, we distribute the calculation of the weight matrix on a parallel network. This method is demonstrated to be effective on a 2D seismic section. Preliminary 3D seismic results are also included.