3D mesh segmentation via L0-constrained random walks

3D mesh segmentation is a challenging problem in computer graphics, computer vision, and multimedia. In this paper, we cast mesh segmentation as a L0 minimization problem using random walks and L0 norm. In random walks method, the probabilities of random walks change smoothly over the whole model, which may lead to inaccurate segmentation boundaries. To attain a perception-aware result, the changes of probabilities should comply with mesh geometry. That is, the changes of probabilities near region boundaries should be more drastic than those inside the regions. Therefore, we introduce a L0 constraint to reflect the sparsity of probability changes, and identify region boundaries more precisely. Experimental results show that the proposed algorithm is effective, robust, and outperforms the state-of-the-art methods on various 3D meshes.