基于单元网格近邻势的聚类方法

为解决有限区域内分布的稀疏不均的、具有一定分布结构数据集的聚类分析难题,在近邻思想、网格化空间划分及万有引力叠加原理的启发下，提出了一种基于单元网格近邻势的聚类方法。该方法是基于近邻势的聚类方法在单元网格层次上的一种可以提高聚类速度的扩展。为有效实现并在时空效率上对这种聚类方法进行尽可能的改进,使用了单元网格、网格密度、多维网格划分法、多维索引树等一些比较重要的概念和方法。仿真实验中， 先使用多个不同类型的人工数据集将提出的算法与几个经典聚类算法在聚类精度、聚类速度等方面进行适当的比 较,采用两副图片在RGB(red， green，blue)色彩空间的像素点集来验证该算法的聚类压缩效果。仿真实验说明了这种方法比一些经典的聚类算法在聚类建设、聚类质量方面具有一定的有效性和优越性

Clustering method based on near neighbour influence of grid cells

To solve the cluster analysis problem of data sets with a certain distribution structure within a limited area, under the concepts of neighboring, grid space and the idea of the superposition principle of universal gravitation, this paper proposes a clustering method based on near neighbour influence of grid cells. It is an extended clustering method based on near neighbour influence on the level of grid cells that can enhance the clustering speeds. To achieve effective improvements, the grid cell, the density of grid cell, the multidimensional grid partition method, and the multidimensional index tree are used. In simulation experiments, we first use some different types of artificial datasets to validate our algorithm with several classical clustering algorithms in clustering accuracy, speed and other aspects of clustering. Then we use the pixel sets of two images by clustering in RGB (red , green, blue) color spatial to validate the effect of clustering compress. From these, we find our algorithm has some validity or superiority in some aspects.