(3,4)-weighted skeleton decomposition for pattern representation and description

Rectification of the chordal axis transform skeleton and criteria for shape decomposition

“…In the latter case, the skeletons computed are actually continuous, being typically made up of piecewise smooth or linear curves. We will refer to such medial axis extensions as semi-discrete medial axes (as only the boundary is discrete and not the skeleton), and reserve the term discrete medial axis to the truly discrete axes (e.g., made up of pixels themselves) derived from digitized shapes as exemplified in [5,6]. Both these methods, however, require uniform or well-sampled representations of the shape or shape boundary to yield satisfactory skeletons.…”

“…The medial axis transform is not defined for discretized (e.g., via pixels) shapes, or shapes specified by discretely sampled contours, as typically encountered in digital imagery. Several useful extensions of the MAT to discrete shapes have been formulated using pixel morphology [5,6] and in the case of discretely sampled boundary, using geometry [7,8]. We note that these two are very different and equally important approaches.…”

Well-shaped skeletons and fast computation of the (3,4) distance transform

“…Again, the removal of these skeletal artifacts is a subject recently studied by others. [12][13][14][15][16] but it is thought that the approach presented here offers increased robustness as will later be explained. Most skeletal artifacts may have to be classified as either bifurcations or elongations.…”

On Reversible Skeletonization Using Anchor-Points from Distance Transforms

“…Pruning methods are described by, e.g., Sanniti di Baja and Thiel in [27] and also by Sanniti di Baja in [23]. These methods take skeleton branches and investigate the effects of branch removal by comparing the recovered regions.…”