Clustering Data on Manifold with Local and Global Consistency

Data clustering aims at finding the hidden patterns in a large collection of data and a large body of effective algorithms have been proposed to partition the data in the past three decades. However, most of the algorithms fail to handle data that expose a manifold structure which is common in many data-driven application, such as interpretation and recognition of video, handwritten character and image data. In this paper, we study the problem of clustering on manifold that aims to partition a set of input data into several clusters each of which contains data points from a simple low-dimensional manifold. We apply the basic assumption of local and global consistency on the manifold. A novel algorithm name CMLGC is proposed to find the proper clusters on the manifold. Our research can also be seen as an instance of manifold learning. The encouraging results on several synthetic and real-world data set are obtained which validate our proposed algorithm.