Semi-supervised clustering with associative-link constraints for 3D models

3D model retrieval and classification is one of the most important topics in the areas of computer graphics and multimedia. Clustering is widely used to organize and to index multimedia databases. However, few of previous works consider semi-supervised approaches. In this paper, we propose a new semi-supervised clustering algorithm and apply it to perform the clustering task in three public 3D model datasets. The proposed algorithm not only utilizes the conventional pairwise constraints, including the Must-link constraint (ML) and the Cannot-link constraint (CL), but also designs a new kind of constraints called the associative-link constraint (AL). Experiments on datasets from three public 3D databases demonstrate advantages of our algorithm beyond the standard Kmeans and two state-of-the-art semi-supervised clustering algorithms.