3D shape co-segmentation via sparse and low rank representations
Yin L Y, Guo K, Zhou B, et al
We propose an effective co-segmentation method, combining both individual shape features and global consistence. Given a 3D shape category, we first utilize each single shape as dictionary to sparsely represent the whole shape category. Next, we force every representation of the feature descriptors with lowrank constraints. Eventually, we utilize representation errors to weight the coefficients and obtain the confident ones. Furthermore, through a simple cluster method and smooth process, we achieve the final co-segmentation results. The experimental results show that our approach can outperform other state-of-the-art methods.