A robust approach to multi-feature based mesh segmentation using adaptive density estimation

Abstract

In this paper, a new and robust approach to mesh segmentation is presented. There are various algorithms which deliver satisfying results on clean 3D models. However, many reverse-engineering applications in computer vision such as 3D reconstruction produce extremely noisy or even incomplete data. The presented segmentation algorithm copes with this challenge by a robust semi-global clustering scheme and a cost-function that is based on a probabilistic model. Vision based reconstruction methods are able to generate colored meshes and it is shown, how the vertex color can be used as a supportive feature. A probabilistic framework allows the algorithm to be easily extended by other user defined features. The segmentation scheme is a local iterative optimization embedded in a hierarchical clustering technique. The presented method has been successfully tested on various real world examples. © 2011 Springer-Verlag.