Combined Central and Subspace Clustering on Computer Vision Applications

Central and subspace clustering methods are at the core of many segmentation problems in computer vision. However, both methods fail to give the correct segmentation in many practical scenarios, e.g., when data are close to the intersection of subspaces or when two cluster centers in different subspaces are spatially close. In this paper, we address these challenges by considering the problem of clustering a set of points lying in a union of n subspaces and distributed around m cluster centers inside each subspace. We proposed a natural generalization of Kmeans and Ksubspaces by combining central and subspace clustering into the same cost function. It results in an algorithm with easy implementation. Experiments on synthetic data compare our method favorably against four other methods. The validation of our method is also demonstrated with real computer vision problems such as face clustering with varying illumination and video shot segmentation of dynamic scenes.