Shape clustering

AbstractThis paper aims to address the problem of shape clustering by discovering the common structure which captures the intrinsic structural information of shapes belonging to the same cluster. It is based on a skeleton graph, named common structure skeleton graph (CSSG), which expresses possible correspondences between nodes of the individual skeletons of the cluster. To construct the CSSG, we derive the correspondences by the optimal subsequence bijection (OSB). To cluster the shape data, we apply an agglomerative clustering scheme, in each iteration, the CSSGs are formed from each cluster and the two closest clusters are merged into one. The proposed agglomerative clustering algorithm has been evaluated on several shape data sets, including three articulated shape data sets, Torsello's data set, and a gesture data set. In all experiments, our method demonstrates effective performance compared to other algorithms. Highlights A method is proposed to discover the common structure of a cluster of shapes. A cluster of shapes are represented by their CSSG. Assigning weights to nodes and edges in CSSGs to help measure their distances. An agglomerative strategy is proposed for shape clustering based on CSSGs.