In 3D segmentation, a deformable model can be efficiently guided by a gradient vector flow (GVF) field. The computation of a GVF field consists mainly of solving a huge discretized system of elliptic partial differential equations. These discrete equations have several properties that can be utilized to accelerate the process of finding an approximate solution. Here, stationary iterative methods, preconditioned conjugate gradient methods, and multigrid methods are considered in order to compute the GVF field at computational times acceptable for interactive 3D segmentation.
Citation:
Erik Vidholm, Per Sundqvist, Ingela Nystrom, "Accelerating the Computation of 3D Gradient Vector Flow Fields," icpr, vol. 3, pp.677-680, 18th International Conference on Pattern Recognition (ICPR'06) Volume 3, 2006