Partitioning Clustering Method Based On Geometric Features

Cluster analysis grouped refers to the set of physical or abstract objects into a similarobject composed of multiple classes analysis, is an important branch of pattern recognition. Itis a typical unsupervised classification method. The current main clustering algorithmincludes hierarchical clustering, density clustering, grid clustering, kernel clustering, spectralradius clustering method. These techniques are used to describe the data, to measure thesimilarity between different data sources, as well as the data sources are classified intodifferent classes. Clustering algorithm based on partition which, density based clusteringalgorithm, clustering algorithm based on grid clustering algorithm based on graph theory,together known as the segmentation based on clustering method. These algorithms either intheory or in practical application have been fully discussed. But as a result of the datastructure, there is not a method can be used to all database. Each algorithm has its owndefects.Firstly, this paper introduces some basic clustering algorithm, including the algorithmsteps and discuss, and the advantages and disadvantages, and then introduce someimprovements in basic clustering algorithm on the algorithm.In the third chapter introduces the k-means algorithm based on the reference points, K-means algorithm based on the reference points is proposed based on the basis of k-meansalgorithm and reference point, mainly to overcome k-means algorithm using random searchand clustering results received sample sequence effects in the search for the initial clusteringcenter, Geometric feature sample reaction mode makes the initial cluster center to better set,In order to reduce the number of iterations, improve the clustering accuracy in clusteringwork. In the fourth chapter proposed the nearest neighbor clustering algorithm of meshdensity maximum, Nearest neighbor clustering algorithm of mesh density is the largest basedon nearest neighbor clustering method and the central grid, Mainly in order to overcome thenearest neighbor clustering results by single disadvantage of larger and the initial clustercenters designated by the effects of the first cluster center, in order to improve the clusteringaccuracy.Experiments show that, compared with the traditional clustering algorithm, this algorithmhas good clustering effect.