På svenska

Presentation Information     2006-05-15 (15:15)   •  The seminar room at Vi2

Speaker Erik Vidholm
Title Accelerating the computation of gradient vector flow fields
Abstract Gradient vector flow (GVF) was introduced by Xu and Prince in 1998 as an external force for segmentation using active contour models (snakes). The basic idea behind GVF is to propagate edge information from strong boundaries into homogeneous regions by diffusing the gradient vectors of an edge map derived from the original image. In 3D, a problem is that the convergence rate of the commonly used numerical scheme to compute GVF does not allow for practical use. For interactive 3D medical image segmentation with GVF, computational times in the order of a minute rather than an hour are needed. In this talk, the aim is to present alternative computation schemes (stationary iterative methods, conjugate gradients, and multigrid methods) and give hints on their advantages and disadvantages.