Geodesic Segmentation

Class: NodeImageGeodesicSegmentation

Node Icon

Segments structures in images based on a user supplied edge potential map. An initial contour is propagated outwards (or inwards) until it “sticks” to the shape boundaries. This is done by using a level set speed function based on a user supplied edge potential map.

This filter requires two inputs. The first input is a initial level set. The initial level set is a real image which contains the initial contour/surface as the zero level set. For example, a signed distance function from the initial contour/surface is typically used. Unlike the simpler Shape Detection Level Set Image Filter the initial contour does not have to lie wholly within the shape to be segmented. The initial contour is allow to overlap the shape boundary. The extra advection term in the update equation behaves like a doublet and attracts the contour to the boundary. This approach for segmentation follows that of Caselles et al (1997).

The second input is the feature image. For this filter, this is the edge potential map. General characteristics of an edge potential map is that it has values close to zero in regions near the edges and values close to one inside the shape itself. Typically, the edge potential map is computed from the image gradient.


Initial Level

Missing description.

Type: Image4DFloat, Required, Single

Edge Potential

Missing description.

Type: Image4DFloat, Required, Single



Missing description.

Type: Image4DFloat


Advection Scaling Number

Missing description.

Curvature Scaling Number

Missing description.

Propagation Scaling Number

Missing description.

Maximum RMS Error Number

Missing description.

Reverse Expansion Direction Boolean

Missing description.

Iterations Integer

Set the number of iterations.


  1. “Geodesic Active Contours”, V. Caselles, R. Kimmel and G. Sapiro. International Journal on Computer Vision, Vol 22, No. 1, pp 61-97, 1997
  2. “The Insight Segmentation and Registration Toolkit”

See also