Flow Shape Clustering Algorithm And Its Application

With the increase of the application area of pattern recognization and machine learning, such as text proccess, voice recognization and compute vision et.al., manifold clustering becomes more and more impotant. In recent years, the related methods have been an active area. Each year, there are a lot of publications appearing in the premier conferences and journals. All the existing algorithms have many restrictions, which depands on the underlying dimensional and can not deal with multi manifolds efficiently.In this paper, we explore manifold clustering techniques and its application. In particlar, we have done the work in the following three aspects.1. We propose a method for clustering data points that originate from multiple lowdimensional manifolds. The algorithm first searches for an optimal data sequence by minimizing the energy of input data using Active Tabu search, where the energy is described primarily with spatial position of points and discrete curvature of underlying manifolds. To distinguish different manifolds, boundary points between manifolds need to be detected along the optimal sequence.2. We mine the local and global feature of data set, construct higher-dimensional signal surface and define a topology distance measure based on topology structure. This measure can make the data in the same cluster compact and disperse the data in the different clusters. A novel framework for manifold cluster algorithm base on this topology distance measure is proposed.3. A novel framework is desined for retrieval in this paper as well as presented its application to image retrieval. The retrieval framework is based on searching for a most ordered cycle from the input data. Particularly, for image retrieval, our framework has an obviously advantage over pervious manifold based methods: our method can directly rank and return relevant images and does not need to learn a mapping from the feature space to the unclear semantic manifold space.