Embodiments of the invention utilize a variational framework for computing curve skeletons (CS) of objects whose cross section is not necessary tubular. Embodiments utilize an energy function, which is proportional to some medialness function, such that the
minimum cost path between any two medial voxels in the shape is a curve skeleton. Different medialness functions include the
Euclidean distance field and a modified version of the magnitude of the gradient
vector flow (GVF), which results in two different energy functions. The first energy function controls the identification of the topological nodes of the shape from which curve skeletons start, while the second one controls the extraction of the curve skeletons. Preferred embodiments are completely automated since all parameters are analytically estimated. Embodiments are highly less sensitive to boundary
noise, are able to extract the entire curve skeletons, as well as only part of it given the starting and the end voxels, and do not require voxels to be isotropic. In addition, computed curve skeletons are highly centered and form a connected graph. Preferred embodiments have been validated the framework both quantitatively and qualitatively against several 3D shapes of different complexity.