一种基于图像去噪的多密度网格聚类算法  

A multi mesh density clustering algorithm based on image denoising

作　　者：田宇[1] 罗辛[1] 

机构地区：[1]东华大学计算机科学与技术学院,上海201620

出　　处：《智能计算机与应用》2016年第1期44-47,共4页Intelligent Computer and Applications

摘　　要：针对传统网格聚类算法仅能够去除空网格的问题,提出一种基于图像分割思想来剔除稀疏数据的多密度网格聚类算法。该算法对原始数据进行网格划分和数据映射,计算网格密度,将每个网格看作图像中的一个像素点,采用Otsu算法确定合适阈值,并给出了阈值应用于网格聚类算法时的阈值折合公式,完成稀疏单元的剔除。在聚类过程中考虑到网格单元内部特征,通过两个网格的相对密度及边界特征得到了相邻网格的相似度度量公式,弥补了网格聚类算法无法应对多密度数据的缺点。在Matlab中进行仿真实验,该算法在聚类之前对网格剔除率为69%,且不需要人工干预,而GAMD和SNN算法未剔除网格。当数据维度增加时,GAMD算法时间远远高于本算法。实验证明,该算法具有较好的数据过滤效果,聚类结果与数据输入顺序无关,在得到任意簇的同时,保证了较高的时间效率且能够广泛应用于各种数据集。Based on the situation that the traditional grid clustering algorithm is only capable of removing empty grid issues,the paper presents an image segmentation thought to weed out the sparse data of multi-density grid clustering algorithm. The algorithm of the original data and data mapping mesh,mesh density calculation,each grid as a picture of a pixel,using Otsu algorithm to determine the appropriate threshold,and gives a reduced threshold algorithm formula while the threshold applies to the grid clustering,and completes culling sparse unit. In the clustering process,taking into account the internal characteristics of grid cells,characterized by the relative density and the border of the two grids,similarity measure formula of adjacent grid is achieved,which makes up for the shortcomings of grid clustering algorithm that the grid can not cope with multi-density data. Conducted in Matlab simulation,grid culling rate is 69%before the algorithm clustering,and does not require human intervention,and GAMD and SNN algorithm does not reject grid. When the data dimension increases,SNN algorithm time is much higher than the present algorithm. Experiments show that the algorithm has better filtering effect data,clustering results are independent of the data input sequence,getting any cluster,while ensuring a high time efficiency,and can be widely applied to various data sets.

关 键 词：网格聚类 多密度 高维稀疏数据 OTSU 聚类算法 

分 类 号：TP391[自动化与计算机技术—计算机应用技术]