Evolutionary 3D-Shape Segmentation Using Satellite Seeds

The aim of 3D-shape segmentation is to divide the surface of an object into meaningful parts. We present a novel version of seed-point-based segmentation including an evolutionary optimization to obtain better segments. At first, some initial seeds are defined. Each of them generates several so-called satellite seeds which enable a more detailed control of the segment boundaries. The locations and weights of the seeds are optimized with an Evolution Strategy. The objective function takes the object's curvature at the segments' boundaries into account as well as the length of these boundaries. An extensive evaluation and comparison with important existing segmentation approaches demonstrates the great potential of our approach.