Pruning algorithm for Voronoi skeletons

A robust and efficient binary image shape skeleton computation procedure in 2D, including a pruning procedure, is presented. The procedure is as follows: first the shape contour is subsampled along its arclength, then the Voronoi skeleton is computed from the resulting reduced contour set of points, and finally a novel two-stage pruning procedure is applied to obtain a simplified skeleton. Besides the information reduction, this pruning makes the skeleton more robust to noise. Pruning can be done in real time thanks to some useful properties of Voronoi skeletons. Here it is proved that if the two end points of a Voronoi segment are inside the shape, then the entire segment is contained in the shape. A prototype implementation performs in real-time (60 frames per second) on an off-the-shelf computer. Testing it on an in-house but publicly available image database shows the stability and robustness of the approach.