Most of the tasks derived from shape analysis rely on the problem of finding meaningful correspondences between two or more shapes. In medical imaging analysis, this problem is a challenging topic due to the need to establish matching features in a given registration process. Besides, a similarity measure between shapes must be computed in order to obtain these correspondences. In this paper, we propose a method for 3D shape correspondences based on groupwise analysis using probabilistic latent variable models. The proposed method finds groupwise correspondences, and can handle multiple shapes with different number of objects (vertex or descriptors for every shape). By assigning a latent vector for each shape descriptor, we can cluster objects in different shapes, and find correspondences between clusters. We use a Dirichlet process prior in order to infer the number of clusters and find groupwise correspondences in an unsupervised manner. The results show that the proposed method can efficiently establish meaningful correspondences without using similarity measures between shapes.
展开▼
