Research On Clustering Algorithm Based On The Nearest Neighbors

As a basic method in unsupervised learning,clustering is widely applied in various disciplines of science and engineering,such as pattern recognition,artificial intelligence,data mining,image processing,biomedicine and so on.Clustering is to divide the sample points into different categories according to the partition criteria,so that the sample points in the same class are similar to each other and the sample points in the different classes differ from each other.So far,many clustering algorithms have been proposed according to different clustering ideas.However,there are still some defects in these algorithms.Firstly,some need to provide the number of cluster in advance.Secondly,some can't accurately discover the clusters with different shapes and different distributions.Moreover,some aren't suitable to deal with the data including the overlapped clusters.In order to overcome these defects,we make a study on the clustering algorithm based on neighborhood according to the relationship of the sample and its nearest neighbors,and the innovative achievements are summarized as follows:(1)Studying the relationship between the points with its neighbors,a clustering evaluation index according to the clustering criteria on the furthest or nearest neighbor is proposed,called Furthest and Nearest Score(FNS)index.Moreover,an automatic clustering algorithm,called Clustering algorithm based on the furthest and nearest score,is proposed according to the designed index.The simulation experiment proves the validity and feasibility of the proposed algorithm.At the same time,the algorithm is applied to the image segmentation problem,and its results fully prove the effectiveness of the algorithm in the image automatic segmentation problem.(2)In order to overcome the bad performance of the existing clustering algorithms in identifying different shape clustering or dealing with the overlapped clusters,we propose a new clustering algorithm based on the connected region generation.The algorithm is an effective robust clustering algorithm according to the connectivity between the points and its nearest neighbors.In the algorithm,the region generation algorithm(CRG)is proposed to get connected regions and a discrete point set,and different connected regions correspond to different clusters.Then,the remaining discrete points are classified according to the regional expansion method and the consistency criterion,and the final clustering results are obtained.The experimental results fully demonstrate the effectiveness and feasibility of the proposed algorithm.(3)According to the relationship between the nearest neighbors and the reverse neighbors,the automatic clustering algorithm based on merging subregions(ACMS)which can automatically classify the points is proposed.In this algorithm,the denseness of the sample is defined and the dense points are selected according to the density of the sample at first.Then,the dense points are divided into subregions according to the criteria and these subregions are combined according to their similarity between groups.Finally,using the edge extension method and the set belonging degree,the non-dense points are divided into the corresponding region,and the clustering results are obtained.The experimental results show the effectiveness and feasibility of the algorithm.