Co-segmentation of 3D shapes

Co-segmentation of 3D shapes in the same category is anintensive topic in computer

graphics. In this paper, we present an unsupervised method tosegment a set of

meshes into corresponding parts in a consistent manner. Giventhe over-segmented

patches as input, the co-segmentation result is generated bygrouping them. In

contrast to the previous method, we formulate the problem as amulti-view spectral

clustering task by co-training a set of affinity matricesderived from different shape

descriptors. For each shape descriptor, the affinity matrix isconstructed via combining

low-rankness and sparse representation. The integration ofmultiple features makes

our method tolerate the large geometry and topology variationsamong the 3D meshes

in a set. Moreover, the low-rank and sparse representation cancapture not only the

global structure but also the local relationship, whichdemonstrate robust to outliers.

The experimental results show that our approach successfullysegments each category

in the benchmark dataset into corresponding parts andgenerates more reliable results

compared with the state-of-the-art.

Results