The most significant difference between 3D image segmentation and 2D image segmentation is during the generation of the weight matrix, the rest of the algorithm is almost identical. When creating the weight matrix, instead of randomly sampling from a circular neighborhood, we sample from a sphere. Of course this means that more points are sampled per node. This, in turn, means, that the sparse matrix is considerably less sparse and the entire algorithm more expensive. Therefore, even with sparse matrices and tight boundaries, we still need to look for ways of reducing the computional-time cost of this algorithm for 3D problems.

4/5/2006