Feature extraction : Crest lines

As defined by Monga, Benayoun and Thirion, Crest Lines are the locus of points on a surface whose largest curvature (in absolute value) is locally maximal in the associated principal direction. Such a line can be viewed as a generalization of edge for smooth surfaces in 3D.

Thirion and Gourdon have developed an algorithm called the marching lines that efficiently extract these lines. In this process, the surface of the interest object is usually obtained as an iso-surface of the original 3D image.

Example of crest lines on the brain surface (from left-top to bottom: left side, frontal and top views).

Lines in red roughly correspond to sulci whereas blue lines could be interpreted as gyri.

Courtesy of G. Subsol

[ Back to substructure matching with frames ]