Voxel-based Part-type Segmentation Of 3D Models

21st century has witnessed huge development multimedia field. Representative of digital models and scene technology, 3D information technology, is gradually becoming mainstream media technology, affecting our life, work and entertainment. Also, development of science and technology together with convenience of information-collecting equipments promotes division of three-dimensional data into surface-based and volume-based information. Specially, volume-based data attains wide application in the fields such as 3D reconstruction and visualization, modeling and retrieval of images, motion tracking, digital scenes, data compression and model segmentation. And the demand for semantic and geometric information of each explicit part of models makes in turn segmentation, especially part-type, obtained more and more attention. However, segmenting algorithms volume-based are far from application.Therefore, our proposed mesh segmentation in the view of volume has important theoretical significance and practical value.We firstly introduce commonly-used surface-based and volume-based 3D mesh part-type segmenting algorithms systemically, which differ in features and segmenting rules with different segmenting effects for modes. Then we voxelize mesh model, including surface voxelization by interpolating algorithm and interior padding. Secondly, we get medial skeleton via boundary propagation algorithm. Among the process, we record iteratively-deleting order of every voxel and propose two shape descriptors for meshes’ thickness, which are Maximum Iterative Depth(MID) and MInimum Iterative Depth(MIID). Thirdly, we get MID-based primary segmenting results and fine segmenting results based on Circled Cutting Perimeter(CCP), which can be gotten from Principal Component Analysis(PCA) and Cumulative Averaged Method(CAM). Finally, we compare our algorithm with algorithms in Benchmark qualitatively and quantitatively, and experiments have proved our proposed method can obtain satisfactory segmenting results.