We have successfully parallelized two key steps of this segmentation method: the calculation of the weigtht matrix and the estimation of the eigenvectors. Now, with a large enough cluster, almost any sized post-stack 3D data set can be globally segmented. This is an exciting development in that this algorithm is becoming significantly more practical.
In many places on salt boundaries the amplitude can become weak and the boundary can more easily be tracked using another attribute such as instantaneous frequency. On first glance it seems straight forward to estimate the weights of the normalized cuts image segmentation method using multiple attributes but to balance the weights in an optimal way may be somewhat challengeing. We hope to address this problem in the near future.