Unbiased watershed hierarchical 3D segmentation

We offer fast and robust 3D segmentation methods using marker-based and hierarchical 2D watershed techniques. These methods are based on the hierarchical queue unbiased 2D watershed algorithm invented by S. Beucher [3] which has been adapted in order to rapidly obtain an accurate watershed of triangulated 3D meshes. An evaluation of usual discrete curvature criteria shows that waterfall algorithms provide best results when using the covariance matrix method and the KMax method [7]. The waterfall hierarchical 3D algorithm is faster and more robust when constructing the saddle point graph than when using the homotopy modification technique. However, it can be observed that too many areas are moved from one hierarchical level to the next. That is why a hybrid algorithm, combining waterfall and hierarchical merging approaches is suggested as this makes it possible to obtain a fast and efficient surface 3D model segmentation process. This algorithm is well adapted to difficulties in industrial segmentation for which, in many cases, the number of regions required to characterize a given 3D model can be assessed by users.